X-Git-Url: https://git.saurik.com/redis.git/blobdiff_plain/2affc3ed20de01efcf9d2aecb86fbabd97666eca..682c73e87e6edfdfe76b9bb4b3d66ce39465b359:/redis.c?ds=sidebyside diff --git a/redis.c b/redis.c index 71f1594f..b926e5a4 100644 --- a/redis.c +++ b/redis.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2006-2009, Salvatore Sanfilippo + * Copyright (c) 2009-2010, Salvatore Sanfilippo * All rights reserved. * * Redistribution and use in source and binary forms, with or without @@ -27,7 +27,7 @@ * POSSIBILITY OF SUCH DAMAGE. */ -#define REDIS_VERSION "1.3.0" +#define REDIS_VERSION "1.3.8" #include "fmacros.h" #include "config.h" @@ -38,6 +38,7 @@ #include #include #define __USE_POSIX199309 +#define __USE_UNIX98 #include #ifdef HAVE_BACKTRACE @@ -59,6 +60,7 @@ #include #include #include +#include #if defined(__sun) #include "solarisfixes.h" @@ -73,6 +75,7 @@ #include "zmalloc.h" /* total memory usage aware version of malloc/free */ #include "lzf.h" /* LZF compression library */ #include "pqsort.h" /* Partial qsort for SORT+LIMIT */ +#include "zipmap.h" /* Error codes */ #define REDIS_OK 0 @@ -83,12 +86,12 @@ #define REDIS_MAXIDLETIME (60*5) /* default client timeout */ #define REDIS_IOBUF_LEN 1024 #define REDIS_LOADBUF_LEN 1024 -#define REDIS_STATIC_ARGS 4 +#define REDIS_STATIC_ARGS 8 #define REDIS_DEFAULT_DBNUM 16 #define REDIS_CONFIGLINE_MAX 1024 #define REDIS_OBJFREELIST_MAX 1000000 /* Max number of objects to cache */ #define REDIS_MAX_SYNC_TIME 60 /* Slave can't take more to sync */ -#define REDIS_EXPIRELOOKUPS_PER_CRON 100 /* try to expire 100 keys/second */ +#define REDIS_EXPIRELOOKUPS_PER_CRON 10 /* lookup 10 expires per loop */ #define REDIS_MAX_WRITE_PER_EVENT (1024*64) #define REDIS_REQUEST_MAX_SIZE (1024*1024*256) /* max bytes in inline command */ @@ -108,6 +111,7 @@ config file and the server is using more than maxmemory bytes of memory. In short this commands are denied on low memory conditions. */ #define REDIS_CMD_DENYOOM 4 +#define REDIS_CMD_FORCE_REPLICATION 8 /* Force replication even if dirty is 0 */ /* Object types */ #define REDIS_STRING 0 @@ -116,9 +120,17 @@ #define REDIS_ZSET 3 #define REDIS_HASH 4 -/* Objects encoding */ +/* Objects encoding. Some kind of objects like Strings and Hashes can be + * internally represented in multiple ways. The 'encoding' field of the object + * is set to one of this fields for this object. */ #define REDIS_ENCODING_RAW 0 /* Raw representation */ #define REDIS_ENCODING_INT 1 /* Encoded as integer */ +#define REDIS_ENCODING_ZIPMAP 2 /* Encoded as zipmap */ +#define REDIS_ENCODING_HT 3 /* Encoded as an hash table */ + +static char* strencoding[] = { + "raw", "int", "zipmap", "hashtable" +}; /* Object types only used for dumping to disk */ #define REDIS_EXPIRETIME 253 @@ -152,13 +164,31 @@ #define REDIS_RDB_ENC_INT32 2 /* 32 bit signed integer */ #define REDIS_RDB_ENC_LZF 3 /* string compressed with FASTLZ */ +/* Virtual memory object->where field. */ +#define REDIS_VM_MEMORY 0 /* The object is on memory */ +#define REDIS_VM_SWAPPED 1 /* The object is on disk */ +#define REDIS_VM_SWAPPING 2 /* Redis is swapping this object on disk */ +#define REDIS_VM_LOADING 3 /* Redis is loading this object from disk */ + +/* Virtual memory static configuration stuff. + * Check vmFindContiguousPages() to know more about this magic numbers. */ +#define REDIS_VM_MAX_NEAR_PAGES 65536 +#define REDIS_VM_MAX_RANDOM_JUMP 4096 +#define REDIS_VM_MAX_THREADS 32 +#define REDIS_THREAD_STACK_SIZE (1024*1024*4) +/* The following is the *percentage* of completed I/O jobs to process when the + * handelr is called. While Virtual Memory I/O operations are performed by + * threads, this operations must be processed by the main thread when completed + * in order to take effect. */ +#define REDIS_MAX_COMPLETED_JOBS_PROCESSED 1 + /* Client flags */ -#define REDIS_CLOSE 1 /* This client connection should be closed ASAP */ -#define REDIS_SLAVE 2 /* This client is a slave server */ -#define REDIS_MASTER 4 /* This client is a master server */ -#define REDIS_MONITOR 8 /* This client is a slave monitor, see MONITOR */ -#define REDIS_MULTI 16 /* This client is in a MULTI context */ -#define REDIS_BLOCKED 32 /* The client is waiting in a blocking operation */ +#define REDIS_SLAVE 1 /* This client is a slave server */ +#define REDIS_MASTER 2 /* This client is a master server */ +#define REDIS_MONITOR 4 /* This client is a slave monitor, see MONITOR */ +#define REDIS_MULTI 8 /* This client is in a MULTI context */ +#define REDIS_BLOCKED 16 /* The client is waiting in a blocking operation */ +#define REDIS_IO_WAIT 32 /* The client is waiting for Virtual Memory I/O */ /* Slave replication state - slave side */ #define REDIS_REPL_NONE 0 /* No active replication */ @@ -186,8 +216,9 @@ /* Log levels */ #define REDIS_DEBUG 0 -#define REDIS_NOTICE 1 -#define REDIS_WARNING 2 +#define REDIS_VERBOSE 1 +#define REDIS_NOTICE 2 +#define REDIS_WARNING 3 /* Anti-warning macro... */ #define REDIS_NOTUSED(V) ((void) V) @@ -200,19 +231,42 @@ #define APPENDFSYNC_ALWAYS 1 #define APPENDFSYNC_EVERYSEC 2 +/* Hashes related defaults */ +#define REDIS_HASH_MAX_ZIPMAP_ENTRIES 64 +#define REDIS_HASH_MAX_ZIPMAP_VALUE 512 + /* We can print the stacktrace, so our assert is defined this way: */ -#define redisAssert(_e) ((_e)?(void)0 : (_redisAssert(#_e),exit(1))) -static void _redisAssert(char *estr); +#define redisAssert(_e) ((_e)?(void)0 : (_redisAssert(#_e,__FILE__,__LINE__),_exit(1))) +#define redisPanic(_e) _redisPanic(#_e,__FILE__,__LINE__),_exit(1) +static void _redisAssert(char *estr, char *file, int line); +static void _redisPanic(char *msg, char *file, int line); /*================================= Data types ============================== */ /* A redis object, that is a type able to hold a string / list / set */ + +/* The VM object structure */ +struct redisObjectVM { + off_t page; /* the page at witch the object is stored on disk */ + off_t usedpages; /* number of pages used on disk */ + time_t atime; /* Last access time */ +} vm; + +/* The actual Redis Object */ typedef struct redisObject { void *ptr; unsigned char type; unsigned char encoding; - unsigned char notused[2]; + unsigned char storage; /* If this object is a key, where is the value? + * REDIS_VM_MEMORY, REDIS_VM_SWAPPED, ... */ + unsigned char vtype; /* If this object is a key, and value is swapped out, + * this is the type of the swapped out object. */ int refcount; + /* VM fields, this are only allocated if VM is active, otherwise the + * object allocation function will just allocate + * sizeof(redisObjct) minus sizeof(redisObjectVM), so using + * Redis without VM active will not have any overhead. */ + struct redisObjectVM vm; } robj; /* Macro used to initalize a Redis object allocated on the stack. @@ -224,12 +278,14 @@ typedef struct redisObject { _var.type = REDIS_STRING; \ _var.encoding = REDIS_ENCODING_RAW; \ _var.ptr = _ptr; \ + if (server.vm_enabled) _var.storage = REDIS_VM_MEMORY; \ } while(0); typedef struct redisDb { dict *dict; /* The keyspace for this DB */ dict *expires; /* Timeout of keys with a timeout set */ dict *blockingkeys; /* Keys with clients waiting for data (BLPOP) */ + dict *io_keys; /* Keys with clients waiting for VM I/O */ int id; } redisDb; @@ -259,8 +315,7 @@ typedef struct redisClient { list *reply; int sentlen; time_t lastinteraction; /* time of the last interaction, used for timeout */ - int flags; /* REDIS_CLOSE | REDIS_SLAVE | REDIS_MONITOR */ - /* REDIS_MULTI */ + int flags; /* REDIS_SLAVE | REDIS_MONITOR | REDIS_MULTI ... */ int slaveseldb; /* slave selected db, if this client is a slave */ int authenticated; /* when requirepass is non-NULL */ int replstate; /* replication state if this is a slave */ @@ -268,10 +323,15 @@ typedef struct redisClient { long repldboff; /* replication DB file offset */ off_t repldbsize; /* replication DB file size */ multiState mstate; /* MULTI/EXEC state */ - robj *blockingkey; /* The key we waiting to terminate a blocking + robj **blockingkeys; /* The key we are waiting to terminate a blocking * operation such as BLPOP. Otherwise NULL. */ + int blockingkeysnum; /* Number of blocking keys */ time_t blockingto; /* Blocking operation timeout. If UNIX current time * is >= blockingto then the operation timed out. */ + list *io_keys; /* Keys this client is waiting to be loaded from the + * swap file in order to continue. */ + dict *pubsub_channels; /* channels a client is interested in (SUBSCRIBE) */ + list *pubsub_patterns; /* patterns a client is interested in (SUBSCRIBE) */ } redisClient; struct saveparam { @@ -284,8 +344,6 @@ struct redisServer { int port; int fd; redisDb *db; - dict *sharingpool; /* Poll used for object sharing */ - unsigned int sharingpoolsize; long long dirty; /* changes to DB from the last save */ list *clients; list *slaves, *monitors; @@ -294,11 +352,11 @@ struct redisServer { int cronloops; /* number of times the cron function run */ list *objfreelist; /* A list of freed objects to avoid malloc() */ time_t lastsave; /* Unix time of last save succeeede */ - size_t usedmemory; /* Used memory in megabytes */ /* Fields used only for stats */ time_t stat_starttime; /* server start time */ long long stat_numcommands; /* number of processed commands */ long long stat_numconnections; /* number of connections received */ + long long stat_expiredkeys; /* number of expired keys */ /* Configuration */ int verbosity; int glueoutputbuf; @@ -323,6 +381,7 @@ struct redisServer { char *requirepass; int shareobjects; int rdbcompression; + int activerehashing; /* Replication related */ int isslave; char *masterauth; @@ -331,20 +390,81 @@ struct redisServer { redisClient *master; /* client that is master for this slave */ int replstate; unsigned int maxclients; - unsigned long maxmemory; + unsigned long long maxmemory; + unsigned int blpop_blocked_clients; + unsigned int vm_blocked_clients; /* Sort parameters - qsort_r() is only available under BSD so we * have to take this state global, in order to pass it to sortCompare() */ int sort_desc; int sort_alpha; int sort_bypattern; + /* Virtual memory configuration */ + int vm_enabled; + char *vm_swap_file; + off_t vm_page_size; + off_t vm_pages; + unsigned long long vm_max_memory; + /* Hashes config */ + size_t hash_max_zipmap_entries; + size_t hash_max_zipmap_value; + /* Virtual memory state */ + FILE *vm_fp; + int vm_fd; + off_t vm_next_page; /* Next probably empty page */ + off_t vm_near_pages; /* Number of pages allocated sequentially */ + unsigned char *vm_bitmap; /* Bitmap of free/used pages */ + time_t unixtime; /* Unix time sampled every second. */ + /* Virtual memory I/O threads stuff */ + /* An I/O thread process an element taken from the io_jobs queue and + * put the result of the operation in the io_done list. While the + * job is being processed, it's put on io_processing queue. */ + list *io_newjobs; /* List of VM I/O jobs yet to be processed */ + list *io_processing; /* List of VM I/O jobs being processed */ + list *io_processed; /* List of VM I/O jobs already processed */ + list *io_ready_clients; /* Clients ready to be unblocked. All keys loaded */ + pthread_mutex_t io_mutex; /* lock to access io_jobs/io_done/io_thread_job */ + pthread_mutex_t obj_freelist_mutex; /* safe redis objects creation/free */ + pthread_mutex_t io_swapfile_mutex; /* So we can lseek + write */ + pthread_attr_t io_threads_attr; /* attributes for threads creation */ + int io_active_threads; /* Number of running I/O threads */ + int vm_max_threads; /* Max number of I/O threads running at the same time */ + /* Our main thread is blocked on the event loop, locking for sockets ready + * to be read or written, so when a threaded I/O operation is ready to be + * processed by the main thread, the I/O thread will use a unix pipe to + * awake the main thread. The followings are the two pipe FDs. */ + int io_ready_pipe_read; + int io_ready_pipe_write; + /* Virtual memory stats */ + unsigned long long vm_stats_used_pages; + unsigned long long vm_stats_swapped_objects; + unsigned long long vm_stats_swapouts; + unsigned long long vm_stats_swapins; + /* Pubsub */ + dict *pubsub_channels; /* Map channels to list of subscribed clients */ + list *pubsub_patterns; /* A list of pubsub_patterns */ + /* Misc */ + FILE *devnull; }; +typedef struct pubsubPattern { + redisClient *client; + robj *pattern; +} pubsubPattern; + typedef void redisCommandProc(redisClient *c); struct redisCommand { char *name; redisCommandProc *proc; int arity; int flags; + /* Use a function to determine which keys need to be loaded + * in the background prior to executing this command. Takes precedence + * over vm_firstkey and others, ignored when NULL */ + redisCommandProc *vm_preload_proc; + /* What keys should be loaded in background when calling this command? */ + int vm_firstkey; /* The first argument that's a key (0 = no keys) */ + int vm_lastkey; /* THe last argument that's a key */ + int vm_keystep; /* The step between first and last key */ }; struct redisFunctionSym { @@ -370,6 +490,7 @@ typedef struct _redisSortOperation { typedef struct zskiplistNode { struct zskiplistNode **forward; struct zskiplistNode *backward; + unsigned int *span; double score; robj *obj; } zskiplistNode; @@ -387,13 +508,16 @@ typedef struct zset { /* Our shared "common" objects */ +#define REDIS_SHARED_INTEGERS 10000 struct sharedObjectsStruct { robj *crlf, *ok, *err, *emptybulk, *czero, *cone, *pong, *space, *colon, *nullbulk, *nullmultibulk, *queued, *emptymultibulk, *wrongtypeerr, *nokeyerr, *syntaxerr, *sameobjecterr, *outofrangeerr, *plus, *select0, *select1, *select2, *select3, *select4, - *select5, *select6, *select7, *select8, *select9; + *select5, *select6, *select7, *select8, *select9, + *messagebulk, *subscribebulk, *unsubscribebulk, *mbulk3, + *psubscribebulk, *punsubscribebulk, *integers[REDIS_SHARED_INTEGERS]; } shared; /* Global vars that are actally used as constants. The following double @@ -402,6 +526,22 @@ struct sharedObjectsStruct { static double R_Zero, R_PosInf, R_NegInf, R_Nan; +/* VM threaded I/O request message */ +#define REDIS_IOJOB_LOAD 0 /* Load from disk to memory */ +#define REDIS_IOJOB_PREPARE_SWAP 1 /* Compute needed pages */ +#define REDIS_IOJOB_DO_SWAP 2 /* Swap from memory to disk */ +typedef struct iojob { + int type; /* Request type, REDIS_IOJOB_* */ + redisDb *db;/* Redis database */ + robj *key; /* This I/O request is about swapping this key */ + robj *val; /* the value to swap for REDIS_IOREQ_*_SWAP, otherwise this + * field is populated by the I/O thread for REDIS_IOREQ_LOAD. */ + off_t page; /* Swap page where to read/write the object */ + off_t pages; /* Swap pages needed to save object. PREPARE_SWAP return val */ + int canceled; /* True if this command was canceled by blocking side of VM */ + pthread_t thread; /* ID of the thread processing this entry */ +} iojob; + /*================================ Prototypes =============================== */ static void freeStringObject(robj *o); @@ -416,15 +556,16 @@ static void addReplySds(redisClient *c, sds s); static void incrRefCount(robj *o); static int rdbSaveBackground(char *filename); static robj *createStringObject(char *ptr, size_t len); -static void replicationFeedSlaves(list *slaves, struct redisCommand *cmd, int dictid, robj **argv, int argc); +static robj *dupStringObject(robj *o); +static void replicationFeedSlaves(list *slaves, int dictid, robj **argv, int argc); static void feedAppendOnlyFile(struct redisCommand *cmd, int dictid, robj **argv, int argc); static int syncWithMaster(void); -static robj *tryObjectSharing(robj *o); -static int tryObjectEncoding(robj *o); +static robj *tryObjectEncoding(robj *o); static robj *getDecodedObject(robj *o); static int removeExpire(redisDb *db, robj *key); static int expireIfNeeded(redisDb *db, robj *key); static int deleteIfVolatile(redisDb *db, robj *key); +static int deleteIfSwapped(redisDb *db, robj *key); static int deleteKey(redisDb *db, robj *key); static time_t getExpire(redisDb *db, robj *key); static int setExpire(redisDb *db, robj *key, time_t when); @@ -443,8 +584,44 @@ static void sendReplyToClientWritev(aeEventLoop *el, int fd, void *privdata, int static void initClientMultiState(redisClient *c); static void freeClientMultiState(redisClient *c); static void queueMultiCommand(redisClient *c, struct redisCommand *cmd); -static void unblockClient(redisClient *c); +static void unblockClientWaitingData(redisClient *c); static int handleClientsWaitingListPush(redisClient *c, robj *key, robj *ele); +static void vmInit(void); +static void vmMarkPagesFree(off_t page, off_t count); +static robj *vmLoadObject(robj *key); +static robj *vmPreviewObject(robj *key); +static int vmSwapOneObjectBlocking(void); +static int vmSwapOneObjectThreaded(void); +static int vmCanSwapOut(void); +static int tryFreeOneObjectFromFreelist(void); +static void acceptHandler(aeEventLoop *el, int fd, void *privdata, int mask); +static void vmThreadedIOCompletedJob(aeEventLoop *el, int fd, void *privdata, int mask); +static void vmCancelThreadedIOJob(robj *o); +static void lockThreadedIO(void); +static void unlockThreadedIO(void); +static int vmSwapObjectThreaded(robj *key, robj *val, redisDb *db); +static void freeIOJob(iojob *j); +static void queueIOJob(iojob *j); +static int vmWriteObjectOnSwap(robj *o, off_t page); +static robj *vmReadObjectFromSwap(off_t page, int type); +static void waitEmptyIOJobsQueue(void); +static void vmReopenSwapFile(void); +static int vmFreePage(off_t page); +static void zunionInterBlockClientOnSwappedKeys(redisClient *c); +static int blockClientOnSwappedKeys(struct redisCommand *cmd, redisClient *c); +static int dontWaitForSwappedKey(redisClient *c, robj *key); +static void handleClientsBlockedOnSwappedKey(redisDb *db, robj *key); +static void readQueryFromClient(aeEventLoop *el, int fd, void *privdata, int mask); +static struct redisCommand *lookupCommand(char *name); +static void call(redisClient *c, struct redisCommand *cmd); +static void resetClient(redisClient *c); +static void convertToRealHash(robj *o); +static int pubsubUnsubscribeAllChannels(redisClient *c, int notify); +static int pubsubUnsubscribeAllPatterns(redisClient *c, int notify); +static void freePubsubPattern(void *p); +static int listMatchPubsubPattern(void *a, void *b); +static int compareStringObjects(robj *a, robj *b); +static void usage(); static void authCommand(redisClient *c); static void pingCommand(redisClient *c); @@ -514,6 +691,7 @@ static void zaddCommand(redisClient *c); static void zincrbyCommand(redisClient *c); static void zrangeCommand(redisClient *c); static void zrangebyscoreCommand(redisClient *c); +static void zcountCommand(redisClient *c); static void zrevrangeCommand(redisClient *c); static void zcardCommand(redisClient *c); static void zremCommand(redisClient *c); @@ -521,99 +699,152 @@ static void zscoreCommand(redisClient *c); static void zremrangebyscoreCommand(redisClient *c); static void multiCommand(redisClient *c); static void execCommand(redisClient *c); +static void discardCommand(redisClient *c); static void blpopCommand(redisClient *c); static void brpopCommand(redisClient *c); +static void appendCommand(redisClient *c); +static void substrCommand(redisClient *c); +static void zrankCommand(redisClient *c); +static void zrevrankCommand(redisClient *c); +static void hsetCommand(redisClient *c); +static void hsetnxCommand(redisClient *c); +static void hgetCommand(redisClient *c); +static void hmsetCommand(redisClient *c); +static void hmgetCommand(redisClient *c); +static void hdelCommand(redisClient *c); +static void hlenCommand(redisClient *c); +static void zremrangebyrankCommand(redisClient *c); +static void zunionCommand(redisClient *c); +static void zinterCommand(redisClient *c); +static void hkeysCommand(redisClient *c); +static void hvalsCommand(redisClient *c); +static void hgetallCommand(redisClient *c); +static void hexistsCommand(redisClient *c); +static void configCommand(redisClient *c); +static void hincrbyCommand(redisClient *c); +static void subscribeCommand(redisClient *c); +static void unsubscribeCommand(redisClient *c); +static void psubscribeCommand(redisClient *c); +static void punsubscribeCommand(redisClient *c); +static void publishCommand(redisClient *c); /*================================= Globals ================================= */ /* Global vars */ static struct redisServer server; /* server global state */ static struct redisCommand cmdTable[] = { - {"get",getCommand,2,REDIS_CMD_INLINE}, - {"set",setCommand,3,REDIS_CMD_BULK|REDIS_CMD_DENYOOM}, - {"setnx",setnxCommand,3,REDIS_CMD_BULK|REDIS_CMD_DENYOOM}, - {"del",delCommand,-2,REDIS_CMD_INLINE}, - {"exists",existsCommand,2,REDIS_CMD_INLINE}, - {"incr",incrCommand,2,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM}, - {"decr",decrCommand,2,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM}, - {"mget",mgetCommand,-2,REDIS_CMD_INLINE}, - {"rpush",rpushCommand,3,REDIS_CMD_BULK|REDIS_CMD_DENYOOM}, - {"lpush",lpushCommand,3,REDIS_CMD_BULK|REDIS_CMD_DENYOOM}, - {"rpop",rpopCommand,2,REDIS_CMD_INLINE}, - {"lpop",lpopCommand,2,REDIS_CMD_INLINE}, - {"brpop",brpopCommand,3,REDIS_CMD_INLINE}, - {"blpop",blpopCommand,3,REDIS_CMD_INLINE}, - {"llen",llenCommand,2,REDIS_CMD_INLINE}, - {"lindex",lindexCommand,3,REDIS_CMD_INLINE}, - {"lset",lsetCommand,4,REDIS_CMD_BULK|REDIS_CMD_DENYOOM}, - {"lrange",lrangeCommand,4,REDIS_CMD_INLINE}, - {"ltrim",ltrimCommand,4,REDIS_CMD_INLINE}, - {"lrem",lremCommand,4,REDIS_CMD_BULK}, - {"rpoplpush",rpoplpushcommand,3,REDIS_CMD_BULK|REDIS_CMD_DENYOOM}, - {"sadd",saddCommand,3,REDIS_CMD_BULK|REDIS_CMD_DENYOOM}, - {"srem",sremCommand,3,REDIS_CMD_BULK}, - {"smove",smoveCommand,4,REDIS_CMD_BULK}, - {"sismember",sismemberCommand,3,REDIS_CMD_BULK}, - {"scard",scardCommand,2,REDIS_CMD_INLINE}, - {"spop",spopCommand,2,REDIS_CMD_INLINE}, - {"srandmember",srandmemberCommand,2,REDIS_CMD_INLINE}, - {"sinter",sinterCommand,-2,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM}, - {"sinterstore",sinterstoreCommand,-3,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM}, - {"sunion",sunionCommand,-2,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM}, - {"sunionstore",sunionstoreCommand,-3,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM}, - {"sdiff",sdiffCommand,-2,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM}, - {"sdiffstore",sdiffstoreCommand,-3,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM}, - {"smembers",sinterCommand,2,REDIS_CMD_INLINE}, - {"zadd",zaddCommand,4,REDIS_CMD_BULK|REDIS_CMD_DENYOOM}, - {"zincrby",zincrbyCommand,4,REDIS_CMD_BULK|REDIS_CMD_DENYOOM}, - {"zrem",zremCommand,3,REDIS_CMD_BULK}, - {"zremrangebyscore",zremrangebyscoreCommand,4,REDIS_CMD_INLINE}, - {"zrange",zrangeCommand,-4,REDIS_CMD_INLINE}, - {"zrangebyscore",zrangebyscoreCommand,-4,REDIS_CMD_INLINE}, - {"zrevrange",zrevrangeCommand,-4,REDIS_CMD_INLINE}, - {"zcard",zcardCommand,2,REDIS_CMD_INLINE}, - {"zscore",zscoreCommand,3,REDIS_CMD_BULK|REDIS_CMD_DENYOOM}, - {"incrby",incrbyCommand,3,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM}, - {"decrby",decrbyCommand,3,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM}, - {"getset",getsetCommand,3,REDIS_CMD_BULK|REDIS_CMD_DENYOOM}, - {"mset",msetCommand,-3,REDIS_CMD_BULK|REDIS_CMD_DENYOOM}, - {"msetnx",msetnxCommand,-3,REDIS_CMD_BULK|REDIS_CMD_DENYOOM}, - {"randomkey",randomkeyCommand,1,REDIS_CMD_INLINE}, - {"select",selectCommand,2,REDIS_CMD_INLINE}, - {"move",moveCommand,3,REDIS_CMD_INLINE}, - {"rename",renameCommand,3,REDIS_CMD_INLINE}, - {"renamenx",renamenxCommand,3,REDIS_CMD_INLINE}, - {"expire",expireCommand,3,REDIS_CMD_INLINE}, - {"expireat",expireatCommand,3,REDIS_CMD_INLINE}, - {"keys",keysCommand,2,REDIS_CMD_INLINE}, - {"dbsize",dbsizeCommand,1,REDIS_CMD_INLINE}, - {"auth",authCommand,2,REDIS_CMD_INLINE}, - {"ping",pingCommand,1,REDIS_CMD_INLINE}, - {"echo",echoCommand,2,REDIS_CMD_BULK}, - {"save",saveCommand,1,REDIS_CMD_INLINE}, - {"bgsave",bgsaveCommand,1,REDIS_CMD_INLINE}, - {"bgrewriteaof",bgrewriteaofCommand,1,REDIS_CMD_INLINE}, - {"shutdown",shutdownCommand,1,REDIS_CMD_INLINE}, - {"lastsave",lastsaveCommand,1,REDIS_CMD_INLINE}, - {"type",typeCommand,2,REDIS_CMD_INLINE}, - {"multi",multiCommand,1,REDIS_CMD_INLINE}, - {"exec",execCommand,1,REDIS_CMD_INLINE}, - {"sync",syncCommand,1,REDIS_CMD_INLINE}, - {"flushdb",flushdbCommand,1,REDIS_CMD_INLINE}, - {"flushall",flushallCommand,1,REDIS_CMD_INLINE}, - {"sort",sortCommand,-2,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM}, - {"info",infoCommand,1,REDIS_CMD_INLINE}, - {"monitor",monitorCommand,1,REDIS_CMD_INLINE}, - {"ttl",ttlCommand,2,REDIS_CMD_INLINE}, - {"slaveof",slaveofCommand,3,REDIS_CMD_INLINE}, - {"debug",debugCommand,-2,REDIS_CMD_INLINE}, - {NULL,NULL,0,0} + {"get",getCommand,2,REDIS_CMD_INLINE,NULL,1,1,1}, + {"set",setCommand,3,REDIS_CMD_BULK|REDIS_CMD_DENYOOM,NULL,0,0,0}, + {"setnx",setnxCommand,3,REDIS_CMD_BULK|REDIS_CMD_DENYOOM,NULL,0,0,0}, + {"append",appendCommand,3,REDIS_CMD_BULK|REDIS_CMD_DENYOOM,NULL,1,1,1}, + {"substr",substrCommand,4,REDIS_CMD_INLINE,NULL,1,1,1}, + {"del",delCommand,-2,REDIS_CMD_INLINE,NULL,0,0,0}, + {"exists",existsCommand,2,REDIS_CMD_INLINE,NULL,1,1,1}, + {"incr",incrCommand,2,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM,NULL,1,1,1}, + {"decr",decrCommand,2,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM,NULL,1,1,1}, + {"mget",mgetCommand,-2,REDIS_CMD_INLINE,NULL,1,-1,1}, + {"rpush",rpushCommand,3,REDIS_CMD_BULK|REDIS_CMD_DENYOOM,NULL,1,1,1}, + {"lpush",lpushCommand,3,REDIS_CMD_BULK|REDIS_CMD_DENYOOM,NULL,1,1,1}, + {"rpop",rpopCommand,2,REDIS_CMD_INLINE,NULL,1,1,1}, + {"lpop",lpopCommand,2,REDIS_CMD_INLINE,NULL,1,1,1}, + {"brpop",brpopCommand,-3,REDIS_CMD_INLINE,NULL,1,1,1}, + {"blpop",blpopCommand,-3,REDIS_CMD_INLINE,NULL,1,1,1}, + {"llen",llenCommand,2,REDIS_CMD_INLINE,NULL,1,1,1}, + {"lindex",lindexCommand,3,REDIS_CMD_INLINE,NULL,1,1,1}, + {"lset",lsetCommand,4,REDIS_CMD_BULK|REDIS_CMD_DENYOOM,NULL,1,1,1}, + {"lrange",lrangeCommand,4,REDIS_CMD_INLINE,NULL,1,1,1}, + {"ltrim",ltrimCommand,4,REDIS_CMD_INLINE,NULL,1,1,1}, + {"lrem",lremCommand,4,REDIS_CMD_BULK,NULL,1,1,1}, + {"rpoplpush",rpoplpushcommand,3,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM,NULL,1,2,1}, + {"sadd",saddCommand,3,REDIS_CMD_BULK|REDIS_CMD_DENYOOM,NULL,1,1,1}, + {"srem",sremCommand,3,REDIS_CMD_BULK,NULL,1,1,1}, + {"smove",smoveCommand,4,REDIS_CMD_BULK,NULL,1,2,1}, + {"sismember",sismemberCommand,3,REDIS_CMD_BULK,NULL,1,1,1}, + {"scard",scardCommand,2,REDIS_CMD_INLINE,NULL,1,1,1}, + {"spop",spopCommand,2,REDIS_CMD_INLINE,NULL,1,1,1}, + {"srandmember",srandmemberCommand,2,REDIS_CMD_INLINE,NULL,1,1,1}, + {"sinter",sinterCommand,-2,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM,NULL,1,-1,1}, + {"sinterstore",sinterstoreCommand,-3,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM,NULL,2,-1,1}, + {"sunion",sunionCommand,-2,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM,NULL,1,-1,1}, + {"sunionstore",sunionstoreCommand,-3,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM,NULL,2,-1,1}, + {"sdiff",sdiffCommand,-2,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM,NULL,1,-1,1}, + {"sdiffstore",sdiffstoreCommand,-3,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM,NULL,2,-1,1}, + {"smembers",sinterCommand,2,REDIS_CMD_INLINE,NULL,1,1,1}, + {"zadd",zaddCommand,4,REDIS_CMD_BULK|REDIS_CMD_DENYOOM,NULL,1,1,1}, + {"zincrby",zincrbyCommand,4,REDIS_CMD_BULK|REDIS_CMD_DENYOOM,NULL,1,1,1}, + {"zrem",zremCommand,3,REDIS_CMD_BULK,NULL,1,1,1}, + {"zremrangebyscore",zremrangebyscoreCommand,4,REDIS_CMD_INLINE,NULL,1,1,1}, + {"zremrangebyrank",zremrangebyrankCommand,4,REDIS_CMD_INLINE,NULL,1,1,1}, + {"zunion",zunionCommand,-4,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM,zunionInterBlockClientOnSwappedKeys,0,0,0}, + {"zinter",zinterCommand,-4,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM,zunionInterBlockClientOnSwappedKeys,0,0,0}, + {"zrange",zrangeCommand,-4,REDIS_CMD_INLINE,NULL,1,1,1}, + {"zrangebyscore",zrangebyscoreCommand,-4,REDIS_CMD_INLINE,NULL,1,1,1}, + {"zcount",zcountCommand,4,REDIS_CMD_INLINE,NULL,1,1,1}, + {"zrevrange",zrevrangeCommand,-4,REDIS_CMD_INLINE,NULL,1,1,1}, + {"zcard",zcardCommand,2,REDIS_CMD_INLINE,NULL,1,1,1}, + {"zscore",zscoreCommand,3,REDIS_CMD_BULK|REDIS_CMD_DENYOOM,NULL,1,1,1}, + {"zrank",zrankCommand,3,REDIS_CMD_BULK,NULL,1,1,1}, + {"zrevrank",zrevrankCommand,3,REDIS_CMD_BULK,NULL,1,1,1}, + {"hset",hsetCommand,4,REDIS_CMD_BULK|REDIS_CMD_DENYOOM,NULL,1,1,1}, + {"hsetnx",hsetnxCommand,4,REDIS_CMD_BULK|REDIS_CMD_DENYOOM,NULL,1,1,1}, + {"hget",hgetCommand,3,REDIS_CMD_BULK,NULL,1,1,1}, + {"hmset",hmsetCommand,-4,REDIS_CMD_BULK|REDIS_CMD_DENYOOM,NULL,1,1,1}, + {"hmget",hmgetCommand,-3,REDIS_CMD_BULK,NULL,1,1,1}, + {"hincrby",hincrbyCommand,4,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM,NULL,1,1,1}, + {"hdel",hdelCommand,3,REDIS_CMD_BULK,NULL,1,1,1}, + {"hlen",hlenCommand,2,REDIS_CMD_INLINE,NULL,1,1,1}, + {"hkeys",hkeysCommand,2,REDIS_CMD_INLINE,NULL,1,1,1}, + {"hvals",hvalsCommand,2,REDIS_CMD_INLINE,NULL,1,1,1}, + {"hgetall",hgetallCommand,2,REDIS_CMD_INLINE,NULL,1,1,1}, + {"hexists",hexistsCommand,3,REDIS_CMD_BULK,NULL,1,1,1}, + {"incrby",incrbyCommand,3,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM,NULL,1,1,1}, + {"decrby",decrbyCommand,3,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM,NULL,1,1,1}, + {"getset",getsetCommand,3,REDIS_CMD_BULK|REDIS_CMD_DENYOOM,NULL,1,1,1}, + {"mset",msetCommand,-3,REDIS_CMD_BULK|REDIS_CMD_DENYOOM,NULL,1,-1,2}, + {"msetnx",msetnxCommand,-3,REDIS_CMD_BULK|REDIS_CMD_DENYOOM,NULL,1,-1,2}, + {"randomkey",randomkeyCommand,1,REDIS_CMD_INLINE,NULL,0,0,0}, + {"select",selectCommand,2,REDIS_CMD_INLINE,NULL,0,0,0}, + {"move",moveCommand,3,REDIS_CMD_INLINE,NULL,1,1,1}, + {"rename",renameCommand,3,REDIS_CMD_INLINE,NULL,1,1,1}, + {"renamenx",renamenxCommand,3,REDIS_CMD_INLINE,NULL,1,1,1}, + {"expire",expireCommand,3,REDIS_CMD_INLINE,NULL,0,0,0}, + {"expireat",expireatCommand,3,REDIS_CMD_INLINE,NULL,0,0,0}, + {"keys",keysCommand,2,REDIS_CMD_INLINE,NULL,0,0,0}, + {"dbsize",dbsizeCommand,1,REDIS_CMD_INLINE,NULL,0,0,0}, + {"auth",authCommand,2,REDIS_CMD_INLINE,NULL,0,0,0}, + {"ping",pingCommand,1,REDIS_CMD_INLINE,NULL,0,0,0}, + {"echo",echoCommand,2,REDIS_CMD_BULK,NULL,0,0,0}, + {"save",saveCommand,1,REDIS_CMD_INLINE,NULL,0,0,0}, + {"bgsave",bgsaveCommand,1,REDIS_CMD_INLINE,NULL,0,0,0}, + {"bgrewriteaof",bgrewriteaofCommand,1,REDIS_CMD_INLINE,NULL,0,0,0}, + {"shutdown",shutdownCommand,1,REDIS_CMD_INLINE,NULL,0,0,0}, + {"lastsave",lastsaveCommand,1,REDIS_CMD_INLINE,NULL,0,0,0}, + {"type",typeCommand,2,REDIS_CMD_INLINE,NULL,1,1,1}, + {"multi",multiCommand,1,REDIS_CMD_INLINE,NULL,0,0,0}, + {"exec",execCommand,1,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM,NULL,0,0,0}, + {"discard",discardCommand,1,REDIS_CMD_INLINE,NULL,0,0,0}, + {"sync",syncCommand,1,REDIS_CMD_INLINE,NULL,0,0,0}, + {"flushdb",flushdbCommand,1,REDIS_CMD_INLINE,NULL,0,0,0}, + {"flushall",flushallCommand,1,REDIS_CMD_INLINE,NULL,0,0,0}, + {"sort",sortCommand,-2,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM,NULL,1,1,1}, + {"info",infoCommand,1,REDIS_CMD_INLINE,NULL,0,0,0}, + {"monitor",monitorCommand,1,REDIS_CMD_INLINE,NULL,0,0,0}, + {"ttl",ttlCommand,2,REDIS_CMD_INLINE,NULL,1,1,1}, + {"slaveof",slaveofCommand,3,REDIS_CMD_INLINE,NULL,0,0,0}, + {"debug",debugCommand,-2,REDIS_CMD_INLINE,NULL,0,0,0}, + {"config",configCommand,-2,REDIS_CMD_BULK,NULL,0,0,0}, + {"subscribe",subscribeCommand,-2,REDIS_CMD_INLINE,NULL,0,0,0}, + {"unsubscribe",unsubscribeCommand,-1,REDIS_CMD_INLINE,NULL,0,0,0}, + {"psubscribe",psubscribeCommand,-2,REDIS_CMD_INLINE,NULL,0,0,0}, + {"punsubscribe",punsubscribeCommand,-1,REDIS_CMD_INLINE,NULL,0,0,0}, + {"publish",publishCommand,3,REDIS_CMD_BULK|REDIS_CMD_FORCE_REPLICATION,NULL,0,0,0}, + {NULL,NULL,0,0,NULL,0,0,0} }; /*============================ Utility functions ============================ */ /* Glob-style pattern matching. */ -int stringmatchlen(const char *pattern, int patternLen, +static int stringmatchlen(const char *pattern, int patternLen, const char *string, int stringLen, int nocase) { while(patternLen) { @@ -735,6 +966,10 @@ int stringmatchlen(const char *pattern, int patternLen, return 0; } +static int stringmatch(const char *pattern, const char *string, int nocase) { + return stringmatchlen(pattern,strlen(pattern),string,strlen(string),nocase); +} + static void redisLog(int level, const char *fmt, ...) { va_list ap; FILE *fp; @@ -744,13 +979,13 @@ static void redisLog(int level, const char *fmt, ...) { va_start(ap, fmt); if (level >= server.verbosity) { - char *c = ".-*"; + char *c = ".-*#"; char buf[64]; time_t now; now = time(NULL); strftime(buf,64,"%d %b %H:%M:%S",localtime(&now)); - fprintf(fp,"%s %c ",buf,c[level]); + fprintf(fp,"[%d] %s %c ",(int)getpid(),buf,c[level]); vfprintf(fp, fmt, ap); fprintf(fp,"\n"); fflush(fp); @@ -794,6 +1029,7 @@ static void dictRedisObjectDestructor(void *privdata, void *val) { DICT_NOTUSED(privdata); + if (val == NULL) return; /* Values of swapped out keys as set to NULL */ decrRefCount(val); } @@ -815,6 +1051,10 @@ static int dictEncObjKeyCompare(void *privdata, const void *key1, robj *o1 = (robj*) key1, *o2 = (robj*) key2; int cmp; + if (o1->encoding == REDIS_ENCODING_INT && + o2->encoding == REDIS_ENCODING_INT && + o1->ptr == o2->ptr) return 1; + o1 = getDecodedObject(o1); o2 = getDecodedObject(o2); cmp = sdsDictKeyCompare(privdata,o1->ptr,o2->ptr); @@ -826,12 +1066,27 @@ static int dictEncObjKeyCompare(void *privdata, const void *key1, static unsigned int dictEncObjHash(const void *key) { robj *o = (robj*) key; - o = getDecodedObject(o); - unsigned int hash = dictGenHashFunction(o->ptr, sdslen((sds)o->ptr)); - decrRefCount(o); - return hash; + if (o->encoding == REDIS_ENCODING_RAW) { + return dictGenHashFunction(o->ptr, sdslen((sds)o->ptr)); + } else { + if (o->encoding == REDIS_ENCODING_INT) { + char buf[32]; + int len; + + len = snprintf(buf,32,"%ld",(long)o->ptr); + return dictGenHashFunction((unsigned char*)buf, len); + } else { + unsigned int hash; + + o = getDecodedObject(o); + hash = dictGenHashFunction(o->ptr, sdslen((sds)o->ptr)); + decrRefCount(o); + return hash; + } + } } +/* Sets type and expires */ static dictType setDictType = { dictEncObjHash, /* hash function */ NULL, /* key dup */ @@ -841,6 +1096,7 @@ static dictType setDictType = { NULL /* val destructor */ }; +/* Sorted sets hash (note: a skiplist is used in addition to the hash table) */ static dictType zsetDictType = { dictEncObjHash, /* hash function */ NULL, /* key dup */ @@ -850,7 +1106,8 @@ static dictType zsetDictType = { dictVanillaFree /* val destructor of malloc(sizeof(double)) */ }; -static dictType hashDictType = { +/* Db->dict */ +static dictType dbDictType = { dictObjHash, /* hash function */ NULL, /* key dup */ NULL, /* val dup */ @@ -859,8 +1116,29 @@ static dictType hashDictType = { dictRedisObjectDestructor /* val destructor */ }; +/* Db->expires */ +static dictType keyptrDictType = { + dictObjHash, /* hash function */ + NULL, /* key dup */ + NULL, /* val dup */ + dictObjKeyCompare, /* key compare */ + dictRedisObjectDestructor, /* key destructor */ + NULL /* val destructor */ +}; + +/* Hash type hash table (note that small hashes are represented with zimpaps) */ +static dictType hashDictType = { + dictEncObjHash, /* hash function */ + NULL, /* key dup */ + NULL, /* val dup */ + dictEncObjKeyCompare, /* key compare */ + dictRedisObjectDestructor, /* key destructor */ + dictRedisObjectDestructor /* val destructor */ +}; + /* Keylist hash table type has unencoded redis objects as keys and - * lists as values. It's used for blocking operations (BLPOP) */ + * lists as values. It's used for blocking operations (BLPOP) and to + * map swapped keys to a list of clients waiting for this keys to be loaded. */ static dictType keylistDictType = { dictObjHash, /* hash function */ NULL, /* key dup */ @@ -870,6 +1148,8 @@ static dictType keylistDictType = { dictListDestructor /* val destructor */ }; +static void version(); + /* ========================= Random utility functions ======================= */ /* Redis generally does not try to recover from out of memory conditions @@ -888,15 +1168,25 @@ static void closeTimedoutClients(void) { redisClient *c; listNode *ln; time_t now = time(NULL); + listIter li; - listRewind(server.clients); - while ((ln = listYield(server.clients)) != NULL) { + listRewind(server.clients,&li); + while ((ln = listNext(&li)) != NULL) { c = listNodeValue(ln); - if (!(c->flags & REDIS_SLAVE) && /* no timeout for slaves */ + if (server.maxidletime && + !(c->flags & REDIS_SLAVE) && /* no timeout for slaves */ !(c->flags & REDIS_MASTER) && /* no timeout for masters */ - (now - c->lastinteraction > server.maxidletime)) { - redisLog(REDIS_DEBUG,"Closing idle client"); + dictSize(c->pubsub_channels) == 0 && /* no timeout for pubsub */ + listLength(c->pubsub_patterns) == 0 && + (now - c->lastinteraction > server.maxidletime)) + { + redisLog(REDIS_VERBOSE,"Closing idle client"); freeClient(c); + } else if (c->flags & REDIS_BLOCKED) { + if (c->blockingto != 0 && c->blockingto < now) { + addReply(c,shared.nullmultibulk); + unblockClientWaitingData(c); + } } } } @@ -916,16 +1206,28 @@ static void tryResizeHashTables(void) { int j; for (j = 0; j < server.dbnum; j++) { - if (htNeedsResize(server.db[j].dict)) { - redisLog(REDIS_DEBUG,"The hash table %d is too sparse, resize it...",j); + if (htNeedsResize(server.db[j].dict)) dictResize(server.db[j].dict); - redisLog(REDIS_DEBUG,"Hash table %d resized.",j); - } if (htNeedsResize(server.db[j].expires)) dictResize(server.db[j].expires); } } +/* Our hash table implementation performs rehashing incrementally while + * we write/read from the hash table. Still if the server is idle, the hash + * table will use two tables for a long time. So we try to use 1 millisecond + * of CPU time at every serverCron() loop in order to rehash some key. */ +static void incrementallyRehash(void) { + int j; + + for (j = 0; j < server.dbnum; j++) { + if (dictIsRehashing(server.db[j].dict)) { + dictRehashMilliseconds(server.db[j].dict,1); + break; /* already used our millisecond for this loop... */ + } + } +} + /* A background saving child (BGSAVE) terminated its work. Handle this. */ void backgroundSaveDoneHandler(int statloc) { int exitcode = WEXITSTATUS(statloc); @@ -940,7 +1242,7 @@ void backgroundSaveDoneHandler(int statloc) { redisLog(REDIS_WARNING, "Background saving error"); } else { redisLog(REDIS_WARNING, - "Background saving terminated by signal"); + "Background saving terminated by signal %d", WTERMSIG(statloc)); rdbRemoveTempFile(server.bgsavechildpid); } server.bgsavechildpid = -1; @@ -1001,7 +1303,8 @@ void backgroundRewriteDoneHandler(int statloc) { redisLog(REDIS_WARNING, "Background append only file rewriting error"); } else { redisLog(REDIS_WARNING, - "Background append only file rewriting terminated by signal"); + "Background append only file rewriting terminated by signal %d", + WTERMSIG(statloc)); } cleanup: sdsfree(server.bgrewritebuf); @@ -1010,14 +1313,30 @@ cleanup: server.bgrewritechildpid = -1; } +/* This function is called once a background process of some kind terminates, + * as we want to avoid resizing the hash tables when there is a child in order + * to play well with copy-on-write (otherwise when a resize happens lots of + * memory pages are copied). The goal of this function is to update the ability + * for dict.c to resize the hash tables accordingly to the fact we have o not + * running childs. */ +static void updateDictResizePolicy(void) { + if (server.bgsavechildpid == -1 && server.bgrewritechildpid == -1) + dictEnableResize(); + else + dictDisableResize(); +} + static int serverCron(struct aeEventLoop *eventLoop, long long id, void *clientData) { int j, loops = server.cronloops++; REDIS_NOTUSED(eventLoop); REDIS_NOTUSED(id); REDIS_NOTUSED(clientData); - /* Update the global state with the amount of used memory */ - server.usedmemory = zmalloc_used_memory(); + /* We take a cached value of the unix time in the global state because + * with virtual memory and aging there is to store the current time + * in objects at every object access, and accuracy is not needed. + * To access a global var is faster than calling time(NULL) */ + server.unixtime = time(NULL); /* Show some info about non-empty databases */ for (j = 0; j < server.dbnum; j++) { @@ -1026,8 +1345,8 @@ static int serverCron(struct aeEventLoop *eventLoop, long long id, void *clientD size = dictSlots(server.db[j].dict); used = dictSize(server.db[j].dict); vkeys = dictSize(server.db[j].expires); - if (!(loops % 5) && (used || vkeys)) { - redisLog(REDIS_DEBUG,"DB %d: %lld keys (%lld volatile) in %lld slots HT.",j,used,vkeys,size); + if (!(loops % 50) && (used || vkeys)) { + redisLog(REDIS_VERBOSE,"DB %d: %lld keys (%lld volatile) in %lld slots HT.",j,used,vkeys,size); /* dictPrintStats(server.dict); */ } } @@ -1038,19 +1357,21 @@ static int serverCron(struct aeEventLoop *eventLoop, long long id, void *clientD * if we resize the HT while there is the saving child at work actually * a lot of memory movements in the parent will cause a lot of pages * copied. */ - if (server.bgsavechildpid == -1) tryResizeHashTables(); + if (server.bgsavechildpid == -1 && server.bgrewritechildpid == -1) { + if (!(loops % 10)) tryResizeHashTables(); + if (server.activerehashing) incrementallyRehash(); + } /* Show information about connected clients */ - if (!(loops % 5)) { - redisLog(REDIS_DEBUG,"%d clients connected (%d slaves), %zu bytes in use, %d shared objects", + if (!(loops % 50)) { + redisLog(REDIS_VERBOSE,"%d clients connected (%d slaves), %zu bytes in use", listLength(server.clients)-listLength(server.slaves), listLength(server.slaves), - server.usedmemory, - dictSize(server.sharingpool)); + zmalloc_used_memory()); } /* Close connections of timedout clients */ - if (server.maxidletime && !(loops % 10)) + if ((server.maxidletime && !(loops % 100)) || server.blpop_blocked_clients) closeTimedoutClients(); /* Check if a background saving or AOF rewrite in progress terminated */ @@ -1064,6 +1385,7 @@ static int serverCron(struct aeEventLoop *eventLoop, long long id, void *clientD } else { backgroundRewriteDoneHandler(statloc); } + updateDictResizePolicy(); } } else { /* If there is not a background saving in progress check if @@ -1093,7 +1415,7 @@ static int serverCron(struct aeEventLoop *eventLoop, long long id, void *clientD /* Continue to expire if at the end of the cycle more than 25% * of the keys were expired. */ do { - int num = dictSize(db->expires); + long num = dictSize(db->expires); time_t now = time(NULL); expired = 0; @@ -1108,22 +1430,83 @@ static int serverCron(struct aeEventLoop *eventLoop, long long id, void *clientD if (now > t) { deleteKey(db,dictGetEntryKey(de)); expired++; + server.stat_expiredkeys++; } } } while (expired > REDIS_EXPIRELOOKUPS_PER_CRON/4); } + /* Swap a few keys on disk if we are over the memory limit and VM + * is enbled. Try to free objects from the free list first. */ + if (vmCanSwapOut()) { + while (server.vm_enabled && zmalloc_used_memory() > + server.vm_max_memory) + { + int retval; + + if (tryFreeOneObjectFromFreelist() == REDIS_OK) continue; + retval = (server.vm_max_threads == 0) ? + vmSwapOneObjectBlocking() : + vmSwapOneObjectThreaded(); + if (retval == REDIS_ERR && !(loops % 300) && + zmalloc_used_memory() > + (server.vm_max_memory+server.vm_max_memory/10)) + { + redisLog(REDIS_WARNING,"WARNING: vm-max-memory limit exceeded by more than 10%% but unable to swap more objects out!"); + } + /* Note that when using threade I/O we free just one object, + * because anyway when the I/O thread in charge to swap this + * object out will finish, the handler of completed jobs + * will try to swap more objects if we are still out of memory. */ + if (retval == REDIS_ERR || server.vm_max_threads > 0) break; + } + } + /* Check if we should connect to a MASTER */ - if (server.replstate == REDIS_REPL_CONNECT) { + if (server.replstate == REDIS_REPL_CONNECT && !(loops % 10)) { redisLog(REDIS_NOTICE,"Connecting to MASTER..."); if (syncWithMaster() == REDIS_OK) { redisLog(REDIS_NOTICE,"MASTER <-> SLAVE sync succeeded"); } } - return 1000; + return 100; +} + +/* This function gets called every time Redis is entering the + * main loop of the event driven library, that is, before to sleep + * for ready file descriptors. */ +static void beforeSleep(struct aeEventLoop *eventLoop) { + REDIS_NOTUSED(eventLoop); + + if (server.vm_enabled && listLength(server.io_ready_clients)) { + listIter li; + listNode *ln; + + listRewind(server.io_ready_clients,&li); + while((ln = listNext(&li))) { + redisClient *c = ln->value; + struct redisCommand *cmd; + + /* Resume the client. */ + listDelNode(server.io_ready_clients,ln); + c->flags &= (~REDIS_IO_WAIT); + server.vm_blocked_clients--; + aeCreateFileEvent(server.el, c->fd, AE_READABLE, + readQueryFromClient, c); + cmd = lookupCommand(c->argv[0]->ptr); + assert(cmd != NULL); + call(c,cmd); + resetClient(c); + /* There may be more data to process in the input buffer. */ + if (c->querybuf && sdslen(c->querybuf) > 0) + processInputBuffer(c); + } + } } static void createSharedObjects(void) { + int j; + shared.crlf = createObject(REDIS_STRING,sdsnew("\r\n")); shared.ok = createObject(REDIS_STRING,sdsnew("+OK\r\n")); shared.err = createObject(REDIS_STRING,sdsnew("-ERR\r\n")); @@ -1158,6 +1541,16 @@ static void createSharedObjects(void) { shared.select7 = createStringObject("select 7\r\n",10); shared.select8 = createStringObject("select 8\r\n",10); shared.select9 = createStringObject("select 9\r\n",10); + shared.messagebulk = createStringObject("$7\r\nmessage\r\n",13); + shared.subscribebulk = createStringObject("$9\r\nsubscribe\r\n",15); + shared.unsubscribebulk = createStringObject("$11\r\nunsubscribe\r\n",18); + shared.psubscribebulk = createStringObject("$10\r\npsubscribe\r\n",17); + shared.punsubscribebulk = createStringObject("$12\r\npunsubscribe\r\n",19); + shared.mbulk3 = createStringObject("*3\r\n",4); + for (j = 0; j < REDIS_SHARED_INTEGERS; j++) { + shared.integers[j] = createObject(REDIS_STRING,(void*)(long)j); + shared.integers[j]->encoding = REDIS_ENCODING_INT; + } } static void appendServerSaveParams(time_t seconds, int changes) { @@ -1176,7 +1569,7 @@ static void resetServerSaveParams() { static void initServerConfig() { server.dbnum = REDIS_DEFAULT_DBNUM; server.port = REDIS_SERVERPORT; - server.verbosity = REDIS_DEBUG; + server.verbosity = REDIS_VERBOSE; server.maxidletime = REDIS_MAXIDLETIME; server.saveparams = NULL; server.logfile = NULL; /* NULL = log on standard output */ @@ -1188,15 +1581,26 @@ static void initServerConfig() { server.lastfsync = time(NULL); server.appendfd = -1; server.appendseldb = -1; /* Make sure the first time will not match */ - server.pidfile = "/var/run/redis.pid"; - server.dbfilename = "dump.rdb"; - server.appendfilename = "appendonly.aof"; + server.pidfile = zstrdup("/var/run/redis.pid"); + server.dbfilename = zstrdup("dump.rdb"); + server.appendfilename = zstrdup("appendonly.aof"); server.requirepass = NULL; server.shareobjects = 0; server.rdbcompression = 1; - server.sharingpoolsize = 1024; + server.activerehashing = 1; server.maxclients = 0; + server.blpop_blocked_clients = 0; server.maxmemory = 0; + server.vm_enabled = 0; + server.vm_swap_file = zstrdup("/tmp/redis-%p.vm"); + server.vm_page_size = 256; /* 256 bytes per page */ + server.vm_pages = 1024*1024*100; /* 104 millions of pages */ + server.vm_max_memory = 1024LL*1024*1024*1; /* 1 GB of RAM */ + server.vm_max_threads = 4; + server.vm_blocked_clients = 0; + server.hash_max_zipmap_entries = REDIS_HASH_MAX_ZIPMAP_ENTRIES; + server.hash_max_zipmap_value = REDIS_HASH_MAX_ZIPMAP_VALUE; + resetServerSaveParams(); appendServerSaveParams(60*60,1); /* save after 1 hour and 1 change */ @@ -1224,6 +1628,11 @@ static void initServer() { signal(SIGPIPE, SIG_IGN); setupSigSegvAction(); + server.devnull = fopen("/dev/null","w"); + if (server.devnull == NULL) { + redisLog(REDIS_WARNING, "Can't open /dev/null: %s", server.neterr); + exit(1); + } server.clients = listCreate(); server.slaves = listCreate(); server.monitors = listCreate(); @@ -1231,29 +1640,37 @@ static void initServer() { createSharedObjects(); server.el = aeCreateEventLoop(); server.db = zmalloc(sizeof(redisDb)*server.dbnum); - server.sharingpool = dictCreate(&setDictType,NULL); server.fd = anetTcpServer(server.neterr, server.port, server.bindaddr); if (server.fd == -1) { redisLog(REDIS_WARNING, "Opening TCP port: %s", server.neterr); exit(1); } for (j = 0; j < server.dbnum; j++) { - server.db[j].dict = dictCreate(&hashDictType,NULL); - server.db[j].expires = dictCreate(&setDictType,NULL); + server.db[j].dict = dictCreate(&dbDictType,NULL); + server.db[j].expires = dictCreate(&keyptrDictType,NULL); server.db[j].blockingkeys = dictCreate(&keylistDictType,NULL); + if (server.vm_enabled) + server.db[j].io_keys = dictCreate(&keylistDictType,NULL); server.db[j].id = j; } + server.pubsub_channels = dictCreate(&keylistDictType,NULL); + server.pubsub_patterns = listCreate(); + listSetFreeMethod(server.pubsub_patterns,freePubsubPattern); + listSetMatchMethod(server.pubsub_patterns,listMatchPubsubPattern); server.cronloops = 0; server.bgsavechildpid = -1; server.bgrewritechildpid = -1; server.bgrewritebuf = sdsempty(); server.lastsave = time(NULL); server.dirty = 0; - server.usedmemory = 0; server.stat_numcommands = 0; server.stat_numconnections = 0; + server.stat_expiredkeys = 0; server.stat_starttime = time(NULL); + server.unixtime = time(NULL); aeCreateTimeEvent(server.el, 1, serverCron, NULL, NULL); + if (aeCreateFileEvent(server.el, server.fd, AE_READABLE, + acceptHandler, NULL) == AE_ERR) oom("creating file event"); if (server.appendonly) { server.appendfd = open(server.appendfilename,O_WRONLY|O_APPEND|O_CREAT,0644); @@ -1263,6 +1680,8 @@ static void initServer() { exit(1); } } + + if (server.vm_enabled) vmInit(); } /* Empty the whole database */ @@ -1296,7 +1715,7 @@ static void loadServerConfig(char *filename) { fp = stdin; else { if ((fp = fopen(filename,"r")) == NULL) { - redisLog(REDIS_WARNING,"Fatal error, can't open config file"); + redisLog(REDIS_WARNING, "Fatal error, can't open config file '%s'", filename); exit(1); } } @@ -1347,6 +1766,7 @@ static void loadServerConfig(char *filename) { } } else if (!strcasecmp(argv[0],"loglevel") && argc == 2) { if (!strcasecmp(argv[1],"debug")) server.verbosity = REDIS_DEBUG; + else if (!strcasecmp(argv[1],"verbose")) server.verbosity = REDIS_VERBOSE; else if (!strcasecmp(argv[1],"notice")) server.verbosity = REDIS_NOTICE; else if (!strcasecmp(argv[1],"warning")) server.verbosity = REDIS_WARNING; else { @@ -1377,6 +1797,8 @@ static void loadServerConfig(char *filename) { if (server.dbnum < 1) { err = "Invalid number of databases"; goto loaderr; } + } else if (!strcasecmp(argv[0],"include") && argc == 2) { + loadServerConfig(argv[1]); } else if (!strcasecmp(argv[0],"maxclients") && argc == 2) { server.maxclients = atoi(argv[1]); } else if (!strcasecmp(argv[0],"maxmemory") && argc == 2) { @@ -1399,10 +1821,9 @@ static void loadServerConfig(char *filename) { if ((server.rdbcompression = yesnotoi(argv[1])) == -1) { err = "argument must be 'yes' or 'no'"; goto loaderr; } - } else if (!strcasecmp(argv[0],"shareobjectspoolsize") && argc == 2) { - server.sharingpoolsize = atoi(argv[1]); - if (server.sharingpoolsize < 1) { - err = "invalid object sharing pool size"; goto loaderr; + } else if (!strcasecmp(argv[0],"activerehashing") && argc == 2) { + if ((server.activerehashing = yesnotoi(argv[1])) == -1) { + err = "argument must be 'yes' or 'no'"; goto loaderr; } } else if (!strcasecmp(argv[0],"daemonize") && argc == 2) { if ((server.daemonize = yesnotoi(argv[1])) == -1) { @@ -1424,11 +1845,34 @@ static void loadServerConfig(char *filename) { goto loaderr; } } else if (!strcasecmp(argv[0],"requirepass") && argc == 2) { - server.requirepass = zstrdup(argv[1]); + server.requirepass = zstrdup(argv[1]); } else if (!strcasecmp(argv[0],"pidfile") && argc == 2) { - server.pidfile = zstrdup(argv[1]); + zfree(server.pidfile); + server.pidfile = zstrdup(argv[1]); } else if (!strcasecmp(argv[0],"dbfilename") && argc == 2) { - server.dbfilename = zstrdup(argv[1]); + zfree(server.dbfilename); + server.dbfilename = zstrdup(argv[1]); + } else if (!strcasecmp(argv[0],"vm-enabled") && argc == 2) { + if ((server.vm_enabled = yesnotoi(argv[1])) == -1) { + err = "argument must be 'yes' or 'no'"; goto loaderr; + } + } else if (!strcasecmp(argv[0],"vm-swap-file") && argc == 2) { + zfree(server.vm_swap_file); + server.vm_swap_file = zstrdup(argv[1]); + } else if (!strcasecmp(argv[0],"vm-max-memory") && argc == 2) { + server.vm_max_memory = strtoll(argv[1], NULL, 10); + } else if (!strcasecmp(argv[0],"vm-page-size") && argc == 2) { + server.vm_page_size = strtoll(argv[1], NULL, 10); + } else if (!strcasecmp(argv[0],"vm-pages") && argc == 2) { + server.vm_pages = strtoll(argv[1], NULL, 10); + } else if (!strcasecmp(argv[0],"vm-max-threads") && argc == 2) { + server.vm_max_threads = strtoll(argv[1], NULL, 10); + } else if (!strcasecmp(argv[0],"hash-max-zipmap-entries") && argc == 2){ + server.hash_max_zipmap_entries = strtol(argv[1], NULL, 10); + } else if (!strcasecmp(argv[0],"hash-max-zipmap-value") && argc == 2){ + server.hash_max_zipmap_value = strtol(argv[1], NULL, 10); + } else if (!strcasecmp(argv[0],"vm-max-threads") && argc == 2) { + server.vm_max_threads = strtoll(argv[1], NULL, 10); } else { err = "Bad directive or wrong number of arguments"; goto loaderr; } @@ -1463,23 +1907,44 @@ static void freeClient(redisClient *c) { listNode *ln; /* Note that if the client we are freeing is blocked into a blocking - * call, we have to set querybuf to NULL *before* to call unblockClient() - * to avoid processInputBuffer() will get called. Also it is important - * to remove the file events after this, because this call adds - * the READABLE event. */ + * call, we have to set querybuf to NULL *before* to call + * unblockClientWaitingData() to avoid processInputBuffer() will get + * called. Also it is important to remove the file events after + * this, because this call adds the READABLE event. */ sdsfree(c->querybuf); c->querybuf = NULL; if (c->flags & REDIS_BLOCKED) - unblockClient(c); - + unblockClientWaitingData(c); + + /* Unsubscribe from all the pubsub channels */ + pubsubUnsubscribeAllChannels(c,0); + pubsubUnsubscribeAllPatterns(c,0); + dictRelease(c->pubsub_channels); + listRelease(c->pubsub_patterns); + /* Obvious cleanup */ aeDeleteFileEvent(server.el,c->fd,AE_READABLE); aeDeleteFileEvent(server.el,c->fd,AE_WRITABLE); listRelease(c->reply); freeClientArgv(c); close(c->fd); + /* Remove from the list of clients */ ln = listSearchKey(server.clients,c); redisAssert(ln != NULL); listDelNode(server.clients,ln); + /* Remove from the list of clients waiting for swapped keys */ + if (c->flags & REDIS_IO_WAIT && listLength(c->io_keys) == 0) { + ln = listSearchKey(server.io_ready_clients,c); + if (ln) { + listDelNode(server.io_ready_clients,ln); + server.vm_blocked_clients--; + } + } + while (server.vm_enabled && listLength(c->io_keys)) { + ln = listFirst(c->io_keys); + dontWaitForSwappedKey(c,ln->value); + } + listRelease(c->io_keys); + /* Master/slave cleanup */ if (c->flags & REDIS_SLAVE) { if (c->replstate == REDIS_REPL_SEND_BULK && c->repldbfd != -1) close(c->repldbfd); @@ -1492,6 +1957,7 @@ static void freeClient(redisClient *c) { server.master = NULL; server.replstate = REDIS_REPL_CONNECT; } + /* Release memory */ zfree(c->argv); zfree(c->mbargv); freeClientMultiState(c); @@ -1503,10 +1969,11 @@ static void glueReplyBuffersIfNeeded(redisClient *c) { int copylen = 0; char buf[GLUEREPLY_UP_TO]; listNode *ln; + listIter li; robj *o; - listRewind(c->reply); - while((ln = listYield(c->reply))) { + listRewind(c->reply,&li); + while((ln = listNext(&li))) { int objlen; o = ln->value; @@ -1534,7 +2001,7 @@ static void sendReplyToClient(aeEventLoop *el, int fd, void *privdata, int mask) /* Use writev() if we have enough buffers to send */ if (!server.glueoutputbuf && - listLength(c->reply) > REDIS_WRITEV_THRESHOLD && + listLength(c->reply) > REDIS_WRITEV_THRESHOLD && !(c->flags & REDIS_MASTER)) { sendReplyToClientWritev(el, fd, privdata, mask); @@ -1578,7 +2045,7 @@ static void sendReplyToClient(aeEventLoop *el, int fd, void *privdata, int mask) if (errno == EAGAIN) { nwritten = 0; } else { - redisLog(REDIS_DEBUG, + redisLog(REDIS_VERBOSE, "Error writing to client: %s", strerror(errno)); freeClient(c); return; @@ -1612,7 +2079,7 @@ static void sendReplyToClientWritev(aeEventLoop *el, int fd, void *privdata, int o = listNodeValue(node); objlen = sdslen(o->ptr); - if (totwritten + objlen - offset > REDIS_MAX_WRITE_PER_EVENT) + if (totwritten + objlen - offset > REDIS_MAX_WRITE_PER_EVENT) break; if(ion == REDIS_WRITEV_IOVEC_COUNT) @@ -1631,7 +2098,7 @@ static void sendReplyToClientWritev(aeEventLoop *el, int fd, void *privdata, int /* write all collected blocks at once */ if((nwritten = writev(fd, iov, ion)) < 0) { if (errno != EAGAIN) { - redisLog(REDIS_DEBUG, + redisLog(REDIS_VERBOSE, "Error writing to client: %s", strerror(errno)); freeClient(c); return; @@ -1660,7 +2127,7 @@ static void sendReplyToClientWritev(aeEventLoop *el, int fd, void *privdata, int } } - if (totwritten > 0) + if (totwritten > 0) c->lastinteraction = time(NULL); if (listLength(c->reply) == 0) { @@ -1691,12 +2158,15 @@ static void call(redisClient *c, struct redisCommand *cmd) { dirty = server.dirty; cmd->proc(c); - if (server.appendonly && server.dirty-dirty) + dirty = server.dirty-dirty; + + if (server.appendonly && dirty) feedAppendOnlyFile(cmd,c->db->id,c->argv,c->argc); - if (server.dirty-dirty && listLength(server.slaves)) - replicationFeedSlaves(server.slaves,cmd,c->db->id,c->argv,c->argc); + if ((dirty || cmd->flags & REDIS_CMD_FORCE_REPLICATION) && + listLength(server.slaves)) + replicationFeedSlaves(server.slaves,c->db->id,c->argv,c->argc); if (listLength(server.monitors)) - replicationFeedSlaves(server.monitors,cmd,c->db->id,c->argv,c->argc); + replicationFeedSlaves(server.monitors,c->db->id,c->argv,c->argc); server.stat_numcommands++; } @@ -1788,6 +2258,9 @@ static int processCommand(redisClient *c) { freeClient(c); return 0; } + + /* Now lookup the command and check ASAP about trivial error conditions + * such wrong arity, bad command name and so forth. */ cmd = lookupCommand(c->argv[0]->ptr); if (!cmd) { addReplySds(c, @@ -1803,11 +2276,8 @@ static int processCommand(redisClient *c) { cmd->name)); resetClient(c); return 1; - } else if (server.maxmemory && cmd->flags & REDIS_CMD_DENYOOM && zmalloc_used_memory() > server.maxmemory) { - addReplySds(c,sdsnew("-ERR command not allowed when used memory > 'maxmemory'\r\n")); - resetClient(c); - return 1; } else if (cmd->flags & REDIS_CMD_BULK && c->bulklen == -1) { + /* This is a bulk command, we have to read the last argument yet. */ int bulklen = atoi(c->argv[c->argc-1]->ptr); decrRefCount(c->argv[c->argc-1]); @@ -1829,18 +2299,14 @@ static int processCommand(redisClient *c) { c->argc++; c->querybuf = sdsrange(c->querybuf,c->bulklen,-1); } else { + /* Otherwise return... there is to read the last argument + * from the socket. */ return 1; } } - /* Let's try to share objects on the command arguments vector */ - if (server.shareobjects) { - int j; - for(j = 1; j < c->argc; j++) - c->argv[j] = tryObjectSharing(c->argv[j]); - } /* Let's try to encode the bulk object to save space. */ if (cmd->flags & REDIS_CMD_BULK) - tryObjectEncoding(c->argv[c->argc-1]); + c->argv[c->argc-1] = tryObjectEncoding(c->argv[c->argc-1]); /* Check if the user is authenticated */ if (server.requirepass && !c->authenticated && cmd->proc != authCommand) { @@ -1849,57 +2315,77 @@ static int processCommand(redisClient *c) { return 1; } + /* Handle the maxmemory directive */ + if (server.maxmemory && (cmd->flags & REDIS_CMD_DENYOOM) && + zmalloc_used_memory() > server.maxmemory) + { + addReplySds(c,sdsnew("-ERR command not allowed when used memory > 'maxmemory'\r\n")); + resetClient(c); + return 1; + } + + /* Only allow SUBSCRIBE and UNSUBSCRIBE in the context of Pub/Sub */ + if ((dictSize(c->pubsub_channels) > 0 || listLength(c->pubsub_patterns) > 0) + && + cmd->proc != subscribeCommand && cmd->proc != unsubscribeCommand && + cmd->proc != psubscribeCommand && cmd->proc != punsubscribeCommand) { + addReplySds(c,sdsnew("-ERR only (P)SUBSCRIBE / (P)UNSUBSCRIBE / QUIT allowed in this context\r\n")); + resetClient(c); + return 1; + } + /* Exec the command */ - if (c->flags & REDIS_MULTI && cmd->proc != execCommand) { + if (c->flags & REDIS_MULTI && cmd->proc != execCommand && cmd->proc != discardCommand) { queueMultiCommand(c,cmd); addReply(c,shared.queued); } else { + if (server.vm_enabled && server.vm_max_threads > 0 && + blockClientOnSwappedKeys(cmd,c)) return 1; call(c,cmd); } /* Prepare the client for the next command */ - if (c->flags & REDIS_CLOSE) { - freeClient(c); - return 0; - } resetClient(c); return 1; } -static void replicationFeedSlaves(list *slaves, struct redisCommand *cmd, int dictid, robj **argv, int argc) { +static void replicationFeedSlaves(list *slaves, int dictid, robj **argv, int argc) { listNode *ln; + listIter li; int outc = 0, j; robj **outv; - /* (args*2)+1 is enough room for args, spaces, newlines */ - robj *static_outv[REDIS_STATIC_ARGS*2+1]; + /* We need 1+(ARGS*3) objects since commands are using the new protocol + * and we one 1 object for the first "*\r\n" multibulk count, then + * for every additional object we have "$\r\n" + object + "\r\n". */ + robj *static_outv[REDIS_STATIC_ARGS*3+1]; + robj *lenobj; if (argc <= REDIS_STATIC_ARGS) { outv = static_outv; } else { - outv = zmalloc(sizeof(robj*)*(argc*2+1)); + outv = zmalloc(sizeof(robj*)*(argc*3+1)); } - - for (j = 0; j < argc; j++) { - if (j != 0) outv[outc++] = shared.space; - if ((cmd->flags & REDIS_CMD_BULK) && j == argc-1) { - robj *lenobj; - lenobj = createObject(REDIS_STRING, - sdscatprintf(sdsempty(),"%lu\r\n", - (unsigned long) stringObjectLen(argv[j]))); - lenobj->refcount = 0; - outv[outc++] = lenobj; - } + lenobj = createObject(REDIS_STRING, + sdscatprintf(sdsempty(), "*%d\r\n", argc)); + lenobj->refcount = 0; + outv[outc++] = lenobj; + for (j = 0; j < argc; j++) { + lenobj = createObject(REDIS_STRING, + sdscatprintf(sdsempty(),"$%lu\r\n", + (unsigned long) stringObjectLen(argv[j]))); + lenobj->refcount = 0; + outv[outc++] = lenobj; outv[outc++] = argv[j]; + outv[outc++] = shared.crlf; } - outv[outc++] = shared.crlf; /* Increment all the refcounts at start and decrement at end in order to * be sure to free objects if there is no slave in a replication state * able to be feed with commands */ for (j = 0; j < outc; j++) incrRefCount(outv[j]); - listRewind(slaves); - while((ln = listYield(slaves))) { + listRewind(slaves,&li); + while((ln = listNext(&li))) { redisClient *slave = ln->value; /* Don't feed slaves that are still waiting for BGSAVE to start */ @@ -1943,7 +2429,7 @@ again: * would not be called at all, but after the execution of the first commands * in the input buffer the client may be blocked, and the "goto again" * will try to reiterate. The following line will make it return asap. */ - if (c->flags & REDIS_BLOCKED) return; + if (c->flags & REDIS_BLOCKED || c->flags & REDIS_IO_WAIT) return; if (c->bulklen == -1) { /* Read the first line of the query */ char *p = strchr(c->querybuf,'\n'); @@ -1952,7 +2438,7 @@ again: if (p) { sds query, *argv; int argc, j; - + query = c->querybuf; c->querybuf = sdsempty(); querylen = 1+(p-(query)); @@ -1992,7 +2478,7 @@ again: } return; } else if (sdslen(c->querybuf) >= REDIS_REQUEST_MAX_SIZE) { - redisLog(REDIS_DEBUG, "Client protocol error"); + redisLog(REDIS_VERBOSE, "Client protocol error"); freeClient(c); return; } @@ -2029,12 +2515,12 @@ static void readQueryFromClient(aeEventLoop *el, int fd, void *privdata, int mas if (errno == EAGAIN) { nread = 0; } else { - redisLog(REDIS_DEBUG, "Reading from client: %s",strerror(errno)); + redisLog(REDIS_VERBOSE, "Reading from client: %s",strerror(errno)); freeClient(c); return; } } else if (nread == 0) { - redisLog(REDIS_DEBUG, "Client closed connection"); + redisLog(REDIS_VERBOSE, "Client closed connection"); freeClient(c); return; } @@ -2056,7 +2542,11 @@ static int selectDb(redisClient *c, int id) { static void *dupClientReplyValue(void *o) { incrRefCount((robj*)o); - return 0; + return o; +} + +static int listMatchObjects(void *a, void *b) { + return compareStringObjects(a,b) == 0; } static redisClient *createClient(int fd) { @@ -2080,9 +2570,16 @@ static redisClient *createClient(int fd) { c->authenticated = 0; c->replstate = REDIS_REPL_NONE; c->reply = listCreate(); - c->blockingkey = NULL; listSetFreeMethod(c->reply,decrRefCount); listSetDupMethod(c->reply,dupClientReplyValue); + c->blockingkeys = NULL; + c->blockingkeysnum = 0; + c->io_keys = listCreate(); + listSetFreeMethod(c->io_keys,decrRefCount); + c->pubsub_channels = dictCreate(&setDictType,NULL); + c->pubsub_patterns = listCreate(); + listSetFreeMethod(c->pubsub_patterns,decrRefCount); + listSetMatchMethod(c->pubsub_patterns,listMatchObjects); if (aeCreateFileEvent(server.el, c->fd, AE_READABLE, readQueryFromClient, c) == AE_ERR) { freeClient(c); @@ -2099,6 +2596,11 @@ static void addReply(redisClient *c, robj *obj) { c->replstate == REDIS_REPL_ONLINE) && aeCreateFileEvent(server.el, c->fd, AE_WRITABLE, sendReplyToClient, c) == AE_ERR) return; + + if (server.vm_enabled && obj->storage != REDIS_VM_MEMORY) { + obj = dupStringObject(obj); + obj->refcount = 0; /* getDecodedObject() will increment the refcount */ + } listAddNodeTail(c->reply,getDecodedObject(obj)); } @@ -2116,6 +2618,51 @@ static void addReplyDouble(redisClient *c, double d) { (unsigned long) strlen(buf),buf)); } +static void addReplyLong(redisClient *c, long l) { + char buf[128]; + size_t len; + + if (l == 0) { + addReply(c,shared.czero); + return; + } else if (l == 1) { + addReply(c,shared.cone); + return; + } + len = snprintf(buf,sizeof(buf),":%ld\r\n",l); + addReplySds(c,sdsnewlen(buf,len)); +} + +static void addReplyLongLong(redisClient *c, long long ll) { + char buf[128]; + size_t len; + + if (ll == 0) { + addReply(c,shared.czero); + return; + } else if (ll == 1) { + addReply(c,shared.cone); + return; + } + len = snprintf(buf,sizeof(buf),":%lld\r\n",ll); + addReplySds(c,sdsnewlen(buf,len)); +} + +static void addReplyUlong(redisClient *c, unsigned long ul) { + char buf[128]; + size_t len; + + if (ul == 0) { + addReply(c,shared.czero); + return; + } else if (ul == 1) { + addReply(c,shared.cone); + return; + } + len = snprintf(buf,sizeof(buf),":%lu\r\n",ul); + addReplySds(c,sdsnewlen(buf,len)); +} + static void addReplyBulkLen(redisClient *c, robj *obj) { size_t len; @@ -2137,6 +2684,23 @@ static void addReplyBulkLen(redisClient *c, robj *obj) { addReplySds(c,sdscatprintf(sdsempty(),"$%lu\r\n",(unsigned long)len)); } +static void addReplyBulk(redisClient *c, robj *obj) { + addReplyBulkLen(c,obj); + addReply(c,obj); + addReply(c,shared.crlf); +} + +/* In the CONFIG command we need to add vanilla C string as bulk replies */ +static void addReplyBulkCString(redisClient *c, char *s) { + if (s == NULL) { + addReply(c,shared.nullbulk); + } else { + robj *o = createStringObject(s,strlen(s)); + addReplyBulk(c,o); + decrRefCount(o); + } +} + static void acceptHandler(aeEventLoop *el, int fd, void *privdata, int mask) { int cport, cfd; char cip[128]; @@ -2147,10 +2711,10 @@ static void acceptHandler(aeEventLoop *el, int fd, void *privdata, int mask) { cfd = anetAccept(server.neterr, fd, cip, &cport); if (cfd == AE_ERR) { - redisLog(REDIS_DEBUG,"Accepting client connection: %s", server.neterr); + redisLog(REDIS_VERBOSE,"Accepting client connection: %s", server.neterr); return; } - redisLog(REDIS_DEBUG,"Accepted %s:%d", cip, cport); + redisLog(REDIS_VERBOSE,"Accepted %s:%d", cip, cport); if ((c = createClient(cfd)) == NULL) { redisLog(REDIS_WARNING,"Error allocating resoures for the client"); close(cfd); /* May be already closed, just ingore errors */ @@ -2178,17 +2742,32 @@ static void acceptHandler(aeEventLoop *el, int fd, void *privdata, int mask) { static robj *createObject(int type, void *ptr) { robj *o; + if (server.vm_enabled) pthread_mutex_lock(&server.obj_freelist_mutex); if (listLength(server.objfreelist)) { listNode *head = listFirst(server.objfreelist); o = listNodeValue(head); listDelNode(server.objfreelist,head); + if (server.vm_enabled) pthread_mutex_unlock(&server.obj_freelist_mutex); } else { - o = zmalloc(sizeof(*o)); + if (server.vm_enabled) { + pthread_mutex_unlock(&server.obj_freelist_mutex); + o = zmalloc(sizeof(*o)); + } else { + o = zmalloc(sizeof(*o)-sizeof(struct redisObjectVM)); + } } o->type = type; o->encoding = REDIS_ENCODING_RAW; o->ptr = ptr; o->refcount = 1; + if (server.vm_enabled) { + /* Note that this code may run in the context of an I/O thread + * and accessing to server.unixtime in theory is an error + * (no locks). But in practice this is safe, and even if we read + * garbage Redis will not fail, as it's just a statistical info */ + o->vm.atime = server.unixtime; + o->storage = REDIS_VM_MEMORY; + } return o; } @@ -2196,6 +2775,28 @@ static robj *createStringObject(char *ptr, size_t len) { return createObject(REDIS_STRING,sdsnewlen(ptr,len)); } +static robj *createStringObjectFromLongLong(long long value) { + robj *o; + if (value >= 0 && value < REDIS_SHARED_INTEGERS) { + incrRefCount(shared.integers[value]); + o = shared.integers[value]; + } else { + o = createObject(REDIS_STRING, NULL); + if (value >= LONG_MIN && value <= LONG_MAX) { + o->encoding = REDIS_ENCODING_INT; + o->ptr = (void*)((long)value); + } else { + o->ptr = sdscatprintf(sdsempty(),"%lld",value); + } + } + return o; +} + +static robj *dupStringObject(robj *o) { + assert(o->encoding == REDIS_ENCODING_RAW); + return createStringObject(o->ptr,sdslen(o->ptr)); +} + static robj *createListObject(void) { list *l = listCreate(); @@ -2208,6 +2809,16 @@ static robj *createSetObject(void) { return createObject(REDIS_SET,d); } +static robj *createHashObject(void) { + /* All the Hashes start as zipmaps. Will be automatically converted + * into hash tables if there are enough elements or big elements + * inside. */ + unsigned char *zm = zipmapNew(); + robj *o = createObject(REDIS_HASH,zm); + o->encoding = REDIS_ENCODING_ZIPMAP; + return o; +} + static robj *createZsetObject(void) { zset *zs = zmalloc(sizeof(*zs)); @@ -2239,42 +2850,97 @@ static void freeZsetObject(robj *o) { } static void freeHashObject(robj *o) { - dictRelease((dict*) o->ptr); + switch (o->encoding) { + case REDIS_ENCODING_HT: + dictRelease((dict*) o->ptr); + break; + case REDIS_ENCODING_ZIPMAP: + zfree(o->ptr); + break; + default: + redisPanic("Unknown hash encoding type"); + break; + } } static void incrRefCount(robj *o) { o->refcount++; -#ifdef DEBUG_REFCOUNT - if (o->type == REDIS_STRING) - printf("Increment '%s'(%p), now is: %d\n",o->ptr,o,o->refcount); -#endif } static void decrRefCount(void *obj) { robj *o = obj; -#ifdef DEBUG_REFCOUNT - if (o->type == REDIS_STRING) - printf("Decrement '%s'(%p), now is: %d\n",o->ptr,o,o->refcount-1); -#endif + if (o->refcount <= 0) redisPanic("decrRefCount against refcount <= 0"); + /* Object is a key of a swapped out value, or in the process of being + * loaded. */ + if (server.vm_enabled && + (o->storage == REDIS_VM_SWAPPED || o->storage == REDIS_VM_LOADING)) + { + if (o->storage == REDIS_VM_LOADING) vmCancelThreadedIOJob(obj); + redisAssert(o->type == REDIS_STRING); + freeStringObject(o); + vmMarkPagesFree(o->vm.page,o->vm.usedpages); + pthread_mutex_lock(&server.obj_freelist_mutex); + if (listLength(server.objfreelist) > REDIS_OBJFREELIST_MAX || + !listAddNodeHead(server.objfreelist,o)) + zfree(o); + pthread_mutex_unlock(&server.obj_freelist_mutex); + server.vm_stats_swapped_objects--; + return; + } + /* Object is in memory, or in the process of being swapped out. */ if (--(o->refcount) == 0) { + if (server.vm_enabled && o->storage == REDIS_VM_SWAPPING) + vmCancelThreadedIOJob(obj); switch(o->type) { case REDIS_STRING: freeStringObject(o); break; case REDIS_LIST: freeListObject(o); break; case REDIS_SET: freeSetObject(o); break; case REDIS_ZSET: freeZsetObject(o); break; case REDIS_HASH: freeHashObject(o); break; - default: redisAssert(0 != 0); break; + default: redisPanic("Unknown object type"); break; } + if (server.vm_enabled) pthread_mutex_lock(&server.obj_freelist_mutex); if (listLength(server.objfreelist) > REDIS_OBJFREELIST_MAX || !listAddNodeHead(server.objfreelist,o)) zfree(o); + if (server.vm_enabled) pthread_mutex_unlock(&server.obj_freelist_mutex); } } static robj *lookupKey(redisDb *db, robj *key) { dictEntry *de = dictFind(db->dict,key); - return de ? dictGetEntryVal(de) : NULL; + if (de) { + robj *key = dictGetEntryKey(de); + robj *val = dictGetEntryVal(de); + + if (server.vm_enabled) { + if (key->storage == REDIS_VM_MEMORY || + key->storage == REDIS_VM_SWAPPING) + { + /* If we were swapping the object out, stop it, this key + * was requested. */ + if (key->storage == REDIS_VM_SWAPPING) + vmCancelThreadedIOJob(key); + /* Update the access time of the key for the aging algorithm. */ + key->vm.atime = server.unixtime; + } else { + int notify = (key->storage == REDIS_VM_LOADING); + + /* Our value was swapped on disk. Bring it at home. */ + redisAssert(val == NULL); + val = vmLoadObject(key); + dictGetEntryVal(de) = val; + + /* Clients blocked by the VM subsystem may be waiting for + * this key... */ + if (notify) handleClientsBlockedOnSwappedKey(db,key); + } + } + return val; + } else { + return NULL; + } } static robj *lookupKeyRead(redisDb *db, robj *key) { @@ -2287,6 +2953,26 @@ static robj *lookupKeyWrite(redisDb *db, robj *key) { return lookupKey(db,key); } +static robj *lookupKeyReadOrReply(redisClient *c, robj *key, robj *reply) { + robj *o = lookupKeyRead(c->db, key); + if (!o) addReply(c,reply); + return o; +} + +static robj *lookupKeyWriteOrReply(redisClient *c, robj *key, robj *reply) { + robj *o = lookupKeyWrite(c->db, key); + if (!o) addReply(c,reply); + return o; +} + +static int checkType(redisClient *c, robj *o, int type) { + if (o->type != type) { + addReply(c,shared.wrongtypeerr); + return 1; + } + return 0; +} + static int deleteKey(redisDb *db, robj *key) { int retval; @@ -2302,51 +2988,6 @@ static int deleteKey(redisDb *db, robj *key) { return retval == DICT_OK; } -/* Try to share an object against the shared objects pool */ -static robj *tryObjectSharing(robj *o) { - struct dictEntry *de; - unsigned long c; - - if (o == NULL || server.shareobjects == 0) return o; - - redisAssert(o->type == REDIS_STRING); - de = dictFind(server.sharingpool,o); - if (de) { - robj *shared = dictGetEntryKey(de); - - c = ((unsigned long) dictGetEntryVal(de))+1; - dictGetEntryVal(de) = (void*) c; - incrRefCount(shared); - decrRefCount(o); - return shared; - } else { - /* Here we are using a stream algorihtm: Every time an object is - * shared we increment its count, everytime there is a miss we - * recrement the counter of a random object. If this object reaches - * zero we remove the object and put the current object instead. */ - if (dictSize(server.sharingpool) >= - server.sharingpoolsize) { - de = dictGetRandomKey(server.sharingpool); - redisAssert(de != NULL); - c = ((unsigned long) dictGetEntryVal(de))-1; - dictGetEntryVal(de) = (void*) c; - if (c == 0) { - dictDelete(server.sharingpool,de->key); - } - } else { - c = 0; /* If the pool is empty we want to add this object */ - } - if (c == 0) { - int retval; - - retval = dictAdd(server.sharingpool,o,(void*)1); - redisAssert(retval == DICT_OK); - incrRefCount(o); - } - return o; - } -} - /* Check if the nul-terminated string 's' can be represented by a long * (that is, is a number that fits into long without any other space or * character before or after the digits). @@ -2357,7 +2998,7 @@ static int isStringRepresentableAsLong(sds s, long *longval) { char buf[32], *endptr; long value; int slen; - + value = strtol(s, &endptr, 10); if (endptr[0] != '\0') return REDIS_ERR; slen = snprintf(buf,32,"%ld",value); @@ -2370,36 +3011,42 @@ static int isStringRepresentableAsLong(sds s, long *longval) { } /* Try to encode a string object in order to save space */ -static int tryObjectEncoding(robj *o) { +static robj *tryObjectEncoding(robj *o) { long value; sds s = o->ptr; if (o->encoding != REDIS_ENCODING_RAW) - return REDIS_ERR; /* Already encoded */ + return o; /* Already encoded */ - /* It's not save to encode shared objects: shared objects can be shared + /* It's not safe to encode shared objects: shared objects can be shared * everywhere in the "object space" of Redis. Encoded objects can only * appear as "values" (and not, for instance, as keys) */ - if (o->refcount > 1) return REDIS_ERR; + if (o->refcount > 1) return o; /* Currently we try to encode only strings */ redisAssert(o->type == REDIS_STRING); /* Check if we can represent this string as a long integer */ - if (isStringRepresentableAsLong(s,&value) == REDIS_ERR) return REDIS_ERR; + if (isStringRepresentableAsLong(s,&value) == REDIS_ERR) return o; /* Ok, this object can be encoded */ - o->encoding = REDIS_ENCODING_INT; - sdsfree(o->ptr); - o->ptr = (void*) value; - return REDIS_OK; + if (value >= 0 && value < REDIS_SHARED_INTEGERS) { + decrRefCount(o); + incrRefCount(shared.integers[value]); + return shared.integers[value]; + } else { + o->encoding = REDIS_ENCODING_INT; + sdsfree(o->ptr); + o->ptr = (void*) value; + return o; + } } /* Get a decoded version of an encoded object (returned as a new object). * If the object is already raw-encoded just increment the ref count. */ static robj *getDecodedObject(robj *o) { robj *dec; - + if (o->encoding == REDIS_ENCODING_RAW) { incrRefCount(o); return o; @@ -2457,25 +3104,116 @@ static size_t stringObjectLen(robj *o) { } } -/*============================ DB saving/loading ============================ */ +static int getDoubleFromObject(robj *o, double *target) { + double value; + char *eptr; -static int rdbSaveType(FILE *fp, unsigned char type) { - if (fwrite(&type,1,1,fp) == 0) return -1; - return 0; -} + if (o == NULL) { + value = 0; + } else { + redisAssert(o->type == REDIS_STRING); + if (o->encoding == REDIS_ENCODING_RAW) { + value = strtod(o->ptr, &eptr); + if (eptr[0] != '\0') return REDIS_ERR; + } else if (o->encoding == REDIS_ENCODING_INT) { + value = (long)o->ptr; + } else { + redisAssert(1 != 1); + } + } -static int rdbSaveTime(FILE *fp, time_t t) { - int32_t t32 = (int32_t) t; - if (fwrite(&t32,4,1,fp) == 0) return -1; - return 0; + *target = value; + return REDIS_OK; } -/* check rdbLoadLen() comments for more info */ -static int rdbSaveLen(FILE *fp, uint32_t len) { - unsigned char buf[2]; +static int getDoubleFromObjectOrReply(redisClient *c, robj *o, double *target, const char *msg) { + double value; + if (getDoubleFromObject(o, &value) != REDIS_OK) { + if (msg != NULL) { + addReplySds(c, sdscatprintf(sdsempty(), "-ERR %s\r\n", msg)); + } else { + addReplySds(c, sdsnew("-ERR value is not a double\r\n")); + } + return REDIS_ERR; + } - if (len < (1<<6)) { - /* Save a 6 bit len */ + *target = value; + return REDIS_OK; +} + +static int getLongLongFromObject(robj *o, long long *target) { + long long value; + char *eptr; + + if (o == NULL) { + value = 0; + } else { + redisAssert(o->type == REDIS_STRING); + if (o->encoding == REDIS_ENCODING_RAW) { + value = strtoll(o->ptr, &eptr, 10); + if (eptr[0] != '\0') return REDIS_ERR; + } else if (o->encoding == REDIS_ENCODING_INT) { + value = (long)o->ptr; + } else { + redisAssert(1 != 1); + } + } + + *target = value; + return REDIS_OK; +} + +static int getLongLongFromObjectOrReply(redisClient *c, robj *o, long long *target, const char *msg) { + long long value; + if (getLongLongFromObject(o, &value) != REDIS_OK) { + if (msg != NULL) { + addReplySds(c, sdscatprintf(sdsempty(), "-ERR %s\r\n", msg)); + } else { + addReplySds(c, sdsnew("-ERR value is not an integer\r\n")); + } + return REDIS_ERR; + } + + *target = value; + return REDIS_OK; +} + +static int getLongFromObjectOrReply(redisClient *c, robj *o, long *target, const char *msg) { + long long value; + + if (getLongLongFromObjectOrReply(c, o, &value, msg) != REDIS_OK) return REDIS_ERR; + if (value < LONG_MIN || value > LONG_MAX) { + if (msg != NULL) { + addReplySds(c, sdscatprintf(sdsempty(), "-ERR %s\r\n", msg)); + } else { + addReplySds(c, sdsnew("-ERR value is out of range\r\n")); + } + return REDIS_ERR; + } + + *target = value; + return REDIS_OK; +} + +/*============================ RDB saving/loading =========================== */ + +static int rdbSaveType(FILE *fp, unsigned char type) { + if (fwrite(&type,1,1,fp) == 0) return -1; + return 0; +} + +static int rdbSaveTime(FILE *fp, time_t t) { + int32_t t32 = (int32_t) t; + if (fwrite(&t32,4,1,fp) == 0) return -1; + return 0; +} + +/* check rdbLoadLen() comments for more info */ +static int rdbSaveLen(FILE *fp, uint32_t len) { + unsigned char buf[2]; + + if (len < (1<<6)) { + /* Save a 6 bit len */ buf[0] = (len&0xFF)|(REDIS_RDB_6BITLEN<<6); if (fwrite(buf,1,1,fp) == 0) return -1; } else if (len < (1<<14)) { @@ -2496,7 +3234,7 @@ static int rdbSaveLen(FILE *fp, uint32_t len) { /* String objects in the form "2391" "-100" without any space and with a * range of values that can fit in an 8, 16 or 32 bit signed value can be * encoded as integers to save space */ -static int rdbTryIntegerEncoding(sds s, unsigned char *enc) { +static int rdbTryIntegerEncoding(char *s, size_t len, unsigned char *enc) { long long value; char *endptr, buf[32]; @@ -2507,7 +3245,7 @@ static int rdbTryIntegerEncoding(sds s, unsigned char *enc) { /* If the number converted back into a string is not identical * then it's not possible to encode the string as integer */ - if (strlen(buf) != sdslen(s) || memcmp(buf,s,sdslen(s))) return 0; + if (strlen(buf) != len || memcmp(buf,s,len)) return 0; /* Finally check if it fits in our ranges */ if (value >= -(1<<7) && value <= (1<<7)-1) { @@ -2531,16 +3269,16 @@ static int rdbTryIntegerEncoding(sds s, unsigned char *enc) { } } -static int rdbSaveLzfStringObject(FILE *fp, robj *obj) { - unsigned int comprlen, outlen; +static int rdbSaveLzfStringObject(FILE *fp, unsigned char *s, size_t len) { + size_t comprlen, outlen; unsigned char byte; void *out; /* We require at least four bytes compression for this to be worth it */ - outlen = sdslen(obj->ptr)-4; - if (outlen <= 0) return 0; + if (len <= 4) return 0; + outlen = len-4; if ((out = zmalloc(outlen+1)) == NULL) return 0; - comprlen = lzf_compress(obj->ptr, sdslen(obj->ptr), out, outlen); + comprlen = lzf_compress(s, len, out, outlen); if (comprlen == 0) { zfree(out); return 0; @@ -2549,7 +3287,7 @@ static int rdbSaveLzfStringObject(FILE *fp, robj *obj) { byte = (REDIS_RDB_ENCVAL<<6)|REDIS_RDB_ENC_LZF; if (fwrite(&byte,1,1,fp) == 0) goto writeerr; if (rdbSaveLen(fp,comprlen) == -1) goto writeerr; - if (rdbSaveLen(fp,sdslen(obj->ptr)) == -1) goto writeerr; + if (rdbSaveLen(fp,len) == -1) goto writeerr; if (fwrite(out,comprlen,1,fp) == 0) goto writeerr; zfree(out); return comprlen; @@ -2561,16 +3299,13 @@ writeerr: /* Save a string objet as [len][data] on disk. If the object is a string * representation of an integer value we try to safe it in a special form */ -static int rdbSaveStringObjectRaw(FILE *fp, robj *obj) { - size_t len; +static int rdbSaveRawString(FILE *fp, unsigned char *s, size_t len) { int enclen; - len = sdslen(obj->ptr); - /* Try integer encoding */ if (len <= 11) { unsigned char buf[5]; - if ((enclen = rdbTryIntegerEncoding(obj->ptr,buf)) > 0) { + if ((enclen = rdbTryIntegerEncoding((char*)s,len,buf)) > 0) { if (fwrite(buf,enclen,1,fp) == 0) return -1; return 0; } @@ -2581,7 +3316,7 @@ static int rdbSaveStringObjectRaw(FILE *fp, robj *obj) { if (server.rdbcompression && len > 20) { int retval; - retval = rdbSaveLzfStringObject(fp,obj); + retval = rdbSaveLzfStringObject(fp,s,len); if (retval == -1) return -1; if (retval > 0) return 0; /* retval == 0 means data can't be compressed, save the old way */ @@ -2589,7 +3324,7 @@ static int rdbSaveStringObjectRaw(FILE *fp, robj *obj) { /* Store verbatim */ if (rdbSaveLen(fp,len) == -1) return -1; - if (len && fwrite(obj->ptr,len,1,fp) == 0) return -1; + if (len && fwrite(s,len,1,fp) == 0) return -1; return 0; } @@ -2597,9 +3332,18 @@ static int rdbSaveStringObjectRaw(FILE *fp, robj *obj) { static int rdbSaveStringObject(FILE *fp, robj *obj) { int retval; - obj = getDecodedObject(obj); - retval = rdbSaveStringObjectRaw(fp,obj); - decrRefCount(obj); + /* Avoid incr/decr ref count business when possible. + * This plays well with copy-on-write given that we are probably + * in a child process (BGSAVE). Also this makes sure key objects + * of swapped objects are not incRefCount-ed (an assert does not allow + * this in order to avoid bugs) */ + if (obj->encoding != REDIS_ENCODING_RAW) { + obj = getDecodedObject(obj); + retval = rdbSaveRawString(fp,obj->ptr,sdslen(obj->ptr)); + decrRefCount(obj); + } else { + retval = rdbSaveRawString(fp,obj->ptr,sdslen(obj->ptr)); + } return retval; } @@ -2630,6 +3374,103 @@ static int rdbSaveDoubleValue(FILE *fp, double val) { return 0; } +/* Save a Redis object. */ +static int rdbSaveObject(FILE *fp, robj *o) { + if (o->type == REDIS_STRING) { + /* Save a string value */ + if (rdbSaveStringObject(fp,o) == -1) return -1; + } else if (o->type == REDIS_LIST) { + /* Save a list value */ + list *list = o->ptr; + listIter li; + listNode *ln; + + if (rdbSaveLen(fp,listLength(list)) == -1) return -1; + listRewind(list,&li); + while((ln = listNext(&li))) { + robj *eleobj = listNodeValue(ln); + + if (rdbSaveStringObject(fp,eleobj) == -1) return -1; + } + } else if (o->type == REDIS_SET) { + /* Save a set value */ + dict *set = o->ptr; + dictIterator *di = dictGetIterator(set); + dictEntry *de; + + if (rdbSaveLen(fp,dictSize(set)) == -1) return -1; + while((de = dictNext(di)) != NULL) { + robj *eleobj = dictGetEntryKey(de); + + if (rdbSaveStringObject(fp,eleobj) == -1) return -1; + } + dictReleaseIterator(di); + } else if (o->type == REDIS_ZSET) { + /* Save a set value */ + zset *zs = o->ptr; + dictIterator *di = dictGetIterator(zs->dict); + dictEntry *de; + + if (rdbSaveLen(fp,dictSize(zs->dict)) == -1) return -1; + while((de = dictNext(di)) != NULL) { + robj *eleobj = dictGetEntryKey(de); + double *score = dictGetEntryVal(de); + + if (rdbSaveStringObject(fp,eleobj) == -1) return -1; + if (rdbSaveDoubleValue(fp,*score) == -1) return -1; + } + dictReleaseIterator(di); + } else if (o->type == REDIS_HASH) { + /* Save a hash value */ + if (o->encoding == REDIS_ENCODING_ZIPMAP) { + unsigned char *p = zipmapRewind(o->ptr); + unsigned int count = zipmapLen(o->ptr); + unsigned char *key, *val; + unsigned int klen, vlen; + + if (rdbSaveLen(fp,count) == -1) return -1; + while((p = zipmapNext(p,&key,&klen,&val,&vlen)) != NULL) { + if (rdbSaveRawString(fp,key,klen) == -1) return -1; + if (rdbSaveRawString(fp,val,vlen) == -1) return -1; + } + } else { + dictIterator *di = dictGetIterator(o->ptr); + dictEntry *de; + + if (rdbSaveLen(fp,dictSize((dict*)o->ptr)) == -1) return -1; + while((de = dictNext(di)) != NULL) { + robj *key = dictGetEntryKey(de); + robj *val = dictGetEntryVal(de); + + if (rdbSaveStringObject(fp,key) == -1) return -1; + if (rdbSaveStringObject(fp,val) == -1) return -1; + } + dictReleaseIterator(di); + } + } else { + redisPanic("Unknown object type"); + } + return 0; +} + +/* Return the length the object will have on disk if saved with + * the rdbSaveObject() function. Currently we use a trick to get + * this length with very little changes to the code. In the future + * we could switch to a faster solution. */ +static off_t rdbSavedObjectLen(robj *o, FILE *fp) { + if (fp == NULL) fp = server.devnull; + rewind(fp); + assert(rdbSaveObject(fp,o) != 1); + return ftello(fp); +} + +/* Return the number of pages required to save this object in the swap file */ +static off_t rdbSavedObjectPages(robj *o, FILE *fp) { + off_t bytes = rdbSavedObjectLen(o,fp); + + return (bytes+(server.vm_page_size-1))/server.vm_page_size; +} + /* Save the DB on disk. Return REDIS_ERR on error, REDIS_OK on success */ static int rdbSave(char *filename) { dictIterator *di = NULL; @@ -2639,6 +3480,12 @@ static int rdbSave(char *filename) { int j; time_t now = time(NULL); + /* Wait for I/O therads to terminate, just in case this is a + * foreground-saving, to avoid seeking the swap file descriptor at the + * same time. */ + if (server.vm_enabled) + waitEmptyIOJobsQueue(); + snprintf(tmpfile,256,"temp-%d.rdb", (int) getpid()); fp = fopen(tmpfile,"w"); if (!fp) { @@ -2673,54 +3520,25 @@ static int rdbSave(char *filename) { if (rdbSaveType(fp,REDIS_EXPIRETIME) == -1) goto werr; if (rdbSaveTime(fp,expiretime) == -1) goto werr; } - /* Save the key and associated value */ - if (rdbSaveType(fp,o->type) == -1) goto werr; - if (rdbSaveStringObject(fp,key) == -1) goto werr; - if (o->type == REDIS_STRING) { - /* Save a string value */ - if (rdbSaveStringObject(fp,o) == -1) goto werr; - } else if (o->type == REDIS_LIST) { - /* Save a list value */ - list *list = o->ptr; - listNode *ln; - - listRewind(list); - if (rdbSaveLen(fp,listLength(list)) == -1) goto werr; - while((ln = listYield(list))) { - robj *eleobj = listNodeValue(ln); - - if (rdbSaveStringObject(fp,eleobj) == -1) goto werr; - } - } else if (o->type == REDIS_SET) { - /* Save a set value */ - dict *set = o->ptr; - dictIterator *di = dictGetIterator(set); - dictEntry *de; - - if (rdbSaveLen(fp,dictSize(set)) == -1) goto werr; - while((de = dictNext(di)) != NULL) { - robj *eleobj = dictGetEntryKey(de); - - if (rdbSaveStringObject(fp,eleobj) == -1) goto werr; - } - dictReleaseIterator(di); - } else if (o->type == REDIS_ZSET) { - /* Save a set value */ - zset *zs = o->ptr; - dictIterator *di = dictGetIterator(zs->dict); - dictEntry *de; - - if (rdbSaveLen(fp,dictSize(zs->dict)) == -1) goto werr; - while((de = dictNext(di)) != NULL) { - robj *eleobj = dictGetEntryKey(de); - double *score = dictGetEntryVal(de); - - if (rdbSaveStringObject(fp,eleobj) == -1) goto werr; - if (rdbSaveDoubleValue(fp,*score) == -1) goto werr; - } - dictReleaseIterator(di); + /* Save the key and associated value. This requires special + * handling if the value is swapped out. */ + if (!server.vm_enabled || key->storage == REDIS_VM_MEMORY || + key->storage == REDIS_VM_SWAPPING) { + /* Save type, key, value */ + if (rdbSaveType(fp,o->type) == -1) goto werr; + if (rdbSaveStringObject(fp,key) == -1) goto werr; + if (rdbSaveObject(fp,o) == -1) goto werr; } else { - redisAssert(0 != 0); + /* REDIS_VM_SWAPPED or REDIS_VM_LOADING */ + robj *po; + /* Get a preview of the object in memory */ + po = vmPreviewObject(key); + /* Save type, key, value */ + if (rdbSaveType(fp,key->vtype) == -1) goto werr; + if (rdbSaveStringObject(fp,key) == -1) goto werr; + if (rdbSaveObject(fp,po) == -1) goto werr; + /* Remove the loaded object from memory */ + decrRefCount(po); } } dictReleaseIterator(di); @@ -2732,7 +3550,7 @@ static int rdbSave(char *filename) { fflush(fp); fsync(fileno(fp)); fclose(fp); - + /* Use RENAME to make sure the DB file is changed atomically only * if the generate DB file is ok. */ if (rename(tmpfile,filename) == -1) { @@ -2757,13 +3575,15 @@ static int rdbSaveBackground(char *filename) { pid_t childpid; if (server.bgsavechildpid != -1) return REDIS_ERR; + if (server.vm_enabled) waitEmptyIOJobsQueue(); if ((childpid = fork()) == 0) { /* Child */ + if (server.vm_enabled) vmReopenSwapFile(); close(server.fd); if (rdbSave(filename) == REDIS_OK) { - exit(0); + _exit(0); } else { - exit(1); + _exit(1); } } else { /* Parent */ @@ -2774,6 +3594,7 @@ static int rdbSaveBackground(char *filename) { } redisLog(REDIS_NOTICE,"Background saving started by pid %d",childpid); server.bgsavechildpid = childpid; + updateDictResizePolicy(); return REDIS_OK; } return REDIS_OK; /* unreached */ @@ -2803,35 +3624,29 @@ static time_t rdbLoadTime(FILE *fp) { * * isencoded is set to 1 if the readed length is not actually a length but * an "encoding type", check the above comments for more info */ -static uint32_t rdbLoadLen(FILE *fp, int rdbver, int *isencoded) { +static uint32_t rdbLoadLen(FILE *fp, int *isencoded) { unsigned char buf[2]; uint32_t len; + int type; if (isencoded) *isencoded = 0; - if (rdbver == 0) { + if (fread(buf,1,1,fp) == 0) return REDIS_RDB_LENERR; + type = (buf[0]&0xC0)>>6; + if (type == REDIS_RDB_6BITLEN) { + /* Read a 6 bit len */ + return buf[0]&0x3F; + } else if (type == REDIS_RDB_ENCVAL) { + /* Read a 6 bit len encoding type */ + if (isencoded) *isencoded = 1; + return buf[0]&0x3F; + } else if (type == REDIS_RDB_14BITLEN) { + /* Read a 14 bit len */ + if (fread(buf+1,1,1,fp) == 0) return REDIS_RDB_LENERR; + return ((buf[0]&0x3F)<<8)|buf[1]; + } else { + /* Read a 32 bit len */ if (fread(&len,4,1,fp) == 0) return REDIS_RDB_LENERR; return ntohl(len); - } else { - int type; - - if (fread(buf,1,1,fp) == 0) return REDIS_RDB_LENERR; - type = (buf[0]&0xC0)>>6; - if (type == REDIS_RDB_6BITLEN) { - /* Read a 6 bit len */ - return buf[0]&0x3F; - } else if (type == REDIS_RDB_ENCVAL) { - /* Read a 6 bit len encoding type */ - if (isencoded) *isencoded = 1; - return buf[0]&0x3F; - } else if (type == REDIS_RDB_14BITLEN) { - /* Read a 14 bit len */ - if (fread(buf+1,1,1,fp) == 0) return REDIS_RDB_LENERR; - return ((buf[0]&0x3F)<<8)|buf[1]; - } else { - /* Read a 32 bit len */ - if (fread(&len,4,1,fp) == 0) return REDIS_RDB_LENERR; - return ntohl(len); - } } } @@ -2854,18 +3669,18 @@ static robj *rdbLoadIntegerObject(FILE *fp, int enctype) { val = (int32_t)v; } else { val = 0; /* anti-warning */ - redisAssert(0!=0); + redisPanic("Unknown RDB integer encoding type"); } return createObject(REDIS_STRING,sdscatprintf(sdsempty(),"%lld",val)); } -static robj *rdbLoadLzfStringObject(FILE*fp, int rdbver) { +static robj *rdbLoadLzfStringObject(FILE*fp) { unsigned int len, clen; unsigned char *c = NULL; sds val = NULL; - if ((clen = rdbLoadLen(fp,rdbver,NULL)) == REDIS_RDB_LENERR) return NULL; - if ((len = rdbLoadLen(fp,rdbver,NULL)) == REDIS_RDB_LENERR) return NULL; + if ((clen = rdbLoadLen(fp,NULL)) == REDIS_RDB_LENERR) return NULL; + if ((len = rdbLoadLen(fp,NULL)) == REDIS_RDB_LENERR) return NULL; if ((c = zmalloc(clen)) == NULL) goto err; if ((val = sdsnewlen(NULL,len)) == NULL) goto err; if (fread(c,clen,1,fp) == 0) goto err; @@ -2878,22 +3693,22 @@ err: return NULL; } -static robj *rdbLoadStringObject(FILE*fp, int rdbver) { +static robj *rdbLoadStringObject(FILE*fp) { int isencoded; uint32_t len; sds val; - len = rdbLoadLen(fp,rdbver,&isencoded); + len = rdbLoadLen(fp,&isencoded); if (isencoded) { switch(len) { case REDIS_RDB_ENC_INT8: case REDIS_RDB_ENC_INT16: case REDIS_RDB_ENC_INT32: - return tryObjectSharing(rdbLoadIntegerObject(fp,len)); + return rdbLoadIntegerObject(fp,len); case REDIS_RDB_ENC_LZF: - return tryObjectSharing(rdbLoadLzfStringObject(fp,rdbver)); + return rdbLoadLzfStringObject(fp); default: - redisAssert(0!=0); + redisPanic("Unknown RDB encoding type"); } } @@ -2903,7 +3718,7 @@ static robj *rdbLoadStringObject(FILE*fp, int rdbver) { sdsfree(val); return NULL; } - return tryObjectSharing(createObject(REDIS_STRING,val)); + return createObject(REDIS_STRING,val); } /* For information about double serialization check rdbSaveDoubleValue() */ @@ -2924,6 +3739,102 @@ static int rdbLoadDoubleValue(FILE *fp, double *val) { } } +/* Load a Redis object of the specified type from the specified file. + * On success a newly allocated object is returned, otherwise NULL. */ +static robj *rdbLoadObject(int type, FILE *fp) { + robj *o; + + redisLog(REDIS_DEBUG,"LOADING OBJECT %d (at %d)\n",type,ftell(fp)); + if (type == REDIS_STRING) { + /* Read string value */ + if ((o = rdbLoadStringObject(fp)) == NULL) return NULL; + o = tryObjectEncoding(o); + } else if (type == REDIS_LIST || type == REDIS_SET) { + /* Read list/set value */ + uint32_t listlen; + + if ((listlen = rdbLoadLen(fp,NULL)) == REDIS_RDB_LENERR) return NULL; + o = (type == REDIS_LIST) ? createListObject() : createSetObject(); + /* It's faster to expand the dict to the right size asap in order + * to avoid rehashing */ + if (type == REDIS_SET && listlen > DICT_HT_INITIAL_SIZE) + dictExpand(o->ptr,listlen); + /* Load every single element of the list/set */ + while(listlen--) { + robj *ele; + + if ((ele = rdbLoadStringObject(fp)) == NULL) return NULL; + ele = tryObjectEncoding(ele); + if (type == REDIS_LIST) { + listAddNodeTail((list*)o->ptr,ele); + } else { + dictAdd((dict*)o->ptr,ele,NULL); + } + } + } else if (type == REDIS_ZSET) { + /* Read list/set value */ + size_t zsetlen; + zset *zs; + + if ((zsetlen = rdbLoadLen(fp,NULL)) == REDIS_RDB_LENERR) return NULL; + o = createZsetObject(); + zs = o->ptr; + /* Load every single element of the list/set */ + while(zsetlen--) { + robj *ele; + double *score = zmalloc(sizeof(double)); + + if ((ele = rdbLoadStringObject(fp)) == NULL) return NULL; + ele = tryObjectEncoding(ele); + if (rdbLoadDoubleValue(fp,score) == -1) return NULL; + dictAdd(zs->dict,ele,score); + zslInsert(zs->zsl,*score,ele); + incrRefCount(ele); /* added to skiplist */ + } + } else if (type == REDIS_HASH) { + size_t hashlen; + + if ((hashlen = rdbLoadLen(fp,NULL)) == REDIS_RDB_LENERR) return NULL; + o = createHashObject(); + /* Too many entries? Use an hash table. */ + if (hashlen > server.hash_max_zipmap_entries) + convertToRealHash(o); + /* Load every key/value, then set it into the zipmap or hash + * table, as needed. */ + while(hashlen--) { + robj *key, *val; + + if ((key = rdbLoadStringObject(fp)) == NULL) return NULL; + if ((val = rdbLoadStringObject(fp)) == NULL) return NULL; + /* If we are using a zipmap and there are too big values + * the object is converted to real hash table encoding. */ + if (o->encoding != REDIS_ENCODING_HT && + (sdslen(key->ptr) > server.hash_max_zipmap_value || + sdslen(val->ptr) > server.hash_max_zipmap_value)) + { + convertToRealHash(o); + } + + if (o->encoding == REDIS_ENCODING_ZIPMAP) { + unsigned char *zm = o->ptr; + + zm = zipmapSet(zm,key->ptr,sdslen(key->ptr), + val->ptr,sdslen(val->ptr),NULL); + o->ptr = zm; + decrRefCount(key); + decrRefCount(val); + } else { + key = tryObjectEncoding(key); + val = tryObjectEncoding(val); + dictAdd((dict*)o->ptr,key,val); + } + } + } else { + redisPanic("Unknown object type"); + } + return o; +} + static int rdbLoad(char *filename) { FILE *fp; robj *keyobj = NULL; @@ -2933,6 +3844,7 @@ static int rdbLoad(char *filename) { redisDb *db = server.db+0; char buf[1024]; time_t expiretime = -1, now = time(NULL); + long long loadedkeys = 0; fp = fopen(filename,"r"); if (!fp) return REDIS_ERR; @@ -2944,7 +3856,7 @@ static int rdbLoad(char *filename) { return REDIS_ERR; } rdbver = atoi(buf+5); - if (rdbver > 1) { + if (rdbver != 1) { fclose(fp); redisLog(REDIS_WARNING,"Can't handle RDB format version %d",rdbver); return REDIS_ERR; @@ -2962,7 +3874,7 @@ static int rdbLoad(char *filename) { if (type == REDIS_EOF) break; /* Handle SELECT DB opcode as a special case */ if (type == REDIS_SELECTDB) { - if ((dbid = rdbLoadLen(fp,rdbver,NULL)) == REDIS_RDB_LENERR) + if ((dbid = rdbLoadLen(fp,NULL)) == REDIS_RDB_LENERR) goto eoferr; if (dbid >= (unsigned)server.dbnum) { redisLog(REDIS_WARNING,"FATAL: Data file was created with a Redis server configured to handle more than %d databases. Exiting\n", server.dbnum); @@ -2973,55 +3885,9 @@ static int rdbLoad(char *filename) { continue; } /* Read key */ - if ((keyobj = rdbLoadStringObject(fp,rdbver)) == NULL) goto eoferr; - - if (type == REDIS_STRING) { - /* Read string value */ - if ((o = rdbLoadStringObject(fp,rdbver)) == NULL) goto eoferr; - tryObjectEncoding(o); - } else if (type == REDIS_LIST || type == REDIS_SET) { - /* Read list/set value */ - uint32_t listlen; - - if ((listlen = rdbLoadLen(fp,rdbver,NULL)) == REDIS_RDB_LENERR) - goto eoferr; - o = (type == REDIS_LIST) ? createListObject() : createSetObject(); - /* Load every single element of the list/set */ - while(listlen--) { - robj *ele; - - if ((ele = rdbLoadStringObject(fp,rdbver)) == NULL) goto eoferr; - tryObjectEncoding(ele); - if (type == REDIS_LIST) { - listAddNodeTail((list*)o->ptr,ele); - } else { - dictAdd((dict*)o->ptr,ele,NULL); - } - } - } else if (type == REDIS_ZSET) { - /* Read list/set value */ - uint32_t zsetlen; - zset *zs; - - if ((zsetlen = rdbLoadLen(fp,rdbver,NULL)) == REDIS_RDB_LENERR) - goto eoferr; - o = createZsetObject(); - zs = o->ptr; - /* Load every single element of the list/set */ - while(zsetlen--) { - robj *ele; - double *score = zmalloc(sizeof(double)); - - if ((ele = rdbLoadStringObject(fp,rdbver)) == NULL) goto eoferr; - tryObjectEncoding(ele); - if (rdbLoadDoubleValue(fp,score) == -1) goto eoferr; - dictAdd(zs->dict,ele,score); - zslInsert(zs->zsl,*score,ele); - incrRefCount(ele); /* added to skiplist */ - } - } else { - redisAssert(0 != 0); - } + if ((keyobj = rdbLoadStringObject(fp)) == NULL) goto eoferr; + /* Read value */ + if ((o = rdbLoadObject(type,fp)) == NULL) goto eoferr; /* Add the new object in the hash table */ retval = dictAdd(d,keyobj,o); if (retval == DICT_ERR) { @@ -3036,6 +3902,13 @@ static int rdbLoad(char *filename) { expiretime = -1; } keyobj = o = NULL; + /* Handle swapping while loading big datasets when VM is on */ + loadedkeys++; + if (server.vm_enabled && (loadedkeys % 5000) == 0) { + while (zmalloc_used_memory() > server.vm_max_memory) { + if (vmSwapOneObjectBlocking() == REDIS_ERR) break; + } + } } fclose(fp); return REDIS_OK; @@ -3064,9 +3937,7 @@ static void pingCommand(redisClient *c) { } static void echoCommand(redisClient *c) { - addReplyBulkLen(c,c->argv[1]); - addReply(c,c->argv[1]); - addReply(c,shared.crlf); + addReplyBulk(c,c->argv[1]); } /*=================================== Strings =============================== */ @@ -3078,6 +3949,12 @@ static void setGenericCommand(redisClient *c, int nx) { retval = dictAdd(c->db->dict,c->argv[1],c->argv[2]); if (retval == DICT_ERR) { if (!nx) { + /* If the key is about a swapped value, we want a new key object + * to overwrite the old. So we delete the old key in the database. + * This will also make sure that swap pages about the old object + * will be marked as free. */ + if (server.vm_enabled && deleteIfSwapped(c->db,c->argv[1])) + incrRefCount(c->argv[1]); dictReplace(c->db->dict,c->argv[1],c->argv[2]); incrRefCount(c->argv[2]); } else { @@ -3102,21 +3979,17 @@ static void setnxCommand(redisClient *c) { } static int getGenericCommand(redisClient *c) { - robj *o = lookupKeyRead(c->db,c->argv[1]); + robj *o; - if (o == NULL) { - addReply(c,shared.nullbulk); + if ((o = lookupKeyReadOrReply(c,c->argv[1],shared.nullbulk)) == NULL) return REDIS_OK; + + if (o->type != REDIS_STRING) { + addReply(c,shared.wrongtypeerr); + return REDIS_ERR; } else { - if (o->type != REDIS_STRING) { - addReply(c,shared.wrongtypeerr); - return REDIS_ERR; - } else { - addReplyBulkLen(c,o); - addReply(c,o); - addReply(c,shared.crlf); - return REDIS_OK; - } + addReplyBulk(c,o); + return REDIS_OK; } } @@ -3138,7 +4011,7 @@ static void getsetCommand(redisClient *c) { static void mgetCommand(redisClient *c) { int j; - + addReplySds(c,sdscatprintf(sdsempty(),"*%d\r\n",c->argc-1)); for (j = 1; j < c->argc; j++) { robj *o = lookupKeyRead(c->db,c->argv[j]); @@ -3148,9 +4021,7 @@ static void mgetCommand(redisClient *c) { if (o->type != REDIS_STRING) { addReply(c,shared.nullbulk); } else { - addReplyBulkLen(c,o); - addReply(c,o); - addReply(c,shared.crlf); + addReplyBulk(c,o); } } } @@ -3180,7 +4051,7 @@ static void msetGenericCommand(redisClient *c, int nx) { for (j = 1; j < c->argc; j += 2) { int retval; - tryObjectEncoding(c->argv[j+1]); + c->argv[j+1] = tryObjectEncoding(c->argv[j+1]); retval = dictAdd(c->db->dict,c->argv[j],c->argv[j+1]); if (retval == DICT_ERR) { dictReplace(c->db->dict,c->argv[j],c->argv[j+1]); @@ -3207,28 +4078,14 @@ static void incrDecrCommand(redisClient *c, long long incr) { long long value; int retval; robj *o; - + o = lookupKeyWrite(c->db,c->argv[1]); - if (o == NULL) { - value = 0; - } else { - if (o->type != REDIS_STRING) { - value = 0; - } else { - char *eptr; - if (o->encoding == REDIS_ENCODING_RAW) - value = strtoll(o->ptr, &eptr, 10); - else if (o->encoding == REDIS_ENCODING_INT) - value = (long)o->ptr; - else - redisAssert(1 != 1); - } - } + if (getLongLongFromObjectOrReply(c, o, &value, NULL) != REDIS_OK) return; value += incr; o = createObject(REDIS_STRING,sdscatprintf(sdsempty(),"%lld",value)); - tryObjectEncoding(o); + o = tryObjectEncoding(o); retval = dictAdd(c->db->dict,c->argv[1],o); if (retval == DICT_ERR) { dictReplace(c->db->dict,c->argv[1],o); @@ -3251,15 +4108,102 @@ static void decrCommand(redisClient *c) { } static void incrbyCommand(redisClient *c) { - long long incr = strtoll(c->argv[2]->ptr, NULL, 10); + long long incr; + + if (getLongLongFromObjectOrReply(c, c->argv[2], &incr, NULL) != REDIS_OK) return; incrDecrCommand(c,incr); } static void decrbyCommand(redisClient *c) { - long long incr = strtoll(c->argv[2]->ptr, NULL, 10); + long long incr; + + if (getLongLongFromObjectOrReply(c, c->argv[2], &incr, NULL) != REDIS_OK) return; incrDecrCommand(c,-incr); } +static void appendCommand(redisClient *c) { + int retval; + size_t totlen; + robj *o; + + o = lookupKeyWrite(c->db,c->argv[1]); + if (o == NULL) { + /* Create the key */ + retval = dictAdd(c->db->dict,c->argv[1],c->argv[2]); + incrRefCount(c->argv[1]); + incrRefCount(c->argv[2]); + totlen = stringObjectLen(c->argv[2]); + } else { + dictEntry *de; + + de = dictFind(c->db->dict,c->argv[1]); + assert(de != NULL); + + o = dictGetEntryVal(de); + if (o->type != REDIS_STRING) { + addReply(c,shared.wrongtypeerr); + return; + } + /* If the object is specially encoded or shared we have to make + * a copy */ + if (o->refcount != 1 || o->encoding != REDIS_ENCODING_RAW) { + robj *decoded = getDecodedObject(o); + + o = createStringObject(decoded->ptr, sdslen(decoded->ptr)); + decrRefCount(decoded); + dictReplace(c->db->dict,c->argv[1],o); + } + /* APPEND! */ + if (c->argv[2]->encoding == REDIS_ENCODING_RAW) { + o->ptr = sdscatlen(o->ptr, + c->argv[2]->ptr, sdslen(c->argv[2]->ptr)); + } else { + o->ptr = sdscatprintf(o->ptr, "%ld", + (unsigned long) c->argv[2]->ptr); + } + totlen = sdslen(o->ptr); + } + server.dirty++; + addReplySds(c,sdscatprintf(sdsempty(),":%lu\r\n",(unsigned long)totlen)); +} + +static void substrCommand(redisClient *c) { + robj *o; + long start = atoi(c->argv[2]->ptr); + long end = atoi(c->argv[3]->ptr); + size_t rangelen, strlen; + sds range; + + if ((o = lookupKeyReadOrReply(c,c->argv[1],shared.nullbulk)) == NULL || + checkType(c,o,REDIS_STRING)) return; + + o = getDecodedObject(o); + strlen = sdslen(o->ptr); + + /* convert negative indexes */ + if (start < 0) start = strlen+start; + if (end < 0) end = strlen+end; + if (start < 0) start = 0; + if (end < 0) end = 0; + + /* indexes sanity checks */ + if (start > end || (size_t)start >= strlen) { + /* Out of range start or start > end result in null reply */ + addReply(c,shared.nullbulk); + decrRefCount(o); + return; + } + if ((size_t)end >= strlen) end = strlen-1; + rangelen = (end-start)+1; + + /* Return the result */ + addReplySds(c,sdscatprintf(sdsempty(),"$%zu\r\n",rangelen)); + range = sdsnewlen((char*)o->ptr+start,rangelen); + addReplySds(c,range); + addReply(c,shared.crlf); + decrRefCount(o); +} + /* ========================= Type agnostic commands ========================= */ static void delCommand(redisClient *c) { @@ -3271,17 +4215,7 @@ static void delCommand(redisClient *c) { deleted++; } } - switch(deleted) { - case 0: - addReply(c,shared.czero); - break; - case 1: - addReply(c,shared.cone); - break; - default: - addReplySds(c,sdscatprintf(sdsempty(),":%d\r\n",deleted)); - break; - } + addReplyLong(c,deleted); } static void existsCommand(redisClient *c) { @@ -3290,7 +4224,7 @@ static void existsCommand(redisClient *c) { static void selectCommand(redisClient *c) { int id = atoi(c->argv[1]->ptr); - + if (selectDb(c,id) == REDIS_ERR) { addReplySds(c,sdsnew("-ERR invalid DB index\r\n")); } else { @@ -3300,7 +4234,7 @@ static void selectCommand(redisClient *c) { static void randomkeyCommand(redisClient *c) { dictEntry *de; - + while(1) { de = dictGetRandomKey(c->db->dict); if (!de || expireIfNeeded(c->db,dictGetEntryKey(de)) == 0) break; @@ -3320,7 +4254,7 @@ static void keysCommand(redisClient *c) { dictEntry *de; sds pattern = c->argv[1]->ptr; int plen = sdslen(pattern); - unsigned long numkeys = 0, keyslen = 0; + unsigned long numkeys = 0; robj *lenobj = createObject(REDIS_STRING,NULL); di = dictGetIterator(c->db->dict); @@ -3333,17 +4267,13 @@ static void keysCommand(redisClient *c) { if ((pattern[0] == '*' && pattern[1] == '\0') || stringmatchlen(pattern,plen,key,sdslen(key),0)) { if (expireIfNeeded(c->db,keyobj) == 0) { - if (numkeys != 0) - addReply(c,shared.space); - addReply(c,keyobj); + addReplyBulk(c,keyobj); numkeys++; - keyslen += sdslen(key); } } } dictReleaseIterator(di); - lenobj->ptr = sdscatprintf(sdsempty(),"$%lu\r\n",keyslen+(numkeys ? (numkeys-1) : 0)); - addReply(c,shared.crlf); + lenobj->ptr = sdscatprintf(sdsempty(),"*%lu\r\n",numkeys); } static void dbsizeCommand(redisClient *c) { @@ -3369,7 +4299,8 @@ static void typeCommand(redisClient *c) { case REDIS_LIST: type = "+list"; break; case REDIS_SET: type = "+set"; break; case REDIS_ZSET: type = "+zset"; break; - default: type = "unknown"; break; + case REDIS_HASH: type = "+hash"; break; + default: type = "+unknown"; break; } } addReplySds(c,sdsnew(type)); @@ -3414,6 +4345,7 @@ static void shutdownCommand(redisClient *c) { if (server.appendonly) { /* Append only file: fsync() the AOF and exit */ fsync(server.appendfd); + if (server.vm_enabled) unlink(server.vm_swap_file); exit(0); } else { /* Snapshotting. Perform a SYNC SAVE and exit */ @@ -3422,14 +4354,17 @@ static void shutdownCommand(redisClient *c) { unlink(server.pidfile); redisLog(REDIS_WARNING,"%zu bytes used at exit",zmalloc_used_memory()); redisLog(REDIS_WARNING,"Server exit now, bye bye..."); + if (server.vm_enabled) unlink(server.vm_swap_file); exit(0); } else { - /* Ooops.. error saving! The best we can do is to continue operating. - * Note that if there was a background saving process, in the next - * cron() Redis will be notified that the background saving aborted, - * handling special stuff like slaves pending for synchronization... */ - redisLog(REDIS_WARNING,"Error trying to save the DB, can't exit"); - addReplySds(c,sdsnew("-ERR can't quit, problems saving the DB\r\n")); + /* Ooops.. error saving! The best we can do is to continue + * operating. Note that if there was a background saving process, + * in the next cron() Redis will be notified that the background + * saving aborted, handling special stuff like slaves pending for + * synchronization... */ + redisLog(REDIS_WARNING,"Error trying to save the DB, can't exit"); + addReplySds(c, + sdsnew("-ERR can't quit, problems saving the DB\r\n")); } } } @@ -3443,11 +4378,9 @@ static void renameGenericCommand(redisClient *c, int nx) { return; } - o = lookupKeyWrite(c->db,c->argv[1]); - if (o == NULL) { - addReply(c,shared.nokeyerr); + if ((o = lookupKeyWriteOrReply(c,c->argv[1],shared.nokeyerr)) == NULL) return; - } + incrRefCount(o); deleteIfVolatile(c->db,c->argv[2]); if (dictAdd(c->db->dict,c->argv[2],o) == DICT_ERR) { @@ -3524,7 +4457,10 @@ static void pushGenericCommand(redisClient *c, int where) { lobj = lookupKeyWrite(c->db,c->argv[1]); if (lobj == NULL) { - if (handleClientsWaitingListPush(c,c->argv[1],c->argv[2])) return; + if (handleClientsWaitingListPush(c,c->argv[1],c->argv[2])) { + addReply(c,shared.cone); + return; + } lobj = createListObject(); list = lobj->ptr; if (where == REDIS_HEAD) { @@ -3540,7 +4476,10 @@ static void pushGenericCommand(redisClient *c, int where) { addReply(c,shared.wrongtypeerr); return; } - if (handleClientsWaitingListPush(c,c->argv[1],c->argv[2])) return; + if (handleClientsWaitingListPush(c,c->argv[1],c->argv[2])) { + addReply(c,shared.cone); + return; + } list = lobj->ptr; if (where == REDIS_HEAD) { listAddNodeHead(list,c->argv[2]); @@ -3550,7 +4489,7 @@ static void pushGenericCommand(redisClient *c, int where) { incrRefCount(c->argv[2]); } server.dirty++; - addReply(c,shared.ok); + addReplySds(c,sdscatprintf(sdsempty(),":%d\r\n",listLength(list))); } static void lpushCommand(redisClient *c) { @@ -3564,107 +4503,79 @@ static void rpushCommand(redisClient *c) { static void llenCommand(redisClient *c) { robj *o; list *l; - - o = lookupKeyRead(c->db,c->argv[1]); - if (o == NULL) { - addReply(c,shared.czero); - return; - } else { - if (o->type != REDIS_LIST) { - addReply(c,shared.wrongtypeerr); - } else { - l = o->ptr; - addReplySds(c,sdscatprintf(sdsempty(),":%d\r\n",listLength(l))); - } - } + + if ((o = lookupKeyReadOrReply(c,c->argv[1],shared.czero)) == NULL || + checkType(c,o,REDIS_LIST)) return; + + l = o->ptr; + addReplyUlong(c,listLength(l)); } static void lindexCommand(redisClient *c) { robj *o; int index = atoi(c->argv[2]->ptr); - - o = lookupKeyRead(c->db,c->argv[1]); - if (o == NULL) { + list *list; + listNode *ln; + + if ((o = lookupKeyReadOrReply(c,c->argv[1],shared.nullbulk)) == NULL || + checkType(c,o,REDIS_LIST)) return; + list = o->ptr; + + ln = listIndex(list, index); + if (ln == NULL) { addReply(c,shared.nullbulk); } else { - if (o->type != REDIS_LIST) { - addReply(c,shared.wrongtypeerr); - } else { - list *list = o->ptr; - listNode *ln; - - ln = listIndex(list, index); - if (ln == NULL) { - addReply(c,shared.nullbulk); - } else { - robj *ele = listNodeValue(ln); - addReplyBulkLen(c,ele); - addReply(c,ele); - addReply(c,shared.crlf); - } - } + robj *ele = listNodeValue(ln); + addReplyBulk(c,ele); } } static void lsetCommand(redisClient *c) { robj *o; int index = atoi(c->argv[2]->ptr); - - o = lookupKeyWrite(c->db,c->argv[1]); - if (o == NULL) { - addReply(c,shared.nokeyerr); + list *list; + listNode *ln; + + if ((o = lookupKeyWriteOrReply(c,c->argv[1],shared.nokeyerr)) == NULL || + checkType(c,o,REDIS_LIST)) return; + list = o->ptr; + + ln = listIndex(list, index); + if (ln == NULL) { + addReply(c,shared.outofrangeerr); } else { - if (o->type != REDIS_LIST) { - addReply(c,shared.wrongtypeerr); - } else { - list *list = o->ptr; - listNode *ln; - - ln = listIndex(list, index); - if (ln == NULL) { - addReply(c,shared.outofrangeerr); - } else { - robj *ele = listNodeValue(ln); + robj *ele = listNodeValue(ln); - decrRefCount(ele); - listNodeValue(ln) = c->argv[3]; - incrRefCount(c->argv[3]); - addReply(c,shared.ok); - server.dirty++; - } - } + decrRefCount(ele); + listNodeValue(ln) = c->argv[3]; + incrRefCount(c->argv[3]); + addReply(c,shared.ok); + server.dirty++; } } static void popGenericCommand(redisClient *c, int where) { robj *o; + list *list; + listNode *ln; - o = lookupKeyWrite(c->db,c->argv[1]); - if (o == NULL) { - addReply(c,shared.nullbulk); - } else { - if (o->type != REDIS_LIST) { - addReply(c,shared.wrongtypeerr); - } else { - list *list = o->ptr; - listNode *ln; + if ((o = lookupKeyWriteOrReply(c,c->argv[1],shared.nullbulk)) == NULL || + checkType(c,o,REDIS_LIST)) return; + list = o->ptr; - if (where == REDIS_HEAD) - ln = listFirst(list); - else - ln = listLast(list); + if (where == REDIS_HEAD) + ln = listFirst(list); + else + ln = listLast(list); - if (ln == NULL) { - addReply(c,shared.nullbulk); - } else { - robj *ele = listNodeValue(ln); - addReplyBulkLen(c,ele); - addReply(c,ele); - addReply(c,shared.crlf); - listDelNode(list,ln); - server.dirty++; - } - } + if (ln == NULL) { + addReply(c,shared.nullbulk); + } else { + robj *ele = listNodeValue(ln); + addReplyBulk(c,ele); + listDelNode(list,ln); + if (listLength(list) == 0) deleteKey(c->db,c->argv[1]); + server.dirty++; } } @@ -3680,46 +4591,39 @@ static void lrangeCommand(redisClient *c) { robj *o; int start = atoi(c->argv[2]->ptr); int end = atoi(c->argv[3]->ptr); + int llen; + int rangelen, j; + list *list; + listNode *ln; + robj *ele; + + if ((o = lookupKeyReadOrReply(c,c->argv[1],shared.emptymultibulk)) == NULL + || checkType(c,o,REDIS_LIST)) return; + list = o->ptr; + llen = listLength(list); + + /* convert negative indexes */ + if (start < 0) start = llen+start; + if (end < 0) end = llen+end; + if (start < 0) start = 0; + if (end < 0) end = 0; + + /* indexes sanity checks */ + if (start > end || start >= llen) { + /* Out of range start or start > end result in empty list */ + addReply(c,shared.emptymultibulk); + return; + } + if (end >= llen) end = llen-1; + rangelen = (end-start)+1; - o = lookupKeyRead(c->db,c->argv[1]); - if (o == NULL) { - addReply(c,shared.nullmultibulk); - } else { - if (o->type != REDIS_LIST) { - addReply(c,shared.wrongtypeerr); - } else { - list *list = o->ptr; - listNode *ln; - int llen = listLength(list); - int rangelen, j; - robj *ele; - - /* convert negative indexes */ - if (start < 0) start = llen+start; - if (end < 0) end = llen+end; - if (start < 0) start = 0; - if (end < 0) end = 0; - - /* indexes sanity checks */ - if (start > end || start >= llen) { - /* Out of range start or start > end result in empty list */ - addReply(c,shared.emptymultibulk); - return; - } - if (end >= llen) end = llen-1; - rangelen = (end-start)+1; - - /* Return the result in form of a multi-bulk reply */ - ln = listIndex(list, start); - addReplySds(c,sdscatprintf(sdsempty(),"*%d\r\n",rangelen)); - for (j = 0; j < rangelen; j++) { - ele = listNodeValue(ln); - addReplyBulkLen(c,ele); - addReply(c,ele); - addReply(c,shared.crlf); - ln = ln->next; - } - } + /* Return the result in form of a multi-bulk reply */ + ln = listIndex(list, start); + addReplySds(c,sdscatprintf(sdsempty(),"*%d\r\n",rangelen)); + for (j = 0; j < rangelen; j++) { + ele = listNodeValue(ln); + addReplyBulk(c,ele); + ln = ln->next; } } @@ -3727,87 +4631,78 @@ static void ltrimCommand(redisClient *c) { robj *o; int start = atoi(c->argv[2]->ptr); int end = atoi(c->argv[3]->ptr); - - o = lookupKeyWrite(c->db,c->argv[1]); - if (o == NULL) { - addReply(c,shared.ok); + int llen; + int j, ltrim, rtrim; + list *list; + listNode *ln; + + if ((o = lookupKeyWriteOrReply(c,c->argv[1],shared.ok)) == NULL || + checkType(c,o,REDIS_LIST)) return; + list = o->ptr; + llen = listLength(list); + + /* convert negative indexes */ + if (start < 0) start = llen+start; + if (end < 0) end = llen+end; + if (start < 0) start = 0; + if (end < 0) end = 0; + + /* indexes sanity checks */ + if (start > end || start >= llen) { + /* Out of range start or start > end result in empty list */ + ltrim = llen; + rtrim = 0; } else { - if (o->type != REDIS_LIST) { - addReply(c,shared.wrongtypeerr); - } else { - list *list = o->ptr; - listNode *ln; - int llen = listLength(list); - int j, ltrim, rtrim; - - /* convert negative indexes */ - if (start < 0) start = llen+start; - if (end < 0) end = llen+end; - if (start < 0) start = 0; - if (end < 0) end = 0; - - /* indexes sanity checks */ - if (start > end || start >= llen) { - /* Out of range start or start > end result in empty list */ - ltrim = llen; - rtrim = 0; - } else { - if (end >= llen) end = llen-1; - ltrim = start; - rtrim = llen-end-1; - } + if (end >= llen) end = llen-1; + ltrim = start; + rtrim = llen-end-1; + } - /* Remove list elements to perform the trim */ - for (j = 0; j < ltrim; j++) { - ln = listFirst(list); - listDelNode(list,ln); - } - for (j = 0; j < rtrim; j++) { - ln = listLast(list); - listDelNode(list,ln); - } - server.dirty++; - addReply(c,shared.ok); - } + /* Remove list elements to perform the trim */ + for (j = 0; j < ltrim; j++) { + ln = listFirst(list); + listDelNode(list,ln); } + for (j = 0; j < rtrim; j++) { + ln = listLast(list); + listDelNode(list,ln); + } + if (listLength(list) == 0) deleteKey(c->db,c->argv[1]); + server.dirty++; + addReply(c,shared.ok); } static void lremCommand(redisClient *c) { robj *o; - - o = lookupKeyWrite(c->db,c->argv[1]); - if (o == NULL) { - addReply(c,shared.czero); - } else { - if (o->type != REDIS_LIST) { - addReply(c,shared.wrongtypeerr); - } else { - list *list = o->ptr; - listNode *ln, *next; - int toremove = atoi(c->argv[2]->ptr); - int removed = 0; - int fromtail = 0; - - if (toremove < 0) { - toremove = -toremove; - fromtail = 1; - } - ln = fromtail ? list->tail : list->head; - while (ln) { - robj *ele = listNodeValue(ln); - - next = fromtail ? ln->prev : ln->next; - if (compareStringObjects(ele,c->argv[3]) == 0) { - listDelNode(list,ln); - server.dirty++; - removed++; - if (toremove && removed == toremove) break; - } - ln = next; - } - addReplySds(c,sdscatprintf(sdsempty(),":%d\r\n",removed)); + list *list; + listNode *ln, *next; + int toremove = atoi(c->argv[2]->ptr); + int removed = 0; + int fromtail = 0; + + if ((o = lookupKeyWriteOrReply(c,c->argv[1],shared.czero)) == NULL || + checkType(c,o,REDIS_LIST)) return; + list = o->ptr; + + if (toremove < 0) { + toremove = -toremove; + fromtail = 1; + } + ln = fromtail ? list->tail : list->head; + while (ln) { + robj *ele = listNodeValue(ln); + + next = fromtail ? ln->prev : ln->next; + if (compareStringObjects(ele,c->argv[3]) == 0) { + listDelNode(list,ln); + server.dirty++; + removed++; + if (toremove && removed == toremove) break; } + ln = next; } + if (listLength(list) == 0) deleteKey(c->db,c->argv[1]); + addReplySds(c,sdscatprintf(sdsempty(),":%d\r\n",removed)); } /* This is the semantic of this command: @@ -3827,53 +4722,50 @@ static void lremCommand(redisClient *c) { */ static void rpoplpushcommand(redisClient *c) { robj *sobj; + list *srclist; + listNode *ln; + + if ((sobj = lookupKeyWriteOrReply(c,c->argv[1],shared.nullbulk)) == NULL || + checkType(c,sobj,REDIS_LIST)) return; + srclist = sobj->ptr; + ln = listLast(srclist); - sobj = lookupKeyWrite(c->db,c->argv[1]); - if (sobj == NULL) { + if (ln == NULL) { addReply(c,shared.nullbulk); } else { - if (sobj->type != REDIS_LIST) { + robj *dobj = lookupKeyWrite(c->db,c->argv[2]); + robj *ele = listNodeValue(ln); + list *dstlist; + + if (dobj && dobj->type != REDIS_LIST) { addReply(c,shared.wrongtypeerr); - } else { - list *srclist = sobj->ptr; - listNode *ln = listLast(srclist); + return; + } - if (ln == NULL) { - addReply(c,shared.nullbulk); - } else { - robj *dobj = lookupKeyWrite(c->db,c->argv[2]); - robj *ele = listNodeValue(ln); - list *dstlist; - - if (dobj == NULL) { - - /* Create the list if the key does not exist */ - dobj = createListObject(); - dictAdd(c->db->dict,c->argv[2],dobj); - incrRefCount(c->argv[2]); - } else if (dobj->type != REDIS_LIST) { - addReply(c,shared.wrongtypeerr); - return; - } - /* Add the element to the target list */ - dstlist = dobj->ptr; - listAddNodeHead(dstlist,ele); - incrRefCount(ele); - - /* Send the element to the client as reply as well */ - addReplyBulkLen(c,ele); - addReply(c,ele); - addReply(c,shared.crlf); - - /* Finally remove the element from the source list */ - listDelNode(srclist,ln); - server.dirty++; + /* Add the element to the target list (unless it's directly + * passed to some BLPOP-ing client */ + if (!handleClientsWaitingListPush(c,c->argv[2],ele)) { + if (dobj == NULL) { + /* Create the list if the key does not exist */ + dobj = createListObject(); + dictAdd(c->db->dict,c->argv[2],dobj); + incrRefCount(c->argv[2]); } + dstlist = dobj->ptr; + listAddNodeHead(dstlist,ele); + incrRefCount(ele); } + + /* Send the element to the client as reply as well */ + addReplyBulk(c,ele); + + /* Finally remove the element from the source list */ + listDelNode(srclist,ln); + if (listLength(srclist) == 0) deleteKey(c->db,c->argv[1]); + server.dirty++; } } - /* ==================================== Sets ================================ */ static void saddCommand(redisClient *c) { @@ -3902,21 +4794,16 @@ static void saddCommand(redisClient *c) { static void sremCommand(redisClient *c) { robj *set; - set = lookupKeyWrite(c->db,c->argv[1]); - if (set == NULL) { - addReply(c,shared.czero); + if ((set = lookupKeyWriteOrReply(c,c->argv[1],shared.czero)) == NULL || + checkType(c,set,REDIS_SET)) return; + + if (dictDelete(set->ptr,c->argv[2]) == DICT_OK) { + server.dirty++; + if (htNeedsResize(set->ptr)) dictResize(set->ptr); + if (dictSize((dict*)set->ptr) == 0) deleteKey(c->db,c->argv[1]); + addReply(c,shared.cone); } else { - if (set->type != REDIS_SET) { - addReply(c,shared.wrongtypeerr); - return; - } - if (dictDelete(set->ptr,c->argv[2]) == DICT_OK) { - server.dirty++; - if (htNeedsResize(set->ptr)) dictResize(set->ptr); - addReply(c,shared.cone); - } else { - addReply(c,shared.czero); - } + addReply(c,shared.czero); } } @@ -3943,6 +4830,8 @@ static void smoveCommand(redisClient *c) { addReply(c,shared.czero); return; } + if (dictSize((dict*)srcset->ptr) == 0 && srcset != dstset) + deleteKey(c->db,c->argv[1]); server.dirty++; /* Add the element to the destination set */ if (!dstset) { @@ -3958,65 +4847,44 @@ static void smoveCommand(redisClient *c) { static void sismemberCommand(redisClient *c) { robj *set; - set = lookupKeyRead(c->db,c->argv[1]); - if (set == NULL) { + if ((set = lookupKeyReadOrReply(c,c->argv[1],shared.czero)) == NULL || + checkType(c,set,REDIS_SET)) return; + + if (dictFind(set->ptr,c->argv[2])) + addReply(c,shared.cone); + else addReply(c,shared.czero); - } else { - if (set->type != REDIS_SET) { - addReply(c,shared.wrongtypeerr); - return; - } - if (dictFind(set->ptr,c->argv[2])) - addReply(c,shared.cone); - else - addReply(c,shared.czero); - } } static void scardCommand(redisClient *c) { robj *o; dict *s; - - o = lookupKeyRead(c->db,c->argv[1]); - if (o == NULL) { - addReply(c,shared.czero); - return; - } else { - if (o->type != REDIS_SET) { - addReply(c,shared.wrongtypeerr); - } else { - s = o->ptr; - addReplySds(c,sdscatprintf(sdsempty(),":%lu\r\n", - dictSize(s))); - } - } + + if ((o = lookupKeyReadOrReply(c,c->argv[1],shared.czero)) == NULL || + checkType(c,o,REDIS_SET)) return; + + s = o->ptr; + addReplyUlong(c,dictSize(s)); } static void spopCommand(redisClient *c) { robj *set; dictEntry *de; - set = lookupKeyWrite(c->db,c->argv[1]); - if (set == NULL) { + if ((set = lookupKeyWriteOrReply(c,c->argv[1],shared.nullbulk)) == NULL || + checkType(c,set,REDIS_SET)) return; + + de = dictGetRandomKey(set->ptr); + if (de == NULL) { addReply(c,shared.nullbulk); } else { - if (set->type != REDIS_SET) { - addReply(c,shared.wrongtypeerr); - return; - } - de = dictGetRandomKey(set->ptr); - if (de == NULL) { - addReply(c,shared.nullbulk); - } else { - robj *ele = dictGetEntryKey(de); + robj *ele = dictGetEntryKey(de); - addReplyBulkLen(c,ele); - addReply(c,ele); - addReply(c,shared.crlf); - dictDelete(set->ptr,ele); - if (htNeedsResize(set->ptr)) dictResize(set->ptr); - server.dirty++; - } + addReplyBulk(c,ele); + dictDelete(set->ptr,ele); + if (htNeedsResize(set->ptr)) dictResize(set->ptr); + if (dictSize((dict*)set->ptr) == 0) deleteKey(c->db,c->argv[1]); + server.dirty++; } } @@ -4024,24 +4892,16 @@ static void srandmemberCommand(redisClient *c) { robj *set; dictEntry *de; - set = lookupKeyRead(c->db,c->argv[1]); - if (set == NULL) { + if ((set = lookupKeyReadOrReply(c,c->argv[1],shared.nullbulk)) == NULL || + checkType(c,set,REDIS_SET)) return; + + de = dictGetRandomKey(set->ptr); + if (de == NULL) { addReply(c,shared.nullbulk); } else { - if (set->type != REDIS_SET) { - addReply(c,shared.wrongtypeerr); - return; - } - de = dictGetRandomKey(set->ptr); - if (de == NULL) { - addReply(c,shared.nullbulk); - } else { - robj *ele = dictGetEntryKey(de); + robj *ele = dictGetEntryKey(de); - addReplyBulkLen(c,ele); - addReply(c,ele); - addReply(c,shared.crlf); - } + addReplyBulk(c,ele); } } @@ -4071,7 +4931,7 @@ static void sinterGenericCommand(redisClient *c, robj **setskeys, unsigned long server.dirty++; addReply(c,shared.czero); } else { - addReply(c,shared.nullmultibulk); + addReply(c,shared.emptymultibulk); } return; } @@ -4115,9 +4975,7 @@ static void sinterGenericCommand(redisClient *c, robj **setskeys, unsigned long continue; /* at least one set does not contain the member */ ele = dictGetEntryKey(de); if (!dstkey) { - addReplyBulkLen(c,ele); - addReply(c,ele); - addReply(c,shared.crlf); + addReplyBulk(c,ele); cardinality++; } else { dictAdd(dstset->ptr,ele,NULL); @@ -4127,18 +4985,20 @@ static void sinterGenericCommand(redisClient *c, robj **setskeys, unsigned long dictReleaseIterator(di); if (dstkey) { - /* Store the resulting set into the target */ + /* Store the resulting set into the target, if the intersection + * is not an empty set. */ deleteKey(c->db,dstkey); - dictAdd(c->db->dict,dstkey,dstset); - incrRefCount(dstkey); - } - - if (!dstkey) { - lenobj->ptr = sdscatprintf(sdsempty(),"*%lu\r\n",cardinality); - } else { - addReplySds(c,sdscatprintf(sdsempty(),":%lu\r\n", - dictSize((dict*)dstset->ptr))); + if (dictSize((dict*)dstset->ptr) > 0) { + dictAdd(c->db->dict,dstkey,dstset); + incrRefCount(dstkey); + addReplyLong(c,dictSize((dict*)dstset->ptr)); + } else { + decrRefCount(dstset); + addReply(c,shared.czero); + } server.dirty++; + } else { + lenobj->ptr = sdscatprintf(sdsempty(),"*%lu\r\n",cardinality); } zfree(dv); } @@ -4153,6 +5013,7 @@ static void sinterstoreCommand(redisClient *c) { #define REDIS_OP_UNION 0 #define REDIS_OP_DIFF 1 +#define REDIS_OP_INTER 2 static void sunionDiffGenericCommand(redisClient *c, robj **setskeys, int setsnum, robj *dstkey, int op) { dict **dv = zmalloc(sizeof(dict*)*setsnum); @@ -4210,7 +5071,8 @@ static void sunionDiffGenericCommand(redisClient *c, robj **setskeys, int setsnu } dictReleaseIterator(di); - if (op == REDIS_OP_DIFF && cardinality == 0) break; /* result set is empty */ + /* result set is empty? Exit asap. */ + if (op == REDIS_OP_DIFF && cardinality == 0) break; } /* Output the content of the resulting set, if not in STORE mode */ @@ -4221,25 +5083,22 @@ static void sunionDiffGenericCommand(redisClient *c, robj **setskeys, int setsnu robj *ele; ele = dictGetEntryKey(de); - addReplyBulkLen(c,ele); - addReply(c,ele); - addReply(c,shared.crlf); + addReplyBulk(c,ele); } dictReleaseIterator(di); + decrRefCount(dstset); } else { /* If we have a target key where to store the resulting set * create this key with the result set inside */ deleteKey(c->db,dstkey); - dictAdd(c->db->dict,dstkey,dstset); - incrRefCount(dstkey); - } - - /* Cleanup */ - if (!dstkey) { - decrRefCount(dstset); - } else { - addReplySds(c,sdscatprintf(sdsempty(),":%lu\r\n", - dictSize((dict*)dstset->ptr))); + if (dictSize((dict*)dstset->ptr) > 0) { + dictAdd(c->db->dict,dstkey,dstset); + incrRefCount(dstkey); + addReplyLong(c,dictSize((dict*)dstset->ptr)); + } else { + decrRefCount(dstset); + addReply(c,shared.czero); + } server.dirty++; } zfree(dv); @@ -4284,6 +5143,8 @@ static zskiplistNode *zslCreateNode(int level, double score, robj *obj) { zskiplistNode *zn = zmalloc(sizeof(*zn)); zn->forward = zmalloc(sizeof(zskiplistNode*) * level); + if (level > 0) + zn->span = zmalloc(sizeof(unsigned int) * (level - 1)); zn->score = score; zn->obj = obj; return zn; @@ -4292,13 +5153,18 @@ static zskiplistNode *zslCreateNode(int level, double score, robj *obj) { static zskiplist *zslCreate(void) { int j; zskiplist *zsl; - + zsl = zmalloc(sizeof(*zsl)); zsl->level = 1; zsl->length = 0; zsl->header = zslCreateNode(ZSKIPLIST_MAXLEVEL,0,NULL); - for (j = 0; j < ZSKIPLIST_MAXLEVEL; j++) + for (j = 0; j < ZSKIPLIST_MAXLEVEL; j++) { zsl->header->forward[j] = NULL; + + /* span has space for ZSKIPLIST_MAXLEVEL-1 elements */ + if (j < ZSKIPLIST_MAXLEVEL-1) + zsl->header->span[j] = 0; + } zsl->header->backward = NULL; zsl->tail = NULL; return zsl; @@ -4307,6 +5173,7 @@ static zskiplist *zslCreate(void) { static void zslFreeNode(zskiplistNode *node) { decrRefCount(node->obj); zfree(node->forward); + zfree(node->span); zfree(node); } @@ -4314,6 +5181,7 @@ static void zslFree(zskiplist *zsl) { zskiplistNode *node = zsl->header->forward[0], *next; zfree(zsl->header->forward); + zfree(zsl->header->span); zfree(zsl->header); while(node) { next = node->forward[0]; @@ -4327,20 +5195,26 @@ static int zslRandomLevel(void) { int level = 1; while ((random()&0xFFFF) < (ZSKIPLIST_P * 0xFFFF)) level += 1; - return level; + return (levelheader; for (i = zsl->level-1; i >= 0; i--) { + /* store rank that is crossed to reach the insert position */ + rank[i] = i == (zsl->level-1) ? 0 : rank[i+1]; + while (x->forward[i] && (x->forward[i]->score < score || (x->forward[i]->score == score && - compareStringObjects(x->forward[i]->obj,obj) < 0))) + compareStringObjects(x->forward[i]->obj,obj) < 0))) { + rank[i] += i > 0 ? x->span[i-1] : 1; x = x->forward[i]; + } update[i] = x; } /* we assume the key is not already inside, since we allow duplicated @@ -4349,15 +5223,30 @@ static void zslInsert(zskiplist *zsl, double score, robj *obj) { * if the element is already inside or not. */ level = zslRandomLevel(); if (level > zsl->level) { - for (i = zsl->level; i < level; i++) + for (i = zsl->level; i < level; i++) { + rank[i] = 0; update[i] = zsl->header; + update[i]->span[i-1] = zsl->length; + } zsl->level = level; } x = zslCreateNode(level,score,obj); for (i = 0; i < level; i++) { x->forward[i] = update[i]->forward[i]; update[i]->forward[i] = x; + + /* update span covered by update[i] as x is inserted here */ + if (i > 0) { + x->span[i-1] = update[i]->span[i-1] - (rank[0] - rank[i]); + update[i]->span[i-1] = (rank[0] - rank[i]) + 1; + } + } + + /* increment span for untouched levels */ + for (i = level; i < zsl->level; i++) { + update[i]->span[i-1]++; } + x->backward = (update[0] == zsl->header) ? NULL : update[0]; if (x->forward[0]) x->forward[0]->backward = x; @@ -4366,6 +5255,31 @@ static void zslInsert(zskiplist *zsl, double score, robj *obj) { zsl->length++; } +/* Internal function used by zslDelete, zslDeleteByScore and zslDeleteByRank */ +void zslDeleteNode(zskiplist *zsl, zskiplistNode *x, zskiplistNode **update) { + int i; + for (i = 0; i < zsl->level; i++) { + if (update[i]->forward[i] == x) { + if (i > 0) { + update[i]->span[i-1] += x->span[i-1] - 1; + } + update[i]->forward[i] = x->forward[i]; + } else { + /* invariant: i > 0, because update[0]->forward[0] + * is always equal to x */ + update[i]->span[i-1] -= 1; + } + } + if (x->forward[0]) { + x->forward[0]->backward = x->backward; + } else { + zsl->tail = x->backward; + } + while(zsl->level > 1 && zsl->header->forward[zsl->level-1] == NULL) + zsl->level--; + zsl->length--; +} + /* Delete an element with matching score/object from the skiplist. */ static int zslDelete(zskiplist *zsl, double score, robj *obj) { zskiplistNode *update[ZSKIPLIST_MAXLEVEL], *x; @@ -4384,20 +5298,8 @@ static int zslDelete(zskiplist *zsl, double score, robj *obj) { * is to find the element with both the right score and object. */ x = x->forward[0]; if (x && score == x->score && compareStringObjects(x->obj,obj) == 0) { - for (i = 0; i < zsl->level; i++) { - if (update[i]->forward[i] != x) break; - update[i]->forward[i] = x->forward[i]; - } - if (x->forward[0]) { - x->forward[0]->backward = (x->backward == zsl->header) ? - NULL : x->backward; - } else { - zsl->tail = x->backward; - } + zslDeleteNode(zsl, x, update); zslFreeNode(x); - while(zsl->level > 1 && zsl->header->forward[zsl->level-1] == NULL) - zsl->level--; - zsl->length--; return 1; } else { return 0; /* not found */ @@ -4409,7 +5311,7 @@ static int zslDelete(zskiplist *zsl, double score, robj *obj) { * Min and mx are inclusive, so a score >= min || score <= max is deleted. * Note that this function takes the reference to the hash table view of the * sorted set, in order to remove the elements from the hash table too. */ -static unsigned long zslDeleteRange(zskiplist *zsl, double min, double max, dict *dict) { +static unsigned long zslDeleteRangeByScore(zskiplist *zsl, double min, double max, dict *dict) { zskiplistNode *update[ZSKIPLIST_MAXLEVEL], *x; unsigned long removed = 0; int i; @@ -4424,45 +5326,108 @@ static unsigned long zslDeleteRange(zskiplist *zsl, double min, double max, dict * is to find the element with both the right score and object. */ x = x->forward[0]; while (x && x->score <= max) { - zskiplistNode *next; - - for (i = 0; i < zsl->level; i++) { - if (update[i]->forward[i] != x) break; - update[i]->forward[i] = x->forward[i]; - } - if (x->forward[0]) { - x->forward[0]->backward = (x->backward == zsl->header) ? - NULL : x->backward; - } else { - zsl->tail = x->backward; - } - next = x->forward[0]; + zskiplistNode *next = x->forward[0]; + zslDeleteNode(zsl, x, update); dictDelete(dict,x->obj); zslFreeNode(x); - while(zsl->level > 1 && zsl->header->forward[zsl->level-1] == NULL) - zsl->level--; - zsl->length--; removed++; x = next; } return removed; /* not found */ } -/* Find the first node having a score equal or greater than the specified one. - * Returns NULL if there is no match. */ -static zskiplistNode *zslFirstWithScore(zskiplist *zsl, double score) { - zskiplistNode *x; +/* Delete all the elements with rank between start and end from the skiplist. + * Start and end are inclusive. Note that start and end need to be 1-based */ +static unsigned long zslDeleteRangeByRank(zskiplist *zsl, unsigned int start, unsigned int end, dict *dict) { + zskiplistNode *update[ZSKIPLIST_MAXLEVEL], *x; + unsigned long traversed = 0, removed = 0; int i; x = zsl->header; for (i = zsl->level-1; i >= 0; i--) { - while (x->forward[i] && x->forward[i]->score < score) + while (x->forward[i] && (traversed + (i > 0 ? x->span[i-1] : 1)) < start) { + traversed += i > 0 ? x->span[i-1] : 1; x = x->forward[i]; + } + update[i] = x; } - /* We may have multiple elements with the same score, what we need - * is to find the element with both the right score and object. */ - return x->forward[0]; -} + + traversed++; + x = x->forward[0]; + while (x && traversed <= end) { + zskiplistNode *next = x->forward[0]; + zslDeleteNode(zsl, x, update); + dictDelete(dict,x->obj); + zslFreeNode(x); + removed++; + traversed++; + x = next; + } + return removed; +} + +/* Find the first node having a score equal or greater than the specified one. + * Returns NULL if there is no match. */ +static zskiplistNode *zslFirstWithScore(zskiplist *zsl, double score) { + zskiplistNode *x; + int i; + + x = zsl->header; + for (i = zsl->level-1; i >= 0; i--) { + while (x->forward[i] && x->forward[i]->score < score) + x = x->forward[i]; + } + /* We may have multiple elements with the same score, what we need + * is to find the element with both the right score and object. */ + return x->forward[0]; +} + +/* Find the rank for an element by both score and key. + * Returns 0 when the element cannot be found, rank otherwise. + * Note that the rank is 1-based due to the span of zsl->header to the + * first element. */ +static unsigned long zslGetRank(zskiplist *zsl, double score, robj *o) { + zskiplistNode *x; + unsigned long rank = 0; + int i; + + x = zsl->header; + for (i = zsl->level-1; i >= 0; i--) { + while (x->forward[i] && + (x->forward[i]->score < score || + (x->forward[i]->score == score && + compareStringObjects(x->forward[i]->obj,o) <= 0))) { + rank += i > 0 ? x->span[i-1] : 1; + x = x->forward[i]; + } + + /* x might be equal to zsl->header, so test if obj is non-NULL */ + if (x->obj && compareStringObjects(x->obj,o) == 0) { + return rank; + } + } + return 0; +} + +/* Finds an element by its rank. The rank argument needs to be 1-based. */ +zskiplistNode* zslGetElementByRank(zskiplist *zsl, unsigned long rank) { + zskiplistNode *x; + unsigned long traversed = 0; + int i; + + x = zsl->header; + for (i = zsl->level-1; i >= 0; i--) { + while (x->forward[i] && (traversed + (i>0 ? x->span[i-1] : 1)) <= rank) + { + traversed += i > 0 ? x->span[i-1] : 1; + x = x->forward[i]; + } + if (traversed == rank) { + return x; + } + } + return NULL; +} /* The actual Z-commands implementations */ @@ -4521,7 +5486,7 @@ static void zaddGenericCommand(redisClient *c, robj *key, robj *ele, double scor } else { dictEntry *de; double *oldscore; - + /* case 2: Score update operation */ de = dictFind(zs->dict,ele); redisAssert(de != NULL); @@ -4550,81 +5515,319 @@ static void zaddGenericCommand(redisClient *c, robj *key, robj *ele, double scor static void zaddCommand(redisClient *c) { double scoreval; - scoreval = strtod(c->argv[2]->ptr,NULL); + if (getDoubleFromObjectOrReply(c, c->argv[2], &scoreval, NULL) != REDIS_OK) return; zaddGenericCommand(c,c->argv[1],c->argv[3],scoreval,0); } static void zincrbyCommand(redisClient *c) { double scoreval; - scoreval = strtod(c->argv[2]->ptr,NULL); + if (getDoubleFromObjectOrReply(c, c->argv[2], &scoreval, NULL) != REDIS_OK) return; zaddGenericCommand(c,c->argv[1],c->argv[3],scoreval,1); } static void zremCommand(redisClient *c) { robj *zsetobj; zset *zs; + dictEntry *de; + double *oldscore; + int deleted; - zsetobj = lookupKeyWrite(c->db,c->argv[1]); - if (zsetobj == NULL) { - addReply(c,shared.czero); - } else { - dictEntry *de; - double *oldscore; - int deleted; - - if (zsetobj->type != REDIS_ZSET) { - addReply(c,shared.wrongtypeerr); - return; - } - zs = zsetobj->ptr; - de = dictFind(zs->dict,c->argv[2]); - if (de == NULL) { - addReply(c,shared.czero); - return; - } - /* Delete from the skiplist */ - oldscore = dictGetEntryVal(de); - deleted = zslDelete(zs->zsl,*oldscore,c->argv[2]); - redisAssert(deleted != 0); + if ((zsetobj = lookupKeyWriteOrReply(c,c->argv[1],shared.czero)) == NULL || + checkType(c,zsetobj,REDIS_ZSET)) return; - /* Delete from the hash table */ - dictDelete(zs->dict,c->argv[2]); - if (htNeedsResize(zs->dict)) dictResize(zs->dict); - server.dirty++; - addReply(c,shared.cone); + zs = zsetobj->ptr; + de = dictFind(zs->dict,c->argv[2]); + if (de == NULL) { + addReply(c,shared.czero); + return; } + /* Delete from the skiplist */ + oldscore = dictGetEntryVal(de); + deleted = zslDelete(zs->zsl,*oldscore,c->argv[2]); + redisAssert(deleted != 0); + + /* Delete from the hash table */ + dictDelete(zs->dict,c->argv[2]); + if (htNeedsResize(zs->dict)) dictResize(zs->dict); + if (dictSize(zs->dict) == 0) deleteKey(c->db,c->argv[1]); + server.dirty++; + addReply(c,shared.cone); } static void zremrangebyscoreCommand(redisClient *c) { - double min = strtod(c->argv[2]->ptr,NULL); - double max = strtod(c->argv[3]->ptr,NULL); + double min; + double max; + long deleted; robj *zsetobj; zset *zs; - zsetobj = lookupKeyWrite(c->db,c->argv[1]); - if (zsetobj == NULL) { + if ((getDoubleFromObjectOrReply(c, c->argv[2], &min, NULL) != REDIS_OK) || + (getDoubleFromObjectOrReply(c, c->argv[3], &max, NULL) != REDIS_OK)) return; + + if ((zsetobj = lookupKeyWriteOrReply(c,c->argv[1],shared.czero)) == NULL || + checkType(c,zsetobj,REDIS_ZSET)) return; + + zs = zsetobj->ptr; + deleted = zslDeleteRangeByScore(zs->zsl,min,max,zs->dict); + if (htNeedsResize(zs->dict)) dictResize(zs->dict); + if (dictSize(zs->dict) == 0) deleteKey(c->db,c->argv[1]); + server.dirty += deleted; + addReplyLong(c,deleted); +} + +static void zremrangebyrankCommand(redisClient *c) { + long start; + long end; + int llen; + long deleted; + robj *zsetobj; + zset *zs; + + if ((getLongFromObjectOrReply(c, c->argv[2], &start, NULL) != REDIS_OK) || + (getLongFromObjectOrReply(c, c->argv[3], &end, NULL) != REDIS_OK)) return; + + if ((zsetobj = lookupKeyWriteOrReply(c,c->argv[1],shared.czero)) == NULL || + checkType(c,zsetobj,REDIS_ZSET)) return; + zs = zsetobj->ptr; + llen = zs->zsl->length; + + /* convert negative indexes */ + if (start < 0) start = llen+start; + if (end < 0) end = llen+end; + if (start < 0) start = 0; + if (end < 0) end = 0; + + /* indexes sanity checks */ + if (start > end || start >= llen) { addReply(c,shared.czero); + return; + } + if (end >= llen) end = llen-1; + + /* increment start and end because zsl*Rank functions + * use 1-based rank */ + deleted = zslDeleteRangeByRank(zs->zsl,start+1,end+1,zs->dict); + if (htNeedsResize(zs->dict)) dictResize(zs->dict); + if (dictSize(zs->dict) == 0) deleteKey(c->db,c->argv[1]); + server.dirty += deleted; + addReplyLong(c, deleted); +} + +typedef struct { + dict *dict; + double weight; +} zsetopsrc; + +static int qsortCompareZsetopsrcByCardinality(const void *s1, const void *s2) { + zsetopsrc *d1 = (void*) s1, *d2 = (void*) s2; + unsigned long size1, size2; + size1 = d1->dict ? dictSize(d1->dict) : 0; + size2 = d2->dict ? dictSize(d2->dict) : 0; + return size1 - size2; +} + +#define REDIS_AGGR_SUM 1 +#define REDIS_AGGR_MIN 2 +#define REDIS_AGGR_MAX 3 + +inline static void zunionInterAggregate(double *target, double val, int aggregate) { + if (aggregate == REDIS_AGGR_SUM) { + *target = *target + val; + } else if (aggregate == REDIS_AGGR_MIN) { + *target = val < *target ? val : *target; + } else if (aggregate == REDIS_AGGR_MAX) { + *target = val > *target ? val : *target; } else { - long deleted; + /* safety net */ + redisPanic("Unknown ZUNION/INTER aggregate type"); + } +} - if (zsetobj->type != REDIS_ZSET) { - addReply(c,shared.wrongtypeerr); - return; +static void zunionInterGenericCommand(redisClient *c, robj *dstkey, int op) { + int i, j, zsetnum; + int aggregate = REDIS_AGGR_SUM; + zsetopsrc *src; + robj *dstobj; + zset *dstzset; + dictIterator *di; + dictEntry *de; + + /* expect zsetnum input keys to be given */ + zsetnum = atoi(c->argv[2]->ptr); + if (zsetnum < 1) { + addReplySds(c,sdsnew("-ERR at least 1 input key is needed for ZUNION/ZINTER\r\n")); + return; + } + + /* test if the expected number of keys would overflow */ + if (3+zsetnum > c->argc) { + addReply(c,shared.syntaxerr); + return; + } + + /* read keys to be used for input */ + src = zmalloc(sizeof(zsetopsrc) * zsetnum); + for (i = 0, j = 3; i < zsetnum; i++, j++) { + robj *zsetobj = lookupKeyWrite(c->db,c->argv[j]); + if (!zsetobj) { + src[i].dict = NULL; + } else { + if (zsetobj->type != REDIS_ZSET) { + zfree(src); + addReply(c,shared.wrongtypeerr); + return; + } + src[i].dict = ((zset*)zsetobj->ptr)->dict; + } + + /* default all weights to 1 */ + src[i].weight = 1.0; + } + + /* parse optional extra arguments */ + if (j < c->argc) { + int remaining = c->argc - j; + + while (remaining) { + if (remaining >= (zsetnum + 1) && !strcasecmp(c->argv[j]->ptr,"weights")) { + j++; remaining--; + for (i = 0; i < zsetnum; i++, j++, remaining--) { + if (getDoubleFromObjectOrReply(c, c->argv[j], &src[i].weight, NULL) != REDIS_OK) + return; + } + } else if (remaining >= 2 && !strcasecmp(c->argv[j]->ptr,"aggregate")) { + j++; remaining--; + if (!strcasecmp(c->argv[j]->ptr,"sum")) { + aggregate = REDIS_AGGR_SUM; + } else if (!strcasecmp(c->argv[j]->ptr,"min")) { + aggregate = REDIS_AGGR_MIN; + } else if (!strcasecmp(c->argv[j]->ptr,"max")) { + aggregate = REDIS_AGGR_MAX; + } else { + zfree(src); + addReply(c,shared.syntaxerr); + return; + } + j++; remaining--; + } else { + zfree(src); + addReply(c,shared.syntaxerr); + return; + } + } + } + + /* sort sets from the smallest to largest, this will improve our + * algorithm's performance */ + qsort(src,zsetnum,sizeof(zsetopsrc), qsortCompareZsetopsrcByCardinality); + + dstobj = createZsetObject(); + dstzset = dstobj->ptr; + + if (op == REDIS_OP_INTER) { + /* skip going over all entries if the smallest zset is NULL or empty */ + if (src[0].dict && dictSize(src[0].dict) > 0) { + /* precondition: as src[0].dict is non-empty and the zsets are ordered + * from small to large, all src[i > 0].dict are non-empty too */ + di = dictGetIterator(src[0].dict); + while((de = dictNext(di)) != NULL) { + double *score = zmalloc(sizeof(double)), value; + *score = src[0].weight * (*(double*)dictGetEntryVal(de)); + + for (j = 1; j < zsetnum; j++) { + dictEntry *other = dictFind(src[j].dict,dictGetEntryKey(de)); + if (other) { + value = src[j].weight * (*(double*)dictGetEntryVal(other)); + zunionInterAggregate(score, value, aggregate); + } else { + break; + } + } + + /* skip entry when not present in every source dict */ + if (j != zsetnum) { + zfree(score); + } else { + robj *o = dictGetEntryKey(de); + dictAdd(dstzset->dict,o,score); + incrRefCount(o); /* added to dictionary */ + zslInsert(dstzset->zsl,*score,o); + incrRefCount(o); /* added to skiplist */ + } + } + dictReleaseIterator(di); + } + } else if (op == REDIS_OP_UNION) { + for (i = 0; i < zsetnum; i++) { + if (!src[i].dict) continue; + + di = dictGetIterator(src[i].dict); + while((de = dictNext(di)) != NULL) { + /* skip key when already processed */ + if (dictFind(dstzset->dict,dictGetEntryKey(de)) != NULL) continue; + + double *score = zmalloc(sizeof(double)), value; + *score = src[i].weight * (*(double*)dictGetEntryVal(de)); + + /* because the zsets are sorted by size, its only possible + * for sets at larger indices to hold this entry */ + for (j = (i+1); j < zsetnum; j++) { + dictEntry *other = dictFind(src[j].dict,dictGetEntryKey(de)); + if (other) { + value = src[j].weight * (*(double*)dictGetEntryVal(other)); + zunionInterAggregate(score, value, aggregate); + } + } + + robj *o = dictGetEntryKey(de); + dictAdd(dstzset->dict,o,score); + incrRefCount(o); /* added to dictionary */ + zslInsert(dstzset->zsl,*score,o); + incrRefCount(o); /* added to skiplist */ + } + dictReleaseIterator(di); } - zs = zsetobj->ptr; - deleted = zslDeleteRange(zs->zsl,min,max,zs->dict); - if (htNeedsResize(zs->dict)) dictResize(zs->dict); - server.dirty += deleted; - addReplySds(c,sdscatprintf(sdsempty(),":%lu\r\n",deleted)); + } else { + /* unknown operator */ + redisAssert(op == REDIS_OP_INTER || op == REDIS_OP_UNION); + } + + deleteKey(c->db,dstkey); + if (dstzset->zsl->length) { + dictAdd(c->db->dict,dstkey,dstobj); + incrRefCount(dstkey); + addReplyLong(c, dstzset->zsl->length); + server.dirty++; + } else { + decrRefCount(dstobj); + addReply(c, shared.czero); } + zfree(src); +} + +static void zunionCommand(redisClient *c) { + zunionInterGenericCommand(c,c->argv[1], REDIS_OP_UNION); +} + +static void zinterCommand(redisClient *c) { + zunionInterGenericCommand(c,c->argv[1], REDIS_OP_INTER); } static void zrangeGenericCommand(redisClient *c, int reverse) { robj *o; - int start = atoi(c->argv[2]->ptr); - int end = atoi(c->argv[3]->ptr); + long start; + long end; int withscores = 0; + int llen; + int rangelen, j; + zset *zsetobj; + zskiplist *zsl; + zskiplistNode *ln; + robj *ele; + + if ((getLongFromObjectOrReply(c, c->argv[2], &start, NULL) != REDIS_OK) || + (getLongFromObjectOrReply(c, c->argv[3], &end, NULL) != REDIS_OK)) return; if (c->argc == 5 && !strcasecmp(c->argv[4]->ptr,"withscores")) { withscores = 1; @@ -4633,59 +5836,45 @@ static void zrangeGenericCommand(redisClient *c, int reverse) { return; } - o = lookupKeyRead(c->db,c->argv[1]); - if (o == NULL) { - addReply(c,shared.nullmultibulk); - } else { - if (o->type != REDIS_ZSET) { - addReply(c,shared.wrongtypeerr); - } else { - zset *zsetobj = o->ptr; - zskiplist *zsl = zsetobj->zsl; - zskiplistNode *ln; + if ((o = lookupKeyReadOrReply(c,c->argv[1],shared.emptymultibulk)) == NULL + || checkType(c,o,REDIS_ZSET)) return; + zsetobj = o->ptr; + zsl = zsetobj->zsl; + llen = zsl->length; - int llen = zsl->length; - int rangelen, j; - robj *ele; + /* convert negative indexes */ + if (start < 0) start = llen+start; + if (end < 0) end = llen+end; + if (start < 0) start = 0; + if (end < 0) end = 0; - /* convert negative indexes */ - if (start < 0) start = llen+start; - if (end < 0) end = llen+end; - if (start < 0) start = 0; - if (end < 0) end = 0; + /* indexes sanity checks */ + if (start > end || start >= llen) { + /* Out of range start or start > end result in empty list */ + addReply(c,shared.emptymultibulk); + return; + } + if (end >= llen) end = llen-1; + rangelen = (end-start)+1; - /* indexes sanity checks */ - if (start > end || start >= llen) { - /* Out of range start or start > end result in empty list */ - addReply(c,shared.emptymultibulk); - return; - } - if (end >= llen) end = llen-1; - rangelen = (end-start)+1; - - /* Return the result in form of a multi-bulk reply */ - if (reverse) { - ln = zsl->tail; - while (start--) - ln = ln->backward; - } else { - ln = zsl->header->forward[0]; - while (start--) - ln = ln->forward[0]; - } + /* check if starting point is trivial, before searching + * the element in log(N) time */ + if (reverse) { + ln = start == 0 ? zsl->tail : zslGetElementByRank(zsl, llen-start); + } else { + ln = start == 0 ? + zsl->header->forward[0] : zslGetElementByRank(zsl, start+1); + } - addReplySds(c,sdscatprintf(sdsempty(),"*%d\r\n", - withscores ? (rangelen*2) : rangelen)); - for (j = 0; j < rangelen; j++) { - ele = ln->obj; - addReplyBulkLen(c,ele); - addReply(c,ele); - addReply(c,shared.crlf); - if (withscores) - addReplyDouble(c,ln->score); - ln = reverse ? ln->backward : ln->forward[0]; - } - } + /* Return the result in form of a multi-bulk reply */ + addReplySds(c,sdscatprintf(sdsempty(),"*%d\r\n", + withscores ? (rangelen*2) : rangelen)); + for (j = 0; j < rangelen; j++) { + ele = ln->obj; + addReplyBulk(c,ele); + if (withscores) + addReplyDouble(c,ln->score); + ln = reverse ? ln->backward : ln->forward[0]; } } @@ -4697,28 +5886,64 @@ static void zrevrangeCommand(redisClient *c) { zrangeGenericCommand(c,1); } -static void zrangebyscoreCommand(redisClient *c) { +/* This command implements both ZRANGEBYSCORE and ZCOUNT. + * If justcount is non-zero, just the count is returned. */ +static void genericZrangebyscoreCommand(redisClient *c, int justcount) { robj *o; - double min = strtod(c->argv[2]->ptr,NULL); - double max = strtod(c->argv[3]->ptr,NULL); + double min, max; + int minex = 0, maxex = 0; /* are min or max exclusive? */ int offset = 0, limit = -1; + int withscores = 0; + int badsyntax = 0; + + /* Parse the min-max interval. If one of the values is prefixed + * by the "(" character, it's considered "open". For instance + * ZRANGEBYSCORE zset (1.5 (2.5 will match min < x < max + * ZRANGEBYSCORE zset 1.5 2.5 will instead match min <= x <= max */ + if (((char*)c->argv[2]->ptr)[0] == '(') { + min = strtod((char*)c->argv[2]->ptr+1,NULL); + minex = 1; + } else { + min = strtod(c->argv[2]->ptr,NULL); + } + if (((char*)c->argv[3]->ptr)[0] == '(') { + max = strtod((char*)c->argv[3]->ptr+1,NULL); + maxex = 1; + } else { + max = strtod(c->argv[3]->ptr,NULL); + } - if (c->argc != 4 && c->argc != 7) { + /* Parse "WITHSCORES": note that if the command was called with + * the name ZCOUNT then we are sure that c->argc == 4, so we'll never + * enter the following paths to parse WITHSCORES and LIMIT. */ + if (c->argc == 5 || c->argc == 8) { + if (strcasecmp(c->argv[c->argc-1]->ptr,"withscores") == 0) + withscores = 1; + else + badsyntax = 1; + } + if (c->argc != (4 + withscores) && c->argc != (7 + withscores)) + badsyntax = 1; + if (badsyntax) { addReplySds(c, sdsnew("-ERR wrong number of arguments for ZRANGEBYSCORE\r\n")); return; - } else if (c->argc == 7 && strcasecmp(c->argv[4]->ptr,"limit")) { + } + + /* Parse "LIMIT" */ + if (c->argc == (7 + withscores) && strcasecmp(c->argv[4]->ptr,"limit")) { addReply(c,shared.syntaxerr); return; - } else if (c->argc == 7) { + } else if (c->argc == (7 + withscores)) { offset = atoi(c->argv[5]->ptr); limit = atoi(c->argv[6]->ptr); if (offset < 0) offset = 0; } + /* Ok, lookup the key and get the range */ o = lookupKeyRead(c->db,c->argv[1]); if (o == NULL) { - addReply(c,shared.nullmultibulk); + addReply(c,justcount ? shared.czero : shared.emptymultibulk); } else { if (o->type != REDIS_ZSET) { addReply(c,shared.wrongtypeerr); @@ -4726,14 +5951,17 @@ static void zrangebyscoreCommand(redisClient *c) { zset *zsetobj = o->ptr; zskiplist *zsl = zsetobj->zsl; zskiplistNode *ln; - robj *ele, *lenobj; - unsigned int rangelen = 0; + robj *ele, *lenobj = NULL; + unsigned long rangelen = 0; - /* Get the first node with the score >= min */ + /* Get the first node with the score >= min, or with + * score > min if 'minex' is true. */ ln = zslFirstWithScore(zsl,min); + while (minex && ln && ln->score == min) ln = ln->forward[0]; + if (ln == NULL) { /* No element matching the speciifed interval */ - addReply(c,shared.emptymultibulk); + addReply(c,justcount ? shared.czero : shared.emptymultibulk); return; } @@ -4741,92 +5969,545 @@ static void zrangebyscoreCommand(redisClient *c) { * are in the list, so we push this object that will represent * the multi-bulk length in the output buffer, and will "fix" * it later */ - lenobj = createObject(REDIS_STRING,NULL); - addReply(c,lenobj); - decrRefCount(lenobj); + if (!justcount) { + lenobj = createObject(REDIS_STRING,NULL); + addReply(c,lenobj); + decrRefCount(lenobj); + } - while(ln && ln->score <= max) { + while(ln && (maxex ? (ln->score < max) : (ln->score <= max))) { if (offset) { offset--; ln = ln->forward[0]; continue; } if (limit == 0) break; - ele = ln->obj; - addReplyBulkLen(c,ele); - addReply(c,ele); - addReply(c,shared.crlf); + if (!justcount) { + ele = ln->obj; + addReplyBulk(c,ele); + if (withscores) + addReplyDouble(c,ln->score); + } ln = ln->forward[0]; rangelen++; if (limit > 0) limit--; } - lenobj->ptr = sdscatprintf(sdsempty(),"*%d\r\n",rangelen); + if (justcount) { + addReplyLong(c,(long)rangelen); + } else { + lenobj->ptr = sdscatprintf(sdsempty(),"*%lu\r\n", + withscores ? (rangelen*2) : rangelen); + } } } } +static void zrangebyscoreCommand(redisClient *c) { + genericZrangebyscoreCommand(c,0); +} + +static void zcountCommand(redisClient *c) { + genericZrangebyscoreCommand(c,1); +} + static void zcardCommand(redisClient *c) { robj *o; zset *zs; - - o = lookupKeyRead(c->db,c->argv[1]); - if (o == NULL) { - addReply(c,shared.czero); - return; - } else { - if (o->type != REDIS_ZSET) { - addReply(c,shared.wrongtypeerr); - } else { - zs = o->ptr; - addReplySds(c,sdscatprintf(sdsempty(),":%lu\r\n",zs->zsl->length)); - } - } + + if ((o = lookupKeyReadOrReply(c,c->argv[1],shared.czero)) == NULL || + checkType(c,o,REDIS_ZSET)) return; + + zs = o->ptr; + addReplyUlong(c,zs->zsl->length); } static void zscoreCommand(redisClient *c) { robj *o; zset *zs; - - o = lookupKeyRead(c->db,c->argv[1]); - if (o == NULL) { + dictEntry *de; + + if ((o = lookupKeyReadOrReply(c,c->argv[1],shared.nullbulk)) == NULL || + checkType(c,o,REDIS_ZSET)) return; + + zs = o->ptr; + de = dictFind(zs->dict,c->argv[2]); + if (!de) { addReply(c,shared.nullbulk); - return; } else { - if (o->type != REDIS_ZSET) { - addReply(c,shared.wrongtypeerr); - } else { - dictEntry *de; + double *score = dictGetEntryVal(de); - zs = o->ptr; - de = dictFind(zs->dict,c->argv[2]); - if (!de) { - addReply(c,shared.nullbulk); - } else { - double *score = dictGetEntryVal(de); + addReplyDouble(c,*score); + } +} - addReplyDouble(c,*score); - } +static void zrankGenericCommand(redisClient *c, int reverse) { + robj *o; + zset *zs; + zskiplist *zsl; + dictEntry *de; + unsigned long rank; + double *score; + + if ((o = lookupKeyReadOrReply(c,c->argv[1],shared.nullbulk)) == NULL || + checkType(c,o,REDIS_ZSET)) return; + + zs = o->ptr; + zsl = zs->zsl; + de = dictFind(zs->dict,c->argv[2]); + if (!de) { + addReply(c,shared.nullbulk); + return; + } + + score = dictGetEntryVal(de); + rank = zslGetRank(zsl, *score, c->argv[2]); + if (rank) { + if (reverse) { + addReplyLong(c, zsl->length - rank); + } else { + addReplyLong(c, rank-1); } + } else { + addReply(c,shared.nullbulk); } } -/* ========================= Non type-specific commands ==================== */ - -static void flushdbCommand(redisClient *c) { - server.dirty += dictSize(c->db->dict); - dictEmpty(c->db->dict); - dictEmpty(c->db->expires); - addReply(c,shared.ok); +static void zrankCommand(redisClient *c) { + zrankGenericCommand(c, 0); } -static void flushallCommand(redisClient *c) { - server.dirty += emptyDb(); - addReply(c,shared.ok); - rdbSave(server.dbfilename); - server.dirty++; +static void zrevrankCommand(redisClient *c) { + zrankGenericCommand(c, 1); } -static redisSortOperation *createSortOperation(int type, robj *pattern) { +/* ========================= Hashes utility functions ======================= */ +#define REDIS_HASH_KEY 1 +#define REDIS_HASH_VALUE 2 + +/* Check the length of a number of objects to see if we need to convert a + * zipmap to a real hash. Note that we only check string encoded objects + * as their string length can be queried in constant time. */ +static void hashTryConversion(robj *subject, robj **argv, int start, int end) { + int i; + if (subject->encoding != REDIS_ENCODING_ZIPMAP) return; + + for (i = start; i <= end; i++) { + if (argv[i]->encoding == REDIS_ENCODING_RAW && + sdslen(argv[i]->ptr) > server.hash_max_zipmap_value) + { + convertToRealHash(subject); + return; + } + } +} + +/* Encode given objects in-place when the hash uses a dict. */ +static void hashTryObjectEncoding(robj *subject, robj **o1, robj **o2) { + if (subject->encoding == REDIS_ENCODING_HT) { + if (o1) *o1 = tryObjectEncoding(*o1); + if (o2) *o2 = tryObjectEncoding(*o2); + } +} + +/* Get the value from a hash identified by key. Returns either a string + * object or NULL if the value cannot be found. The refcount of the object + * is always increased by 1 when the value was found. */ +static robj *hashGet(robj *o, robj *key) { + robj *value = NULL; + if (o->encoding == REDIS_ENCODING_ZIPMAP) { + unsigned char *v; + unsigned int vlen; + key = getDecodedObject(key); + if (zipmapGet(o->ptr,key->ptr,sdslen(key->ptr),&v,&vlen)) { + value = createStringObject((char*)v,vlen); + } + decrRefCount(key); + } else { + dictEntry *de = dictFind(o->ptr,key); + if (de != NULL) { + value = dictGetEntryVal(de); + incrRefCount(value); + } + } + return value; +} + +/* Test if the key exists in the given hash. Returns 1 if the key + * exists and 0 when it doesn't. */ +static int hashExists(robj *o, robj *key) { + if (o->encoding == REDIS_ENCODING_ZIPMAP) { + key = getDecodedObject(key); + if (zipmapExists(o->ptr,key->ptr,sdslen(key->ptr))) { + decrRefCount(key); + return 1; + } + decrRefCount(key); + } else { + if (dictFind(o->ptr,key) != NULL) { + return 1; + } + } + return 0; +} + +/* Add an element, discard the old if the key already exists. + * Return 0 on insert and 1 on update. */ +static int hashSet(robj *o, robj *key, robj *value) { + int update = 0; + if (o->encoding == REDIS_ENCODING_ZIPMAP) { + key = getDecodedObject(key); + value = getDecodedObject(value); + o->ptr = zipmapSet(o->ptr, + key->ptr,sdslen(key->ptr), + value->ptr,sdslen(value->ptr), &update); + decrRefCount(key); + decrRefCount(value); + + /* Check if the zipmap needs to be upgraded to a real hash table */ + if (zipmapLen(o->ptr) > server.hash_max_zipmap_entries) + convertToRealHash(o); + } else { + if (dictReplace(o->ptr,key,value)) { + /* Insert */ + incrRefCount(key); + } else { + /* Update */ + update = 1; + } + incrRefCount(value); + } + return update; +} + +/* Delete an element from a hash. + * Return 1 on deleted and 0 on not found. */ +static int hashDelete(robj *o, robj *key) { + int deleted = 0; + if (o->encoding == REDIS_ENCODING_ZIPMAP) { + key = getDecodedObject(key); + o->ptr = zipmapDel(o->ptr,key->ptr,sdslen(key->ptr), &deleted); + decrRefCount(key); + } else { + deleted = dictDelete((dict*)o->ptr,key) == DICT_OK; + /* Always check if the dictionary needs a resize after a delete. */ + if (deleted && htNeedsResize(o->ptr)) dictResize(o->ptr); + } + return deleted; +} + +/* Return the number of elements in a hash. */ +static unsigned long hashLength(robj *o) { + return (o->encoding == REDIS_ENCODING_ZIPMAP) ? + zipmapLen((unsigned char*)o->ptr) : dictSize((dict*)o->ptr); +} + +/* Structure to hold hash iteration abstration. Note that iteration over + * hashes involves both fields and values. Because it is possible that + * not both are required, store pointers in the iterator to avoid + * unnecessary memory allocation for fields/values. */ +typedef struct { + int encoding; + unsigned char *zi; + unsigned char *zk, *zv; + unsigned int zklen, zvlen; + + dictIterator *di; + dictEntry *de; +} hashIterator; + +static hashIterator *hashInitIterator(robj *subject) { + hashIterator *hi = zmalloc(sizeof(hashIterator)); + hi->encoding = subject->encoding; + if (hi->encoding == REDIS_ENCODING_ZIPMAP) { + hi->zi = zipmapRewind(subject->ptr); + } else if (hi->encoding == REDIS_ENCODING_HT) { + hi->di = dictGetIterator(subject->ptr); + } else { + redisAssert(NULL); + } + return hi; +} + +static void hashReleaseIterator(hashIterator *hi) { + if (hi->encoding == REDIS_ENCODING_HT) { + dictReleaseIterator(hi->di); + } + zfree(hi); +} + +/* Move to the next entry in the hash. Return REDIS_OK when the next entry + * could be found and REDIS_ERR when the iterator reaches the end. */ +static int hashNext(hashIterator *hi) { + if (hi->encoding == REDIS_ENCODING_ZIPMAP) { + if ((hi->zi = zipmapNext(hi->zi, &hi->zk, &hi->zklen, + &hi->zv, &hi->zvlen)) == NULL) return REDIS_ERR; + } else { + if ((hi->de = dictNext(hi->di)) == NULL) return REDIS_ERR; + } + return REDIS_OK; +} + +/* Get key or value object at current iteration position. + * This increases the refcount of the field object by 1. */ +static robj *hashCurrent(hashIterator *hi, int what) { + robj *o; + if (hi->encoding == REDIS_ENCODING_ZIPMAP) { + if (what & REDIS_HASH_KEY) { + o = createStringObject((char*)hi->zk,hi->zklen); + } else { + o = createStringObject((char*)hi->zv,hi->zvlen); + } + } else { + if (what & REDIS_HASH_KEY) { + o = dictGetEntryKey(hi->de); + } else { + o = dictGetEntryVal(hi->de); + } + incrRefCount(o); + } + return o; +} + +static robj *hashLookupWriteOrCreate(redisClient *c, robj *key) { + robj *o = lookupKeyWrite(c->db,key); + if (o == NULL) { + o = createHashObject(); + dictAdd(c->db->dict,key,o); + incrRefCount(key); + } else { + if (o->type != REDIS_HASH) { + addReply(c,shared.wrongtypeerr); + return NULL; + } + } + return o; +} + +/* ============================= Hash commands ============================== */ +static void hsetCommand(redisClient *c) { + int update; + robj *o; + + if ((o = hashLookupWriteOrCreate(c,c->argv[1])) == NULL) return; + hashTryConversion(o,c->argv,2,3); + hashTryObjectEncoding(o,&c->argv[2], &c->argv[3]); + update = hashSet(o,c->argv[2],c->argv[3]); + addReply(c, update ? shared.czero : shared.cone); + server.dirty++; +} + +static void hsetnxCommand(redisClient *c) { + robj *o; + if ((o = hashLookupWriteOrCreate(c,c->argv[1])) == NULL) return; + hashTryConversion(o,c->argv,2,3); + + if (hashExists(o, c->argv[2])) { + addReply(c, shared.czero); + } else { + hashTryObjectEncoding(o,&c->argv[2], &c->argv[3]); + hashSet(o,c->argv[2],c->argv[3]); + addReply(c, shared.cone); + server.dirty++; + } +} + +static void hmsetCommand(redisClient *c) { + int i; + robj *o; + + if ((c->argc % 2) == 1) { + addReplySds(c,sdsnew("-ERR wrong number of arguments for HMSET\r\n")); + return; + } + + if ((o = hashLookupWriteOrCreate(c,c->argv[1])) == NULL) return; + hashTryConversion(o,c->argv,2,c->argc-1); + for (i = 2; i < c->argc; i += 2) { + hashTryObjectEncoding(o,&c->argv[i], &c->argv[i+1]); + hashSet(o,c->argv[i],c->argv[i+1]); + } + addReply(c, shared.ok); + server.dirty++; +} + +static void hincrbyCommand(redisClient *c) { + long long value, incr; + robj *o, *current, *new; + + if (getLongLongFromObjectOrReply(c,c->argv[3],&incr,NULL) != REDIS_OK) return; + if ((o = hashLookupWriteOrCreate(c,c->argv[1])) == NULL) return; + if ((current = hashGet(o,c->argv[2])) != NULL) { + if (current->encoding == REDIS_ENCODING_RAW) + value = strtoll(current->ptr,NULL,10); + else if (current->encoding == REDIS_ENCODING_INT) + value = (long)current->ptr; + else + redisAssert(1 != 1); + decrRefCount(current); + } else { + value = 0; + } + + value += incr; + new = createStringObjectFromLongLong(value); + hashTryObjectEncoding(o,&c->argv[2],NULL); + hashSet(o,c->argv[2],new); + decrRefCount(new); + addReplyLongLong(c,value); + server.dirty++; +} + +static void hgetCommand(redisClient *c) { + robj *o, *value; + if ((o = lookupKeyReadOrReply(c,c->argv[1],shared.nullbulk)) == NULL || + checkType(c,o,REDIS_HASH)) return; + + if ((value = hashGet(o,c->argv[2])) != NULL) { + addReplyBulk(c,value); + decrRefCount(value); + } else { + addReply(c,shared.nullbulk); + } +} + +static void hmgetCommand(redisClient *c) { + int i; + robj *o, *value; + o = lookupKeyRead(c->db,c->argv[1]); + if (o != NULL && o->type != REDIS_HASH) { + addReply(c,shared.wrongtypeerr); + } + + /* Note the check for o != NULL happens inside the loop. This is + * done because objects that cannot be found are considered to be + * an empty hash. The reply should then be a series of NULLs. */ + addReplySds(c,sdscatprintf(sdsempty(),"*%d\r\n",c->argc-2)); + for (i = 2; i < c->argc; i++) { + if (o != NULL && (value = hashGet(o,c->argv[i])) != NULL) { + addReplyBulk(c,value); + decrRefCount(value); + } else { + addReply(c,shared.nullbulk); + } + } +} + +static void hdelCommand(redisClient *c) { + robj *o; + if ((o = lookupKeyWriteOrReply(c,c->argv[1],shared.czero)) == NULL || + checkType(c,o,REDIS_HASH)) return; + + if (hashDelete(o,c->argv[2])) { + if (hashLength(o) == 0) deleteKey(c->db,c->argv[1]); + addReply(c,shared.cone); + server.dirty++; + } else { + addReply(c,shared.czero); + } +} + +static void hlenCommand(redisClient *c) { + robj *o; + if ((o = lookupKeyReadOrReply(c,c->argv[1],shared.czero)) == NULL || + checkType(c,o,REDIS_HASH)) return; + + addReplyUlong(c,hashLength(o)); +} + +static void genericHgetallCommand(redisClient *c, int flags) { + robj *o, *lenobj, *obj; + unsigned long count = 0; + hashIterator *hi; + + if ((o = lookupKeyReadOrReply(c,c->argv[1],shared.emptymultibulk)) == NULL + || checkType(c,o,REDIS_HASH)) return; + + lenobj = createObject(REDIS_STRING,NULL); + addReply(c,lenobj); + decrRefCount(lenobj); + + hi = hashInitIterator(o); + while (hashNext(hi) != REDIS_ERR) { + if (flags & REDIS_HASH_KEY) { + obj = hashCurrent(hi,REDIS_HASH_KEY); + addReplyBulk(c,obj); + decrRefCount(obj); + count++; + } + if (flags & REDIS_HASH_VALUE) { + obj = hashCurrent(hi,REDIS_HASH_VALUE); + addReplyBulk(c,obj); + decrRefCount(obj); + count++; + } + } + hashReleaseIterator(hi); + + lenobj->ptr = sdscatprintf(sdsempty(),"*%lu\r\n",count); +} + +static void hkeysCommand(redisClient *c) { + genericHgetallCommand(c,REDIS_HASH_KEY); +} + +static void hvalsCommand(redisClient *c) { + genericHgetallCommand(c,REDIS_HASH_VALUE); +} + +static void hgetallCommand(redisClient *c) { + genericHgetallCommand(c,REDIS_HASH_KEY|REDIS_HASH_VALUE); +} + +static void hexistsCommand(redisClient *c) { + robj *o; + if ((o = lookupKeyReadOrReply(c,c->argv[1],shared.czero)) == NULL || + checkType(c,o,REDIS_HASH)) return; + + addReply(c, hashExists(o,c->argv[2]) ? shared.cone : shared.czero); +} + +static void convertToRealHash(robj *o) { + unsigned char *key, *val, *p, *zm = o->ptr; + unsigned int klen, vlen; + dict *dict = dictCreate(&hashDictType,NULL); + + assert(o->type == REDIS_HASH && o->encoding != REDIS_ENCODING_HT); + p = zipmapRewind(zm); + while((p = zipmapNext(p,&key,&klen,&val,&vlen)) != NULL) { + robj *keyobj, *valobj; + + keyobj = createStringObject((char*)key,klen); + valobj = createStringObject((char*)val,vlen); + keyobj = tryObjectEncoding(keyobj); + valobj = tryObjectEncoding(valobj); + dictAdd(dict,keyobj,valobj); + } + o->encoding = REDIS_ENCODING_HT; + o->ptr = dict; + zfree(zm); +} + +/* ========================= Non type-specific commands ==================== */ + +static void flushdbCommand(redisClient *c) { + server.dirty += dictSize(c->db->dict); + dictEmpty(c->db->dict); + dictEmpty(c->db->expires); + addReply(c,shared.ok); +} + +static void flushallCommand(redisClient *c) { + server.dirty += emptyDb(); + addReply(c,shared.ok); + if (server.bgsavechildpid != -1) { + kill(server.bgsavechildpid,SIGKILL); + rdbRemoveTempFile(server.bgsavechildpid); + } + rdbSave(server.dbfilename); + server.dirty++; +} + +static redisSortOperation *createSortOperation(int type, robj *pattern) { redisSortOperation *so = zmalloc(sizeof(*so)); so->type = type; so->pattern = pattern; @@ -4834,23 +6515,26 @@ static redisSortOperation *createSortOperation(int type, robj *pattern) { } /* Return the value associated to the key with a name obtained - * substituting the first occurence of '*' in 'pattern' with 'subst' */ + * substituting the first occurence of '*' in 'pattern' with 'subst'. + * The returned object will always have its refcount increased by 1 + * when it is non-NULL. */ static robj *lookupKeyByPattern(redisDb *db, robj *pattern, robj *subst) { - char *p; + char *p, *f; sds spat, ssub; - robj keyobj; - int prefixlen, sublen, postfixlen; + robj keyobj, fieldobj, *o; + int prefixlen, sublen, postfixlen, fieldlen; /* Expoit the internal sds representation to create a sds string allocated on the stack in order to make this function faster */ struct { long len; long free; char buf[REDIS_SORTKEY_MAX+1]; - } keyname; + } keyname, fieldname; /* If the pattern is "#" return the substitution object itself in order * to implement the "SORT ... GET #" feature. */ spat = pattern->ptr; if (spat[0] == '#' && spat[1] == '\0') { + incrRefCount(subst); return subst; } @@ -4867,20 +6551,47 @@ static robj *lookupKeyByPattern(redisDb *db, robj *pattern, robj *subst) { return NULL; } + /* Find out if we're dealing with a hash dereference. */ + if ((f = strstr(p+1, "->")) != NULL) { + fieldlen = sdslen(spat)-(f-spat); + /* this also copies \0 character */ + memcpy(fieldname.buf,f+2,fieldlen-1); + fieldname.len = fieldlen-2; + } else { + fieldlen = 0; + } + prefixlen = p-spat; sublen = sdslen(ssub); - postfixlen = sdslen(spat)-(prefixlen+1); + postfixlen = sdslen(spat)-(prefixlen+1)-fieldlen; memcpy(keyname.buf,spat,prefixlen); memcpy(keyname.buf+prefixlen,ssub,sublen); memcpy(keyname.buf+prefixlen+sublen,p+1,postfixlen); keyname.buf[prefixlen+sublen+postfixlen] = '\0'; keyname.len = prefixlen+sublen+postfixlen; - - initStaticStringObject(keyobj,((char*)&keyname)+(sizeof(long)*2)) decrRefCount(subst); - /* printf("lookup '%s' => %p\n", keyname.buf,de); */ - return lookupKeyRead(db,&keyobj); + /* Lookup substituted key */ + initStaticStringObject(keyobj,((char*)&keyname)+(sizeof(long)*2)); + o = lookupKeyRead(db,&keyobj); + if (o == NULL) return NULL; + + if (fieldlen > 0) { + if (o->type != REDIS_HASH || fieldname.len < 1) return NULL; + + /* Retrieve value from hash by the field name. This operation + * already increases the refcount of the returned object. */ + initStaticStringObject(fieldobj,((char*)&fieldname)+(sizeof(long)*2)); + o = hashGet(o, &fieldobj); + } else { + if (o->type != REDIS_STRING) return NULL; + + /* Every object that this function returns needs to have its refcount + * increased. sortCommand decreases it again. */ + incrRefCount(o); + } + + return o; } /* sortCompare() is used by qsort in sortCommand(). Given that qsort_r with @@ -4915,14 +6626,9 @@ static int sortCompare(const void *s1, const void *s2) { cmp = strcoll(so1->u.cmpobj->ptr,so2->u.cmpobj->ptr); } } else { - /* Compare elements directly */ - robj *dec1, *dec2; - - dec1 = getDecodedObject(so1->obj); - dec2 = getDecodedObject(so2->obj); - cmp = strcoll(dec1->ptr,dec2->ptr); - decrRefCount(dec1); - decrRefCount(dec2); + /* Compare elements directly. Note that these objects already + * need to be non-encoded (see sortCommand). */ + cmp = strcoll(so1->obj->ptr,so2->obj->ptr); } } return server.sort_desc ? -cmp : cmp; @@ -4943,7 +6649,7 @@ static void sortCommand(redisClient *c) { /* Lookup the key to sort. It must be of the right types */ sortval = lookupKeyRead(c->db,c->argv[1]); if (sortval == NULL) { - addReply(c,shared.nullmultibulk); + addReply(c,shared.emptymultibulk); return; } if (sortval->type != REDIS_SET && sortval->type != REDIS_LIST && @@ -5005,7 +6711,7 @@ static void sortCommand(redisClient *c) { case REDIS_LIST: vectorlen = listLength((list*)sortval->ptr); break; case REDIS_SET: vectorlen = dictSize((dict*)sortval->ptr); break; case REDIS_ZSET: vectorlen = dictSize(((zset*)sortval->ptr)->dict); break; - default: vectorlen = 0; redisAssert(0); /* Avoid GCC warning */ + default: vectorlen = 0; redisPanic("Bad SORT type"); /* Avoid GCC warning */ } vector = zmalloc(sizeof(redisSortObject)*vectorlen); j = 0; @@ -5013,9 +6719,10 @@ static void sortCommand(redisClient *c) { if (sortval->type == REDIS_LIST) { list *list = sortval->ptr; listNode *ln; + listIter li; - listRewind(list); - while((ln = listYield(list))) { + listRewind(list,&li); + while((ln = listNext(&li))) { robj *ele = ln->value; vector[j].obj = ele; vector[j].u.score = 0; @@ -5048,38 +6755,36 @@ static void sortCommand(redisClient *c) { /* Now it's time to load the right scores in the sorting vector */ if (dontsort == 0) { for (j = 0; j < vectorlen; j++) { + robj *byval; if (sortby) { - robj *byval; - + /* lookup value to sort by */ byval = lookupKeyByPattern(c->db,sortby,vector[j].obj); - if (!byval || byval->type != REDIS_STRING) continue; - if (alpha) { - vector[j].u.cmpobj = getDecodedObject(byval); - } else { - if (byval->encoding == REDIS_ENCODING_RAW) { - vector[j].u.score = strtod(byval->ptr,NULL); - } else { - /* Don't need to decode the object if it's - * integer-encoded (the only encoding supported) so - * far. We can just cast it */ - if (byval->encoding == REDIS_ENCODING_INT) { - vector[j].u.score = (long)byval->ptr; - } else - redisAssert(1 != 1); - } - } + if (!byval) continue; } else { - if (!alpha) { - if (vector[j].obj->encoding == REDIS_ENCODING_RAW) - vector[j].u.score = strtod(vector[j].obj->ptr,NULL); - else { - if (vector[j].obj->encoding == REDIS_ENCODING_INT) - vector[j].u.score = (long) vector[j].obj->ptr; - else - redisAssert(1 != 1); - } + /* use object itself to sort by */ + byval = vector[j].obj; + } + + if (alpha) { + vector[j].u.cmpobj = getDecodedObject(byval); + } else { + if (byval->encoding == REDIS_ENCODING_RAW) { + vector[j].u.score = strtod(byval->ptr,NULL); + } else if (byval->encoding == REDIS_ENCODING_INT) { + /* Don't need to decode the object if it's + * integer-encoded (the only encoding supported) so + * far. We can just cast it */ + vector[j].u.score = (long)byval->ptr; + } else { + redisAssert(1 != 1); } } + + /* when the object was retrieved using lookupKeyByPattern, + * its refcount needs to be decreased. */ + if (sortby) { + decrRefCount(byval); + } } } @@ -5111,24 +6816,21 @@ static void sortCommand(redisClient *c) { addReplySds(c,sdscatprintf(sdsempty(),"*%d\r\n",outputlen)); for (j = start; j <= end; j++) { listNode *ln; - if (!getop) { - addReplyBulkLen(c,vector[j].obj); - addReply(c,vector[j].obj); - addReply(c,shared.crlf); - } - listRewind(operations); - while((ln = listYield(operations))) { + listIter li; + + if (!getop) addReplyBulk(c,vector[j].obj); + listRewind(operations,&li); + while((ln = listNext(&li))) { redisSortOperation *sop = ln->value; robj *val = lookupKeyByPattern(c->db,sop->pattern, vector[j].obj); if (sop->type == REDIS_SORT_GET) { - if (!val || val->type != REDIS_STRING) { + if (!val) { addReply(c,shared.nullbulk); } else { - addReplyBulkLen(c,val); - addReply(c,val); - addReply(c,shared.crlf); + addReplyBulk(c,val); + decrRefCount(val); } } else { redisAssert(sop->type == REDIS_SORT_GET); /* always fails */ @@ -5142,22 +6844,27 @@ static void sortCommand(redisClient *c) { /* STORE option specified, set the sorting result as a List object */ for (j = start; j <= end; j++) { listNode *ln; + listIter li; + if (!getop) { listAddNodeTail(listPtr,vector[j].obj); incrRefCount(vector[j].obj); } - listRewind(operations); - while((ln = listYield(operations))) { + listRewind(operations,&li); + while((ln = listNext(&li))) { redisSortOperation *sop = ln->value; robj *val = lookupKeyByPattern(c->db,sop->pattern, vector[j].obj); if (sop->type == REDIS_SORT_GET) { - if (!val || val->type != REDIS_STRING) { + if (!val) { listAddNodeTail(listPtr,createStringObject("",0)); } else { + /* We should do a incrRefCount on val because it is + * added to the list, but also a decrRefCount because + * it is returned by lookupKeyByPattern. This results + * in doing nothing at all. */ listAddNodeTail(listPtr,val); - incrRefCount(val); } } else { redisAssert(sop->type == REDIS_SORT_GET); /* always fails */ @@ -5178,12 +6885,33 @@ static void sortCommand(redisClient *c) { decrRefCount(sortval); listRelease(operations); for (j = 0; j < vectorlen; j++) { - if (sortby && alpha && vector[j].u.cmpobj) + if (alpha && vector[j].u.cmpobj) decrRefCount(vector[j].u.cmpobj); } zfree(vector); } +/* Convert an amount of bytes into a human readable string in the form + * of 100B, 2G, 100M, 4K, and so forth. */ +static void bytesToHuman(char *s, unsigned long long n) { + double d; + + if (n < 1024) { + /* Bytes */ + sprintf(s,"%lluB",n); + return; + } else if (n < (1024*1024)) { + d = (double)n/(1024); + sprintf(s,"%.2fK",d); + } else if (n < (1024LL*1024*1024)) { + d = (double)n/(1024*1024); + sprintf(s,"%.2fM",d); + } else if (n < (1024LL*1024*1024*1024)) { + d = (double)n/(1024LL*1024*1024); + sprintf(s,"%.2fG",d); + } +} + /* Create the string returned by the INFO command. This is decoupled * by the INFO command itself as we need to report the same information * on memory corruption problems. */ @@ -5191,37 +6919,57 @@ static sds genRedisInfoString(void) { sds info; time_t uptime = time(NULL)-server.stat_starttime; int j; - + char hmem[64]; + + bytesToHuman(hmem,zmalloc_used_memory()); info = sdscatprintf(sdsempty(), "redis_version:%s\r\n" "arch_bits:%s\r\n" "multiplexing_api:%s\r\n" + "process_id:%ld\r\n" "uptime_in_seconds:%ld\r\n" "uptime_in_days:%ld\r\n" "connected_clients:%d\r\n" "connected_slaves:%d\r\n" + "blocked_clients:%d\r\n" "used_memory:%zu\r\n" + "used_memory_human:%s\r\n" "changes_since_last_save:%lld\r\n" "bgsave_in_progress:%d\r\n" "last_save_time:%ld\r\n" "bgrewriteaof_in_progress:%d\r\n" "total_connections_received:%lld\r\n" "total_commands_processed:%lld\r\n" + "expired_keys:%lld\r\n" + "hash_max_zipmap_entries:%ld\r\n" + "hash_max_zipmap_value:%ld\r\n" + "pubsub_channels:%ld\r\n" + "pubsub_patterns:%u\r\n" + "vm_enabled:%d\r\n" "role:%s\r\n" ,REDIS_VERSION, (sizeof(long) == 8) ? "64" : "32", aeGetApiName(), + (long) getpid(), uptime, uptime/(3600*24), listLength(server.clients)-listLength(server.slaves), listLength(server.slaves), - server.usedmemory, + server.blpop_blocked_clients, + zmalloc_used_memory(), + hmem, server.dirty, server.bgsavechildpid != -1, server.lastsave, server.bgrewritechildpid != -1, server.stat_numconnections, server.stat_numcommands, + server.stat_expiredkeys, + server.hash_max_zipmap_entries, + server.hash_max_zipmap_value, + dictSize(server.pubsub_channels), + listLength(server.pubsub_patterns), + server.vm_enabled != 0, server.masterhost == NULL ? "master" : "slave" ); if (server.masterhost) { @@ -5237,6 +6985,36 @@ static sds genRedisInfoString(void) { server.master ? ((int)(time(NULL)-server.master->lastinteraction)) : -1 ); } + if (server.vm_enabled) { + lockThreadedIO(); + info = sdscatprintf(info, + "vm_conf_max_memory:%llu\r\n" + "vm_conf_page_size:%llu\r\n" + "vm_conf_pages:%llu\r\n" + "vm_stats_used_pages:%llu\r\n" + "vm_stats_swapped_objects:%llu\r\n" + "vm_stats_swappin_count:%llu\r\n" + "vm_stats_swappout_count:%llu\r\n" + "vm_stats_io_newjobs_len:%lu\r\n" + "vm_stats_io_processing_len:%lu\r\n" + "vm_stats_io_processed_len:%lu\r\n" + "vm_stats_io_active_threads:%lu\r\n" + "vm_stats_blocked_clients:%lu\r\n" + ,(unsigned long long) server.vm_max_memory, + (unsigned long long) server.vm_page_size, + (unsigned long long) server.vm_pages, + (unsigned long long) server.vm_stats_used_pages, + (unsigned long long) server.vm_stats_swapped_objects, + (unsigned long long) server.vm_stats_swapins, + (unsigned long long) server.vm_stats_swapouts, + (unsigned long) listLength(server.io_newjobs), + (unsigned long) listLength(server.io_processing), + (unsigned long) listLength(server.io_processed), + (unsigned long) server.io_active_threads, + (unsigned long) server.vm_blocked_clients + ); + unlockThreadedIO(); + } for (j = 0; j < server.dbnum; j++) { long long keys, vkeys; @@ -5312,6 +7090,7 @@ static int expireIfNeeded(redisDb *db, robj *key) { /* Delete the key */ dictDelete(db->expires,key); + server.stat_expiredkeys++; return dictDelete(db->dict,key) == DICT_OK; } @@ -5324,19 +7103,25 @@ static int deleteIfVolatile(redisDb *db, robj *key) { /* Delete the key */ server.dirty++; + server.stat_expiredkeys++; dictDelete(db->expires,key); return dictDelete(db->dict,key) == DICT_OK; } -static void expireGenericCommand(redisClient *c, robj *key, time_t seconds) { +static void expireGenericCommand(redisClient *c, robj *key, robj *param, long offset) { dictEntry *de; + time_t seconds; + + if (getLongFromObjectOrReply(c, param, &seconds, NULL) != REDIS_OK) return; + + seconds -= offset; de = dictFind(c->db->dict,key); if (de == NULL) { addReply(c,shared.czero); return; } - if (seconds < 0) { + if (seconds <= 0) { if (deleteKey(c->db,key)) server.dirty++; addReply(c, shared.cone); return; @@ -5353,11 +7138,11 @@ static void expireGenericCommand(redisClient *c, robj *key, time_t seconds) { } static void expireCommand(redisClient *c) { - expireGenericCommand(c,c->argv[1],strtol(c->argv[2]->ptr,NULL,10)); + expireGenericCommand(c,c->argv[1],c->argv[2],0); } static void expireatCommand(redisClient *c) { - expireGenericCommand(c,c->argv[1],strtol(c->argv[2]->ptr,NULL,10)-time(NULL)); + expireGenericCommand(c,c->argv[1],c->argv[2],time(NULL)); } static void ttlCommand(redisClient *c) { @@ -5417,6 +7202,18 @@ static void multiCommand(redisClient *c) { addReply(c,shared.ok); } +static void discardCommand(redisClient *c) { + if (!(c->flags & REDIS_MULTI)) { + addReplySds(c,sdsnew("-ERR DISCARD without MULTI\r\n")); + return; + } + + freeClientMultiState(c); + initClientMultiState(c); + c->flags &= (~REDIS_MULTI); + addReply(c,shared.ok); +} + static void execCommand(redisClient *c) { int j; robj **orig_argv; @@ -5462,7 +7259,7 @@ static void execCommand(redisClient *c) { * empty we need to block. In order to do so we remove the notification for * new data to read in the client socket (so that we'll not serve new * requests if the blocking request is not served). Also we put the client - * in a dictionary (server.blockingkeys) mapping keys to a list of clients + * in a dictionary (db->blockingkeys) mapping keys to a list of clients * blocking for this keys. * - If a PUSH operation against a key with blocked clients waiting is * performed, we serve the first in the list: basically instead to push @@ -5475,57 +7272,68 @@ static void execCommand(redisClient *c) { /* Set a client in blocking mode for the specified key, with the specified * timeout */ -static void blockForKey(redisClient *c, robj *key, time_t timeout) { +static void blockForKeys(redisClient *c, robj **keys, int numkeys, time_t timeout) { dictEntry *de; list *l; + int j; - c->blockingkey = key; - incrRefCount(key); + c->blockingkeys = zmalloc(sizeof(robj*)*numkeys); + c->blockingkeysnum = numkeys; c->blockingto = timeout; - de = dictFind(c->db->blockingkeys,key); - if (de == NULL) { - int retval; + for (j = 0; j < numkeys; j++) { + /* Add the key in the client structure, to map clients -> keys */ + c->blockingkeys[j] = keys[j]; + incrRefCount(keys[j]); - l = listCreate(); - retval = dictAdd(c->db->blockingkeys,key,l); - assert(retval == DICT_OK); - } else { - l = dictGetEntryVal(de); + /* And in the other "side", to map keys -> clients */ + de = dictFind(c->db->blockingkeys,keys[j]); + if (de == NULL) { + int retval; + + /* For every key we take a list of clients blocked for it */ + l = listCreate(); + retval = dictAdd(c->db->blockingkeys,keys[j],l); + incrRefCount(keys[j]); + assert(retval == DICT_OK); + } else { + l = dictGetEntryVal(de); + } + listAddNodeTail(l,c); } - listAddNodeTail(l,c); + /* Mark the client as a blocked client */ c->flags |= REDIS_BLOCKED; - aeDeleteFileEvent(server.el,c->fd,AE_READABLE); + server.blpop_blocked_clients++; } /* Unblock a client that's waiting in a blocking operation such as BLPOP */ -static void unblockClient(redisClient *c) { +static void unblockClientWaitingData(redisClient *c) { dictEntry *de; list *l; + int j; - /* Remove this client from the list of clients waiting for this key. */ - assert(c->blockingkey != NULL); - de = dictFind(c->db->blockingkeys,c->blockingkey); - assert(de != NULL); - l = dictGetEntryVal(de); - listDelNode(l,listSearchKey(l,c)); - /* If the list is empty we need to remove it to avoid wasting memory */ - if (listLength(l) == 0) - dictDelete(c->db->blockingkeys,c->blockingkey); - /* Finally set the right flags in the client structure */ - decrRefCount(c->blockingkey); - c->blockingkey = NULL; + assert(c->blockingkeys != NULL); + /* The client may wait for multiple keys, so unblock it for every key. */ + for (j = 0; j < c->blockingkeysnum; j++) { + /* Remove this client from the list of clients waiting for this key. */ + de = dictFind(c->db->blockingkeys,c->blockingkeys[j]); + assert(de != NULL); + l = dictGetEntryVal(de); + listDelNode(l,listSearchKey(l,c)); + /* If the list is empty we need to remove it to avoid wasting memory */ + if (listLength(l) == 0) + dictDelete(c->db->blockingkeys,c->blockingkeys[j]); + decrRefCount(c->blockingkeys[j]); + } + /* Cleanup the client structure */ + zfree(c->blockingkeys); + c->blockingkeys = NULL; c->flags &= (~REDIS_BLOCKED); - /* Ok now we are ready to get read events from socket, note that we - * can't trap errors here as it's possible that unblockClients() is - * called from freeClient() itself, and the only thing we can do - * if we failed to register the READABLE event is to kill the client. - * Still the following function should never fail in the real world as - * we are sure the file descriptor is sane, and we exit on out of mem. */ - aeCreateFileEvent(server.el, c->fd, AE_READABLE, readQueryFromClient, c); - /* As a final step we want to process data if there is some command waiting - * in the input buffer. Note that this is safe even if unblockClient() - * gets called from freeClient() because freeClient() will be smart - * enough to call this function *after* c->querybuf was set to NULL. */ + server.blpop_blocked_clients--; + /* We want to process data if there is some command waiting + * in the input buffer. Note that this is safe even if + * unblockClientWaitingData() gets called from freeClient() because + * freeClient() will be smart enough to call this function + * *after* c->querybuf was set to NULL. */ if (c->querybuf && sdslen(c->querybuf) > 0) processInputBuffer(c); } @@ -5551,14 +7359,11 @@ static int handleClientsWaitingListPush(redisClient *c, robj *key, robj *ele) { ln = listFirst(l); assert(ln != NULL); receiver = ln->value; - listDelNode(l,ln); - if (listLength(l) == 0) - dictDelete(c->db->blockingkeys,key); - addReplyBulkLen(receiver,ele); - addReply(receiver,ele); - addReply(receiver,shared.crlf); - unblockClient(receiver); + addReplySds(receiver,sdsnew("*2\r\n")); + addReplyBulk(receiver,key); + addReplyBulk(receiver,ele); + unblockClientWaitingData(receiver); return 1; } @@ -5566,26 +7371,51 @@ static int handleClientsWaitingListPush(redisClient *c, robj *key, robj *ele) { static void blockingPopGenericCommand(redisClient *c, int where) { robj *o; time_t timeout; + int j; - o = lookupKeyWrite(c->db,c->argv[1]); - if (o != NULL) { - if (o->type != REDIS_LIST) { - popGenericCommand(c,where); - return; - } else { - list *list = o->ptr; - if (listLength(list) != 0) { - /* If the list contains elements fall back to the usual - * non-blocking POP operation */ - popGenericCommand(c,where); + for (j = 1; j < c->argc-1; j++) { + o = lookupKeyWrite(c->db,c->argv[j]); + if (o != NULL) { + if (o->type != REDIS_LIST) { + addReply(c,shared.wrongtypeerr); return; + } else { + list *list = o->ptr; + if (listLength(list) != 0) { + /* If the list contains elements fall back to the usual + * non-blocking POP operation */ + robj *argv[2], **orig_argv; + int orig_argc; + + /* We need to alter the command arguments before to call + * popGenericCommand() as the command takes a single key. */ + orig_argv = c->argv; + orig_argc = c->argc; + argv[1] = c->argv[j]; + c->argv = argv; + c->argc = 2; + + /* Also the return value is different, we need to output + * the multi bulk reply header and the key name. The + * "real" command will add the last element (the value) + * for us. If this souds like an hack to you it's just + * because it is... */ + addReplySds(c,sdsnew("*2\r\n")); + addReplyBulk(c,argv[1]); + popGenericCommand(c,where); + + /* Fix the client structure with the original stuff */ + c->argv = orig_argv; + c->argc = orig_argc; + return; + } } } } /* If the list is empty or the key does not exists we must block */ - timeout = strtol(c->argv[2]->ptr,NULL,10); + timeout = strtol(c->argv[c->argc-1]->ptr,NULL,10); if (timeout > 0) timeout += time(NULL); - blockForKey(c,c->argv[1],timeout); + blockForKeys(c,c->argv+1,c->argc-2,timeout); } static void blpopCommand(redisClient *c) { @@ -5682,9 +7512,10 @@ static void syncCommand(redisClient *c) { * registering differences since the server forked to save */ redisClient *slave; listNode *ln; + listIter li; - listRewind(server.slaves); - while((ln = listYield(server.slaves))) { + listRewind(server.slaves,&li); + while((ln = listNext(&li))) { slave = ln->value; if (slave->replstate == REDIS_REPL_WAIT_BGSAVE_END) break; } @@ -5751,7 +7582,7 @@ static void sendBulkToSlave(aeEventLoop *el, int fd, void *privdata, int mask) { return; } if ((nwritten = write(fd,buf,buflen)) == -1) { - redisLog(REDIS_DEBUG,"Write error sending DB to slave: %s", + redisLog(REDIS_VERBOSE,"Write error sending DB to slave: %s", strerror(errno)); freeClient(slave); return; @@ -5781,9 +7612,10 @@ static void sendBulkToSlave(aeEventLoop *el, int fd, void *privdata, int mask) { static void updateSlavesWaitingBgsave(int bgsaveerr) { listNode *ln; int startbgsave = 0; + listIter li; - listRewind(server.slaves); - while((ln = listYield(server.slaves))) { + listRewind(server.slaves,&li); + while((ln = listNext(&li))) { redisClient *slave = ln->value; if (slave->replstate == REDIS_REPL_WAIT_BGSAVE_START) { @@ -5791,7 +7623,7 @@ static void updateSlavesWaitingBgsave(int bgsaveerr) { slave->replstate = REDIS_REPL_WAIT_BGSAVE_END; } else if (slave->replstate == REDIS_REPL_WAIT_BGSAVE_END) { struct redis_stat buf; - + if (bgsaveerr != REDIS_OK) { freeClient(slave); redisLog(REDIS_WARNING,"SYNC failed. BGSAVE child returned an error"); @@ -5815,9 +7647,11 @@ static void updateSlavesWaitingBgsave(int bgsaveerr) { } if (startbgsave) { if (rdbSaveBackground(server.dbfilename) != REDIS_OK) { - listRewind(server.slaves); + listIter li; + + listRewind(server.slaves,&li); redisLog(REDIS_WARNING,"SYNC failed. BGSAVE failed"); - while((ln = listYield(server.slaves))) { + while((ln = listNext(&li))) { redisClient *slave = ln->value; if (slave->replstate == REDIS_REPL_WAIT_BGSAVE_START) @@ -5829,9 +7663,9 @@ static void updateSlavesWaitingBgsave(int bgsaveerr) { static int syncWithMaster(void) { char buf[1024], tmpfile[256], authcmd[1024]; - int dumpsize; + long dumpsize; int fd = anetTcpConnect(NULL,server.masterhost,server.masterport); - int dfd; + int dfd, maxtries = 5; if (fd == -1) { redisLog(REDIS_WARNING,"Unable to connect to MASTER: %s", @@ -5881,11 +7715,16 @@ static int syncWithMaster(void) { redisLog(REDIS_WARNING,"Bad protocol from MASTER, the first byte is not '$', are you sure the host and port are right?"); return REDIS_ERR; } - dumpsize = atoi(buf+1); - redisLog(REDIS_NOTICE,"Receiving %d bytes data dump from MASTER",dumpsize); + dumpsize = strtol(buf+1,NULL,10); + redisLog(REDIS_NOTICE,"Receiving %ld bytes data dump from MASTER",dumpsize); /* Read the bulk write data on a temp file */ - snprintf(tmpfile,256,"temp-%d.%ld.rdb",(int)time(NULL),(long int)random()); - dfd = open(tmpfile,O_CREAT|O_WRONLY,0644); + while(maxtries--) { + snprintf(tmpfile,256, + "temp-%d.%ld.rdb",(int)time(NULL),(long int)getpid()); + dfd = open(tmpfile,O_CREAT|O_WRONLY|O_EXCL,0644); + if (dfd != -1) break; + sleep(1); + } if (dfd == -1) { close(fd); redisLog(REDIS_WARNING,"Opening the temp file needed for MASTER <-> SLAVE synchronization: %s",strerror(errno)); @@ -5955,6 +7794,27 @@ static void slaveofCommand(redisClient *c) { /* ============================ Maxmemory directive ======================== */ +/* Try to free one object form the pre-allocated objects free list. + * This is useful under low mem conditions as by default we take 1 million + * free objects allocated. On success REDIS_OK is returned, otherwise + * REDIS_ERR. */ +static int tryFreeOneObjectFromFreelist(void) { + robj *o; + + if (server.vm_enabled) pthread_mutex_lock(&server.obj_freelist_mutex); + if (listLength(server.objfreelist)) { + listNode *head = listFirst(server.objfreelist); + o = listNodeValue(head); + listDelNode(server.objfreelist,head); + if (server.vm_enabled) pthread_mutex_unlock(&server.obj_freelist_mutex); + zfree(o); + return REDIS_OK; + } else { + if (server.vm_enabled) pthread_mutex_unlock(&server.obj_freelist_mutex); + return REDIS_ERR; + } +} + /* This function gets called when 'maxmemory' is set on the config file to limit * the max memory used by the server, and we are out of memory. * This function will try to, in order: @@ -5968,40 +7828,32 @@ static void slaveofCommand(redisClient *c) { */ static void freeMemoryIfNeeded(void) { while (server.maxmemory && zmalloc_used_memory() > server.maxmemory) { - if (listLength(server.objfreelist)) { - robj *o; - - listNode *head = listFirst(server.objfreelist); - o = listNodeValue(head); - listDelNode(server.objfreelist,head); - zfree(o); - } else { - int j, k, freed = 0; - - for (j = 0; j < server.dbnum; j++) { - int minttl = -1; - robj *minkey = NULL; - struct dictEntry *de; - - if (dictSize(server.db[j].expires)) { - freed = 1; - /* From a sample of three keys drop the one nearest to - * the natural expire */ - for (k = 0; k < 3; k++) { - time_t t; - - de = dictGetRandomKey(server.db[j].expires); - t = (time_t) dictGetEntryVal(de); - if (minttl == -1 || t < minttl) { - minkey = dictGetEntryKey(de); - minttl = t; - } + int j, k, freed = 0; + + if (tryFreeOneObjectFromFreelist() == REDIS_OK) continue; + for (j = 0; j < server.dbnum; j++) { + int minttl = -1; + robj *minkey = NULL; + struct dictEntry *de; + + if (dictSize(server.db[j].expires)) { + freed = 1; + /* From a sample of three keys drop the one nearest to + * the natural expire */ + for (k = 0; k < 3; k++) { + time_t t; + + de = dictGetRandomKey(server.db[j].expires); + t = (time_t) dictGetEntryVal(de); + if (minttl == -1 || t < minttl) { + minkey = dictGetEntryKey(de); + minttl = t; } - deleteKey(server.db+j,minkey); } + deleteKey(server.db+j,minkey); } - if (!freed) return; /* nothing to free... */ } + if (!freed) return; /* nothing to free... */ } } @@ -6125,6 +7977,7 @@ int loadAppendOnlyFile(char *filename) { struct redisClient *fakeClient; FILE *fp = fopen(filename,"r"); struct redis_stat sb; + unsigned long long loadedkeys = 0; if (redis_fstat(fileno(fp),&sb) != -1 && sb.st_size == 0) return REDIS_ERR; @@ -6168,14 +8021,9 @@ int loadAppendOnlyFile(char *filename) { redisLog(REDIS_WARNING,"Unknown command '%s' reading the append only file", argv[0]->ptr); exit(1); } - /* Try object sharing and encoding */ - if (server.shareobjects) { - int j; - for(j = 1; j < argc; j++) - argv[j] = tryObjectSharing(argv[j]); - } + /* Try object encoding */ if (cmd->flags & REDIS_CMD_BULK) - tryObjectEncoding(argv[argc-1]); + argv[argc-1] = tryObjectEncoding(argv[argc-1]); /* Run the command in the context of a fake client */ fakeClient->argc = argc; fakeClient->argv = argv; @@ -6186,6 +8034,13 @@ int loadAppendOnlyFile(char *filename) { /* Clean up, ready for the next command */ for (j = 0; j < argc; j++) decrRefCount(argv[j]); zfree(argv); + /* Handle swapping while loading big datasets when VM is on */ + loadedkeys++; + if (server.vm_enabled && (loadedkeys % 5000) == 0) { + while (zmalloc_used_memory() > server.vm_max_memory) { + if (vmSwapOneObjectBlocking() == REDIS_ERR) break; + } + } } fclose(fp); freeFakeClient(fakeClient); @@ -6204,241 +8059,1747 @@ fmterr: } /* Write an object into a file in the bulk format $\r\n\r\n */ -static int fwriteBulk(FILE *fp, robj *obj) { +static int fwriteBulkObject(FILE *fp, robj *obj) { char buf[128]; - obj = getDecodedObject(obj); + int decrrc = 0; + + /* Avoid the incr/decr ref count business if possible to help + * copy-on-write (we are often in a child process when this function + * is called). + * Also makes sure that key objects don't get incrRefCount-ed when VM + * is enabled */ + if (obj->encoding != REDIS_ENCODING_RAW) { + obj = getDecodedObject(obj); + decrrc = 1; + } snprintf(buf,sizeof(buf),"$%ld\r\n",(long)sdslen(obj->ptr)); if (fwrite(buf,strlen(buf),1,fp) == 0) goto err; if (sdslen(obj->ptr) && fwrite(obj->ptr,sdslen(obj->ptr),1,fp) == 0) goto err; if (fwrite("\r\n",2,1,fp) == 0) goto err; - decrRefCount(obj); + if (decrrc) decrRefCount(obj); return 1; err: - decrRefCount(obj); + if (decrrc) decrRefCount(obj); return 0; } +/* Write binary-safe string into a file in the bulkformat + * $\r\n\r\n */ +static int fwriteBulkString(FILE *fp, char *s, unsigned long len) { + char buf[128]; + + snprintf(buf,sizeof(buf),"$%ld\r\n",(unsigned long)len); + if (fwrite(buf,strlen(buf),1,fp) == 0) return 0; + if (len && fwrite(s,len,1,fp) == 0) return 0; + if (fwrite("\r\n",2,1,fp) == 0) return 0; + return 1; +} + /* Write a double value in bulk format $\r\n\r\n */ static int fwriteBulkDouble(FILE *fp, double d) { char buf[128], dbuf[128]; - snprintf(dbuf,sizeof(dbuf),"%.17g\r\n",d); - snprintf(buf,sizeof(buf),"$%lu\r\n",(unsigned long)strlen(dbuf)-2); - if (fwrite(buf,strlen(buf),1,fp) == 0) return 0; - if (fwrite(dbuf,strlen(dbuf),1,fp) == 0) return 0; - return 1; + snprintf(dbuf,sizeof(dbuf),"%.17g\r\n",d); + snprintf(buf,sizeof(buf),"$%lu\r\n",(unsigned long)strlen(dbuf)-2); + if (fwrite(buf,strlen(buf),1,fp) == 0) return 0; + if (fwrite(dbuf,strlen(dbuf),1,fp) == 0) return 0; + return 1; +} + +/* Write a long value in bulk format $\r\n\r\n */ +static int fwriteBulkLong(FILE *fp, long l) { + char buf[128], lbuf[128]; + + snprintf(lbuf,sizeof(lbuf),"%ld\r\n",l); + snprintf(buf,sizeof(buf),"$%lu\r\n",(unsigned long)strlen(lbuf)-2); + if (fwrite(buf,strlen(buf),1,fp) == 0) return 0; + if (fwrite(lbuf,strlen(lbuf),1,fp) == 0) return 0; + return 1; +} + +/* Write a sequence of commands able to fully rebuild the dataset into + * "filename". Used both by REWRITEAOF and BGREWRITEAOF. */ +static int rewriteAppendOnlyFile(char *filename) { + dictIterator *di = NULL; + dictEntry *de; + FILE *fp; + char tmpfile[256]; + int j; + time_t now = time(NULL); + + /* Note that we have to use a different temp name here compared to the + * one used by rewriteAppendOnlyFileBackground() function. */ + snprintf(tmpfile,256,"temp-rewriteaof-%d.aof", (int) getpid()); + fp = fopen(tmpfile,"w"); + if (!fp) { + redisLog(REDIS_WARNING, "Failed rewriting the append only file: %s", strerror(errno)); + return REDIS_ERR; + } + for (j = 0; j < server.dbnum; j++) { + char selectcmd[] = "*2\r\n$6\r\nSELECT\r\n"; + redisDb *db = server.db+j; + dict *d = db->dict; + if (dictSize(d) == 0) continue; + di = dictGetIterator(d); + if (!di) { + fclose(fp); + return REDIS_ERR; + } + + /* SELECT the new DB */ + if (fwrite(selectcmd,sizeof(selectcmd)-1,1,fp) == 0) goto werr; + if (fwriteBulkLong(fp,j) == 0) goto werr; + + /* Iterate this DB writing every entry */ + while((de = dictNext(di)) != NULL) { + robj *key, *o; + time_t expiretime; + int swapped; + + key = dictGetEntryKey(de); + /* If the value for this key is swapped, load a preview in memory. + * We use a "swapped" flag to remember if we need to free the + * value object instead to just increment the ref count anyway + * in order to avoid copy-on-write of pages if we are forked() */ + if (!server.vm_enabled || key->storage == REDIS_VM_MEMORY || + key->storage == REDIS_VM_SWAPPING) { + o = dictGetEntryVal(de); + swapped = 0; + } else { + o = vmPreviewObject(key); + swapped = 1; + } + expiretime = getExpire(db,key); + + /* Save the key and associated value */ + if (o->type == REDIS_STRING) { + /* Emit a SET command */ + char cmd[]="*3\r\n$3\r\nSET\r\n"; + if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr; + /* Key and value */ + if (fwriteBulkObject(fp,key) == 0) goto werr; + if (fwriteBulkObject(fp,o) == 0) goto werr; + } else if (o->type == REDIS_LIST) { + /* Emit the RPUSHes needed to rebuild the list */ + list *list = o->ptr; + listNode *ln; + listIter li; + + listRewind(list,&li); + while((ln = listNext(&li))) { + char cmd[]="*3\r\n$5\r\nRPUSH\r\n"; + robj *eleobj = listNodeValue(ln); + + if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr; + if (fwriteBulkObject(fp,key) == 0) goto werr; + if (fwriteBulkObject(fp,eleobj) == 0) goto werr; + } + } else if (o->type == REDIS_SET) { + /* Emit the SADDs needed to rebuild the set */ + dict *set = o->ptr; + dictIterator *di = dictGetIterator(set); + dictEntry *de; + + while((de = dictNext(di)) != NULL) { + char cmd[]="*3\r\n$4\r\nSADD\r\n"; + robj *eleobj = dictGetEntryKey(de); + + if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr; + if (fwriteBulkObject(fp,key) == 0) goto werr; + if (fwriteBulkObject(fp,eleobj) == 0) goto werr; + } + dictReleaseIterator(di); + } else if (o->type == REDIS_ZSET) { + /* Emit the ZADDs needed to rebuild the sorted set */ + zset *zs = o->ptr; + dictIterator *di = dictGetIterator(zs->dict); + dictEntry *de; + + while((de = dictNext(di)) != NULL) { + char cmd[]="*4\r\n$4\r\nZADD\r\n"; + robj *eleobj = dictGetEntryKey(de); + double *score = dictGetEntryVal(de); + + if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr; + if (fwriteBulkObject(fp,key) == 0) goto werr; + if (fwriteBulkDouble(fp,*score) == 0) goto werr; + if (fwriteBulkObject(fp,eleobj) == 0) goto werr; + } + dictReleaseIterator(di); + } else if (o->type == REDIS_HASH) { + char cmd[]="*4\r\n$4\r\nHSET\r\n"; + + /* Emit the HSETs needed to rebuild the hash */ + if (o->encoding == REDIS_ENCODING_ZIPMAP) { + unsigned char *p = zipmapRewind(o->ptr); + unsigned char *field, *val; + unsigned int flen, vlen; + + while((p = zipmapNext(p,&field,&flen,&val,&vlen)) != NULL) { + if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr; + if (fwriteBulkObject(fp,key) == 0) goto werr; + if (fwriteBulkString(fp,(char*)field,flen) == -1) + return -1; + if (fwriteBulkString(fp,(char*)val,vlen) == -1) + return -1; + } + } else { + dictIterator *di = dictGetIterator(o->ptr); + dictEntry *de; + + while((de = dictNext(di)) != NULL) { + robj *field = dictGetEntryKey(de); + robj *val = dictGetEntryVal(de); + + if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr; + if (fwriteBulkObject(fp,key) == 0) goto werr; + if (fwriteBulkObject(fp,field) == -1) return -1; + if (fwriteBulkObject(fp,val) == -1) return -1; + } + dictReleaseIterator(di); + } + } else { + redisPanic("Unknown object type"); + } + /* Save the expire time */ + if (expiretime != -1) { + char cmd[]="*3\r\n$8\r\nEXPIREAT\r\n"; + /* If this key is already expired skip it */ + if (expiretime < now) continue; + if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr; + if (fwriteBulkObject(fp,key) == 0) goto werr; + if (fwriteBulkLong(fp,expiretime) == 0) goto werr; + } + if (swapped) decrRefCount(o); + } + dictReleaseIterator(di); + } + + /* Make sure data will not remain on the OS's output buffers */ + fflush(fp); + fsync(fileno(fp)); + fclose(fp); + + /* Use RENAME to make sure the DB file is changed atomically only + * if the generate DB file is ok. */ + if (rename(tmpfile,filename) == -1) { + redisLog(REDIS_WARNING,"Error moving temp append only file on the final destination: %s", strerror(errno)); + unlink(tmpfile); + return REDIS_ERR; + } + redisLog(REDIS_NOTICE,"SYNC append only file rewrite performed"); + return REDIS_OK; + +werr: + fclose(fp); + unlink(tmpfile); + redisLog(REDIS_WARNING,"Write error writing append only file on disk: %s", strerror(errno)); + if (di) dictReleaseIterator(di); + return REDIS_ERR; +} + +/* This is how rewriting of the append only file in background works: + * + * 1) The user calls BGREWRITEAOF + * 2) Redis calls this function, that forks(): + * 2a) the child rewrite the append only file in a temp file. + * 2b) the parent accumulates differences in server.bgrewritebuf. + * 3) When the child finished '2a' exists. + * 4) The parent will trap the exit code, if it's OK, will append the + * data accumulated into server.bgrewritebuf into the temp file, and + * finally will rename(2) the temp file in the actual file name. + * The the new file is reopened as the new append only file. Profit! + */ +static int rewriteAppendOnlyFileBackground(void) { + pid_t childpid; + + if (server.bgrewritechildpid != -1) return REDIS_ERR; + if (server.vm_enabled) waitEmptyIOJobsQueue(); + if ((childpid = fork()) == 0) { + /* Child */ + char tmpfile[256]; + + if (server.vm_enabled) vmReopenSwapFile(); + close(server.fd); + snprintf(tmpfile,256,"temp-rewriteaof-bg-%d.aof", (int) getpid()); + if (rewriteAppendOnlyFile(tmpfile) == REDIS_OK) { + _exit(0); + } else { + _exit(1); + } + } else { + /* Parent */ + if (childpid == -1) { + redisLog(REDIS_WARNING, + "Can't rewrite append only file in background: fork: %s", + strerror(errno)); + return REDIS_ERR; + } + redisLog(REDIS_NOTICE, + "Background append only file rewriting started by pid %d",childpid); + server.bgrewritechildpid = childpid; + updateDictResizePolicy(); + /* We set appendseldb to -1 in order to force the next call to the + * feedAppendOnlyFile() to issue a SELECT command, so the differences + * accumulated by the parent into server.bgrewritebuf will start + * with a SELECT statement and it will be safe to merge. */ + server.appendseldb = -1; + return REDIS_OK; + } + return REDIS_OK; /* unreached */ +} + +static void bgrewriteaofCommand(redisClient *c) { + if (server.bgrewritechildpid != -1) { + addReplySds(c,sdsnew("-ERR background append only file rewriting already in progress\r\n")); + return; + } + if (rewriteAppendOnlyFileBackground() == REDIS_OK) { + char *status = "+Background append only file rewriting started\r\n"; + addReplySds(c,sdsnew(status)); + } else { + addReply(c,shared.err); + } +} + +static void aofRemoveTempFile(pid_t childpid) { + char tmpfile[256]; + + snprintf(tmpfile,256,"temp-rewriteaof-bg-%d.aof", (int) childpid); + unlink(tmpfile); +} + +/* Virtual Memory is composed mainly of two subsystems: + * - Blocking Virutal Memory + * - Threaded Virtual Memory I/O + * The two parts are not fully decoupled, but functions are split among two + * different sections of the source code (delimited by comments) in order to + * make more clear what functionality is about the blocking VM and what about + * the threaded (not blocking) VM. + * + * Redis VM design: + * + * Redis VM is a blocking VM (one that blocks reading swapped values from + * disk into memory when a value swapped out is needed in memory) that is made + * unblocking by trying to examine the command argument vector in order to + * load in background values that will likely be needed in order to exec + * the command. The command is executed only once all the relevant keys + * are loaded into memory. + * + * This basically is almost as simple of a blocking VM, but almost as parallel + * as a fully non-blocking VM. + */ + +/* =================== Virtual Memory - Blocking Side ====================== */ + +/* substitute the first occurrence of '%p' with the process pid in the + * swap file name. */ +static void expandVmSwapFilename(void) { + char *p = strstr(server.vm_swap_file,"%p"); + sds new; + + if (!p) return; + new = sdsempty(); + *p = '\0'; + new = sdscat(new,server.vm_swap_file); + new = sdscatprintf(new,"%ld",(long) getpid()); + new = sdscat(new,p+2); + zfree(server.vm_swap_file); + server.vm_swap_file = new; +} + +static void vmInit(void) { + off_t totsize; + int pipefds[2]; + size_t stacksize; + + if (server.vm_max_threads != 0) + zmalloc_enable_thread_safeness(); /* we need thread safe zmalloc() */ + + expandVmSwapFilename(); + redisLog(REDIS_NOTICE,"Using '%s' as swap file",server.vm_swap_file); + if ((server.vm_fp = fopen(server.vm_swap_file,"r+b")) == NULL) { + server.vm_fp = fopen(server.vm_swap_file,"w+b"); + } + if (server.vm_fp == NULL) { + redisLog(REDIS_WARNING, + "Impossible to open the swap file: %s. Exiting.", + strerror(errno)); + exit(1); + } + server.vm_fd = fileno(server.vm_fp); + server.vm_next_page = 0; + server.vm_near_pages = 0; + server.vm_stats_used_pages = 0; + server.vm_stats_swapped_objects = 0; + server.vm_stats_swapouts = 0; + server.vm_stats_swapins = 0; + totsize = server.vm_pages*server.vm_page_size; + redisLog(REDIS_NOTICE,"Allocating %lld bytes of swap file",totsize); + if (ftruncate(server.vm_fd,totsize) == -1) { + redisLog(REDIS_WARNING,"Can't ftruncate swap file: %s. Exiting.", + strerror(errno)); + exit(1); + } else { + redisLog(REDIS_NOTICE,"Swap file allocated with success"); + } + server.vm_bitmap = zmalloc((server.vm_pages+7)/8); + redisLog(REDIS_VERBOSE,"Allocated %lld bytes page table for %lld pages", + (long long) (server.vm_pages+7)/8, server.vm_pages); + memset(server.vm_bitmap,0,(server.vm_pages+7)/8); + + /* Initialize threaded I/O (used by Virtual Memory) */ + server.io_newjobs = listCreate(); + server.io_processing = listCreate(); + server.io_processed = listCreate(); + server.io_ready_clients = listCreate(); + pthread_mutex_init(&server.io_mutex,NULL); + pthread_mutex_init(&server.obj_freelist_mutex,NULL); + pthread_mutex_init(&server.io_swapfile_mutex,NULL); + server.io_active_threads = 0; + if (pipe(pipefds) == -1) { + redisLog(REDIS_WARNING,"Unable to intialized VM: pipe(2): %s. Exiting." + ,strerror(errno)); + exit(1); + } + server.io_ready_pipe_read = pipefds[0]; + server.io_ready_pipe_write = pipefds[1]; + redisAssert(anetNonBlock(NULL,server.io_ready_pipe_read) != ANET_ERR); + /* LZF requires a lot of stack */ + pthread_attr_init(&server.io_threads_attr); + pthread_attr_getstacksize(&server.io_threads_attr, &stacksize); + while (stacksize < REDIS_THREAD_STACK_SIZE) stacksize *= 2; + pthread_attr_setstacksize(&server.io_threads_attr, stacksize); + /* Listen for events in the threaded I/O pipe */ + if (aeCreateFileEvent(server.el, server.io_ready_pipe_read, AE_READABLE, + vmThreadedIOCompletedJob, NULL) == AE_ERR) + oom("creating file event"); +} + +/* Mark the page as used */ +static void vmMarkPageUsed(off_t page) { + off_t byte = page/8; + int bit = page&7; + redisAssert(vmFreePage(page) == 1); + server.vm_bitmap[byte] |= 1<= server.vm_pages) { + this -= server.vm_pages; + if (this == 0) { + /* Just overflowed, what we found on tail is no longer + * interesting, as it's no longer contiguous. */ + numfree = 0; + } + } + if (vmFreePage(this)) { + /* This is a free page */ + numfree++; + /* Already got N free pages? Return to the caller, with success */ + if (numfree == n) { + *first = this-(n-1); + server.vm_next_page = this+1; + redisLog(REDIS_DEBUG, "FOUND CONTIGUOUS PAGES: %lld pages at %lld\n", (long long) n, (long long) *first); + return REDIS_OK; + } + } else { + /* The current one is not a free page */ + numfree = 0; + } + + /* Fast-forward if the current page is not free and we already + * searched enough near this place. */ + since_jump++; + if (!numfree && since_jump >= REDIS_VM_MAX_RANDOM_JUMP/4) { + offset += random() % REDIS_VM_MAX_RANDOM_JUMP; + since_jump = 0; + /* Note that even if we rewind after the jump, we are don't need + * to make sure numfree is set to zero as we only jump *if* it + * is set to zero. */ + } else { + /* Otherwise just check the next page */ + offset++; + } + } + return REDIS_ERR; +} + +/* Write the specified object at the specified page of the swap file */ +static int vmWriteObjectOnSwap(robj *o, off_t page) { + if (server.vm_enabled) pthread_mutex_lock(&server.io_swapfile_mutex); + if (fseeko(server.vm_fp,page*server.vm_page_size,SEEK_SET) == -1) { + if (server.vm_enabled) pthread_mutex_unlock(&server.io_swapfile_mutex); + redisLog(REDIS_WARNING, + "Critical VM problem in vmWriteObjectOnSwap(): can't seek: %s", + strerror(errno)); + return REDIS_ERR; + } + rdbSaveObject(server.vm_fp,o); + fflush(server.vm_fp); + if (server.vm_enabled) pthread_mutex_unlock(&server.io_swapfile_mutex); + return REDIS_OK; +} + +/* Swap the 'val' object relative to 'key' into disk. Store all the information + * needed to later retrieve the object into the key object. + * If we can't find enough contiguous empty pages to swap the object on disk + * REDIS_ERR is returned. */ +static int vmSwapObjectBlocking(robj *key, robj *val) { + off_t pages = rdbSavedObjectPages(val,NULL); + off_t page; + + assert(key->storage == REDIS_VM_MEMORY); + assert(key->refcount == 1); + if (vmFindContiguousPages(&page,pages) == REDIS_ERR) return REDIS_ERR; + if (vmWriteObjectOnSwap(val,page) == REDIS_ERR) return REDIS_ERR; + key->vm.page = page; + key->vm.usedpages = pages; + key->storage = REDIS_VM_SWAPPED; + key->vtype = val->type; + decrRefCount(val); /* Deallocate the object from memory. */ + vmMarkPagesUsed(page,pages); + redisLog(REDIS_DEBUG,"VM: object %s swapped out at %lld (%lld pages)", + (unsigned char*) key->ptr, + (unsigned long long) page, (unsigned long long) pages); + server.vm_stats_swapped_objects++; + server.vm_stats_swapouts++; + return REDIS_OK; +} + +static robj *vmReadObjectFromSwap(off_t page, int type) { + robj *o; + + if (server.vm_enabled) pthread_mutex_lock(&server.io_swapfile_mutex); + if (fseeko(server.vm_fp,page*server.vm_page_size,SEEK_SET) == -1) { + redisLog(REDIS_WARNING, + "Unrecoverable VM problem in vmReadObjectFromSwap(): can't seek: %s", + strerror(errno)); + _exit(1); + } + o = rdbLoadObject(type,server.vm_fp); + if (o == NULL) { + redisLog(REDIS_WARNING, "Unrecoverable VM problem in vmReadObjectFromSwap(): can't load object from swap file: %s", strerror(errno)); + _exit(1); + } + if (server.vm_enabled) pthread_mutex_unlock(&server.io_swapfile_mutex); + return o; +} + +/* Load the value object relative to the 'key' object from swap to memory. + * The newly allocated object is returned. + * + * If preview is true the unserialized object is returned to the caller but + * no changes are made to the key object, nor the pages are marked as freed */ +static robj *vmGenericLoadObject(robj *key, int preview) { + robj *val; + + redisAssert(key->storage == REDIS_VM_SWAPPED || key->storage == REDIS_VM_LOADING); + val = vmReadObjectFromSwap(key->vm.page,key->vtype); + if (!preview) { + key->storage = REDIS_VM_MEMORY; + key->vm.atime = server.unixtime; + vmMarkPagesFree(key->vm.page,key->vm.usedpages); + redisLog(REDIS_DEBUG, "VM: object %s loaded from disk", + (unsigned char*) key->ptr); + server.vm_stats_swapped_objects--; + } else { + redisLog(REDIS_DEBUG, "VM: object %s previewed from disk", + (unsigned char*) key->ptr); + } + server.vm_stats_swapins++; + return val; +} + +/* Plain object loading, from swap to memory */ +static robj *vmLoadObject(robj *key) { + /* If we are loading the object in background, stop it, we + * need to load this object synchronously ASAP. */ + if (key->storage == REDIS_VM_LOADING) + vmCancelThreadedIOJob(key); + return vmGenericLoadObject(key,0); +} + +/* Just load the value on disk, without to modify the key. + * This is useful when we want to perform some operation on the value + * without to really bring it from swap to memory, like while saving the + * dataset or rewriting the append only log. */ +static robj *vmPreviewObject(robj *key) { + return vmGenericLoadObject(key,1); +} + +/* How a good candidate is this object for swapping? + * The better candidate it is, the greater the returned value. + * + * Currently we try to perform a fast estimation of the object size in + * memory, and combine it with aging informations. + * + * Basically swappability = idle-time * log(estimated size) + * + * Bigger objects are preferred over smaller objects, but not + * proportionally, this is why we use the logarithm. This algorithm is + * just a first try and will probably be tuned later. */ +static double computeObjectSwappability(robj *o) { + time_t age = server.unixtime - o->vm.atime; + long asize = 0; + list *l; + dict *d; + struct dictEntry *de; + int z; + + if (age <= 0) return 0; + switch(o->type) { + case REDIS_STRING: + if (o->encoding != REDIS_ENCODING_RAW) { + asize = sizeof(*o); + } else { + asize = sdslen(o->ptr)+sizeof(*o)+sizeof(long)*2; + } + break; + case REDIS_LIST: + l = o->ptr; + listNode *ln = listFirst(l); + + asize = sizeof(list); + if (ln) { + robj *ele = ln->value; + long elesize; + + elesize = (ele->encoding == REDIS_ENCODING_RAW) ? + (sizeof(*o)+sdslen(ele->ptr)) : + sizeof(*o); + asize += (sizeof(listNode)+elesize)*listLength(l); + } + break; + case REDIS_SET: + case REDIS_ZSET: + z = (o->type == REDIS_ZSET); + d = z ? ((zset*)o->ptr)->dict : o->ptr; + + asize = sizeof(dict)+(sizeof(struct dictEntry*)*dictSlots(d)); + if (z) asize += sizeof(zset)-sizeof(dict); + if (dictSize(d)) { + long elesize; + robj *ele; + + de = dictGetRandomKey(d); + ele = dictGetEntryKey(de); + elesize = (ele->encoding == REDIS_ENCODING_RAW) ? + (sizeof(*o)+sdslen(ele->ptr)) : + sizeof(*o); + asize += (sizeof(struct dictEntry)+elesize)*dictSize(d); + if (z) asize += sizeof(zskiplistNode)*dictSize(d); + } + break; + case REDIS_HASH: + if (o->encoding == REDIS_ENCODING_ZIPMAP) { + unsigned char *p = zipmapRewind((unsigned char*)o->ptr); + unsigned int len = zipmapLen((unsigned char*)o->ptr); + unsigned int klen, vlen; + unsigned char *key, *val; + + if ((p = zipmapNext(p,&key,&klen,&val,&vlen)) == NULL) { + klen = 0; + vlen = 0; + } + asize = len*(klen+vlen+3); + } else if (o->encoding == REDIS_ENCODING_HT) { + d = o->ptr; + asize = sizeof(dict)+(sizeof(struct dictEntry*)*dictSlots(d)); + if (dictSize(d)) { + long elesize; + robj *ele; + + de = dictGetRandomKey(d); + ele = dictGetEntryKey(de); + elesize = (ele->encoding == REDIS_ENCODING_RAW) ? + (sizeof(*o)+sdslen(ele->ptr)) : + sizeof(*o); + ele = dictGetEntryVal(de); + elesize = (ele->encoding == REDIS_ENCODING_RAW) ? + (sizeof(*o)+sdslen(ele->ptr)) : + sizeof(*o); + asize += (sizeof(struct dictEntry)+elesize)*dictSize(d); + } + } + break; + } + return (double)age*log(1+asize); +} + +/* Try to swap an object that's a good candidate for swapping. + * Returns REDIS_OK if the object was swapped, REDIS_ERR if it's not possible + * to swap any object at all. + * + * If 'usethreaded' is true, Redis will try to swap the object in background + * using I/O threads. */ +static int vmSwapOneObject(int usethreads) { + int j, i; + struct dictEntry *best = NULL; + double best_swappability = 0; + redisDb *best_db = NULL; + robj *key, *val; + + for (j = 0; j < server.dbnum; j++) { + redisDb *db = server.db+j; + /* Why maxtries is set to 100? + * Because this way (usually) we'll find 1 object even if just 1% - 2% + * are swappable objects */ + int maxtries = 100; + + if (dictSize(db->dict) == 0) continue; + for (i = 0; i < 5; i++) { + dictEntry *de; + double swappability; + + if (maxtries) maxtries--; + de = dictGetRandomKey(db->dict); + key = dictGetEntryKey(de); + val = dictGetEntryVal(de); + /* Only swap objects that are currently in memory. + * + * Also don't swap shared objects if threaded VM is on, as we + * try to ensure that the main thread does not touch the + * object while the I/O thread is using it, but we can't + * control other keys without adding additional mutex. */ + if (key->storage != REDIS_VM_MEMORY || + (server.vm_max_threads != 0 && val->refcount != 1)) { + if (maxtries) i--; /* don't count this try */ + continue; + } + swappability = computeObjectSwappability(val); + if (!best || swappability > best_swappability) { + best = de; + best_swappability = swappability; + best_db = db; + } + } + } + if (best == NULL) return REDIS_ERR; + key = dictGetEntryKey(best); + val = dictGetEntryVal(best); + + redisLog(REDIS_DEBUG,"Key with best swappability: %s, %f", + key->ptr, best_swappability); + + /* Unshare the key if needed */ + if (key->refcount > 1) { + robj *newkey = dupStringObject(key); + decrRefCount(key); + key = dictGetEntryKey(best) = newkey; + } + /* Swap it */ + if (usethreads) { + vmSwapObjectThreaded(key,val,best_db); + return REDIS_OK; + } else { + if (vmSwapObjectBlocking(key,val) == REDIS_OK) { + dictGetEntryVal(best) = NULL; + return REDIS_OK; + } else { + return REDIS_ERR; + } + } +} + +static int vmSwapOneObjectBlocking() { + return vmSwapOneObject(0); +} + +static int vmSwapOneObjectThreaded() { + return vmSwapOneObject(1); +} + +/* Return true if it's safe to swap out objects in a given moment. + * Basically we don't want to swap objects out while there is a BGSAVE + * or a BGAEOREWRITE running in backgroud. */ +static int vmCanSwapOut(void) { + return (server.bgsavechildpid == -1 && server.bgrewritechildpid == -1); +} + +/* Delete a key if swapped. Returns 1 if the key was found, was swapped + * and was deleted. Otherwise 0 is returned. */ +static int deleteIfSwapped(redisDb *db, robj *key) { + dictEntry *de; + robj *foundkey; + + if ((de = dictFind(db->dict,key)) == NULL) return 0; + foundkey = dictGetEntryKey(de); + if (foundkey->storage == REDIS_VM_MEMORY) return 0; + deleteKey(db,key); + return 1; +} + +/* =================== Virtual Memory - Threaded I/O ======================= */ + +static void freeIOJob(iojob *j) { + if ((j->type == REDIS_IOJOB_PREPARE_SWAP || + j->type == REDIS_IOJOB_DO_SWAP || + j->type == REDIS_IOJOB_LOAD) && j->val != NULL) + decrRefCount(j->val); + /* We don't decrRefCount the j->key field as we did't incremented + * the count creating IO Jobs. This is because the key field here is + * just used as an indentifier and if a key is removed the Job should + * never be touched again. */ + zfree(j); +} + +/* Every time a thread finished a Job, it writes a byte into the write side + * of an unix pipe in order to "awake" the main thread, and this function + * is called. */ +static void vmThreadedIOCompletedJob(aeEventLoop *el, int fd, void *privdata, + int mask) +{ + char buf[1]; + int retval, processed = 0, toprocess = -1, trytoswap = 1; + REDIS_NOTUSED(el); + REDIS_NOTUSED(mask); + REDIS_NOTUSED(privdata); + + /* For every byte we read in the read side of the pipe, there is one + * I/O job completed to process. */ + while((retval = read(fd,buf,1)) == 1) { + iojob *j; + listNode *ln; + robj *key; + struct dictEntry *de; + + redisLog(REDIS_DEBUG,"Processing I/O completed job"); + + /* Get the processed element (the oldest one) */ + lockThreadedIO(); + assert(listLength(server.io_processed) != 0); + if (toprocess == -1) { + toprocess = (listLength(server.io_processed)*REDIS_MAX_COMPLETED_JOBS_PROCESSED)/100; + if (toprocess <= 0) toprocess = 1; + } + ln = listFirst(server.io_processed); + j = ln->value; + listDelNode(server.io_processed,ln); + unlockThreadedIO(); + /* If this job is marked as canceled, just ignore it */ + if (j->canceled) { + freeIOJob(j); + continue; + } + /* Post process it in the main thread, as there are things we + * can do just here to avoid race conditions and/or invasive locks */ + redisLog(REDIS_DEBUG,"Job %p type: %d, key at %p (%s) refcount: %d\n", (void*) j, j->type, (void*)j->key, (char*)j->key->ptr, j->key->refcount); + de = dictFind(j->db->dict,j->key); + assert(de != NULL); + key = dictGetEntryKey(de); + if (j->type == REDIS_IOJOB_LOAD) { + redisDb *db; + + /* Key loaded, bring it at home */ + key->storage = REDIS_VM_MEMORY; + key->vm.atime = server.unixtime; + vmMarkPagesFree(key->vm.page,key->vm.usedpages); + redisLog(REDIS_DEBUG, "VM: object %s loaded from disk (threaded)", + (unsigned char*) key->ptr); + server.vm_stats_swapped_objects--; + server.vm_stats_swapins++; + dictGetEntryVal(de) = j->val; + incrRefCount(j->val); + db = j->db; + freeIOJob(j); + /* Handle clients waiting for this key to be loaded. */ + handleClientsBlockedOnSwappedKey(db,key); + } else if (j->type == REDIS_IOJOB_PREPARE_SWAP) { + /* Now we know the amount of pages required to swap this object. + * Let's find some space for it, and queue this task again + * rebranded as REDIS_IOJOB_DO_SWAP. */ + if (!vmCanSwapOut() || + vmFindContiguousPages(&j->page,j->pages) == REDIS_ERR) + { + /* Ooops... no space or we can't swap as there is + * a fork()ed Redis trying to save stuff on disk. */ + freeIOJob(j); + key->storage = REDIS_VM_MEMORY; /* undo operation */ + } else { + /* Note that we need to mark this pages as used now, + * if the job will be canceled, we'll mark them as freed + * again. */ + vmMarkPagesUsed(j->page,j->pages); + j->type = REDIS_IOJOB_DO_SWAP; + lockThreadedIO(); + queueIOJob(j); + unlockThreadedIO(); + } + } else if (j->type == REDIS_IOJOB_DO_SWAP) { + robj *val; + + /* Key swapped. We can finally free some memory. */ + if (key->storage != REDIS_VM_SWAPPING) { + printf("key->storage: %d\n",key->storage); + printf("key->name: %s\n",(char*)key->ptr); + printf("key->refcount: %d\n",key->refcount); + printf("val: %p\n",(void*)j->val); + printf("val->type: %d\n",j->val->type); + printf("val->ptr: %s\n",(char*)j->val->ptr); + } + redisAssert(key->storage == REDIS_VM_SWAPPING); + val = dictGetEntryVal(de); + key->vm.page = j->page; + key->vm.usedpages = j->pages; + key->storage = REDIS_VM_SWAPPED; + key->vtype = j->val->type; + decrRefCount(val); /* Deallocate the object from memory. */ + dictGetEntryVal(de) = NULL; + redisLog(REDIS_DEBUG, + "VM: object %s swapped out at %lld (%lld pages) (threaded)", + (unsigned char*) key->ptr, + (unsigned long long) j->page, (unsigned long long) j->pages); + server.vm_stats_swapped_objects++; + server.vm_stats_swapouts++; + freeIOJob(j); + /* Put a few more swap requests in queue if we are still + * out of memory */ + if (trytoswap && vmCanSwapOut() && + zmalloc_used_memory() > server.vm_max_memory) + { + int more = 1; + while(more) { + lockThreadedIO(); + more = listLength(server.io_newjobs) < + (unsigned) server.vm_max_threads; + unlockThreadedIO(); + /* Don't waste CPU time if swappable objects are rare. */ + if (vmSwapOneObjectThreaded() == REDIS_ERR) { + trytoswap = 0; + break; + } + } + } + } + processed++; + if (processed == toprocess) return; + } + if (retval < 0 && errno != EAGAIN) { + redisLog(REDIS_WARNING, + "WARNING: read(2) error in vmThreadedIOCompletedJob() %s", + strerror(errno)); + } +} + +static void lockThreadedIO(void) { + pthread_mutex_lock(&server.io_mutex); +} + +static void unlockThreadedIO(void) { + pthread_mutex_unlock(&server.io_mutex); +} + +/* Remove the specified object from the threaded I/O queue if still not + * processed, otherwise make sure to flag it as canceled. */ +static void vmCancelThreadedIOJob(robj *o) { + list *lists[3] = { + server.io_newjobs, /* 0 */ + server.io_processing, /* 1 */ + server.io_processed /* 2 */ + }; + int i; + + assert(o->storage == REDIS_VM_LOADING || o->storage == REDIS_VM_SWAPPING); +again: + lockThreadedIO(); + /* Search for a matching key in one of the queues */ + for (i = 0; i < 3; i++) { + listNode *ln; + listIter li; + + listRewind(lists[i],&li); + while ((ln = listNext(&li)) != NULL) { + iojob *job = ln->value; + + if (job->canceled) continue; /* Skip this, already canceled. */ + if (job->key == o) { + redisLog(REDIS_DEBUG,"*** CANCELED %p (%s) (type %d) (LIST ID %d)\n", + (void*)job, (char*)o->ptr, job->type, i); + /* Mark the pages as free since the swap didn't happened + * or happened but is now discarded. */ + if (i != 1 && job->type == REDIS_IOJOB_DO_SWAP) + vmMarkPagesFree(job->page,job->pages); + /* Cancel the job. It depends on the list the job is + * living in. */ + switch(i) { + case 0: /* io_newjobs */ + /* If the job was yet not processed the best thing to do + * is to remove it from the queue at all */ + freeIOJob(job); + listDelNode(lists[i],ln); + break; + case 1: /* io_processing */ + /* Oh Shi- the thread is messing with the Job: + * + * Probably it's accessing the object if this is a + * PREPARE_SWAP or DO_SWAP job. + * If it's a LOAD job it may be reading from disk and + * if we don't wait for the job to terminate before to + * cancel it, maybe in a few microseconds data can be + * corrupted in this pages. So the short story is: + * + * Better to wait for the job to move into the + * next queue (processed)... */ + + /* We try again and again until the job is completed. */ + unlockThreadedIO(); + /* But let's wait some time for the I/O thread + * to finish with this job. After all this condition + * should be very rare. */ + usleep(1); + goto again; + case 2: /* io_processed */ + /* The job was already processed, that's easy... + * just mark it as canceled so that we'll ignore it + * when processing completed jobs. */ + job->canceled = 1; + break; + } + /* Finally we have to adjust the storage type of the object + * in order to "UNDO" the operaiton. */ + if (o->storage == REDIS_VM_LOADING) + o->storage = REDIS_VM_SWAPPED; + else if (o->storage == REDIS_VM_SWAPPING) + o->storage = REDIS_VM_MEMORY; + unlockThreadedIO(); + return; + } + } + } + unlockThreadedIO(); + assert(1 != 1); /* We should never reach this */ +} + +static void *IOThreadEntryPoint(void *arg) { + iojob *j; + listNode *ln; + REDIS_NOTUSED(arg); + + pthread_detach(pthread_self()); + while(1) { + /* Get a new job to process */ + lockThreadedIO(); + if (listLength(server.io_newjobs) == 0) { + /* No new jobs in queue, exit. */ + redisLog(REDIS_DEBUG,"Thread %ld exiting, nothing to do", + (long) pthread_self()); + server.io_active_threads--; + unlockThreadedIO(); + return NULL; + } + ln = listFirst(server.io_newjobs); + j = ln->value; + listDelNode(server.io_newjobs,ln); + /* Add the job in the processing queue */ + j->thread = pthread_self(); + listAddNodeTail(server.io_processing,j); + ln = listLast(server.io_processing); /* We use ln later to remove it */ + unlockThreadedIO(); + redisLog(REDIS_DEBUG,"Thread %ld got a new job (type %d): %p about key '%s'", + (long) pthread_self(), j->type, (void*)j, (char*)j->key->ptr); + + /* Process the Job */ + if (j->type == REDIS_IOJOB_LOAD) { + j->val = vmReadObjectFromSwap(j->page,j->key->vtype); + } else if (j->type == REDIS_IOJOB_PREPARE_SWAP) { + FILE *fp = fopen("/dev/null","w+"); + j->pages = rdbSavedObjectPages(j->val,fp); + fclose(fp); + } else if (j->type == REDIS_IOJOB_DO_SWAP) { + if (vmWriteObjectOnSwap(j->val,j->page) == REDIS_ERR) + j->canceled = 1; + } + + /* Done: insert the job into the processed queue */ + redisLog(REDIS_DEBUG,"Thread %ld completed the job: %p (key %s)", + (long) pthread_self(), (void*)j, (char*)j->key->ptr); + lockThreadedIO(); + listDelNode(server.io_processing,ln); + listAddNodeTail(server.io_processed,j); + unlockThreadedIO(); + + /* Signal the main thread there is new stuff to process */ + assert(write(server.io_ready_pipe_write,"x",1) == 1); + } + return NULL; /* never reached */ +} + +static void spawnIOThread(void) { + pthread_t thread; + sigset_t mask, omask; + int err; + + sigemptyset(&mask); + sigaddset(&mask,SIGCHLD); + sigaddset(&mask,SIGHUP); + sigaddset(&mask,SIGPIPE); + pthread_sigmask(SIG_SETMASK, &mask, &omask); + while ((err = pthread_create(&thread,&server.io_threads_attr,IOThreadEntryPoint,NULL)) != 0) { + redisLog(REDIS_WARNING,"Unable to spawn an I/O thread: %s", + strerror(err)); + usleep(1000000); + } + pthread_sigmask(SIG_SETMASK, &omask, NULL); + server.io_active_threads++; +} + +/* We need to wait for the last thread to exit before we are able to + * fork() in order to BGSAVE or BGREWRITEAOF. */ +static void waitEmptyIOJobsQueue(void) { + while(1) { + int io_processed_len; + + lockThreadedIO(); + if (listLength(server.io_newjobs) == 0 && + listLength(server.io_processing) == 0 && + server.io_active_threads == 0) + { + unlockThreadedIO(); + return; + } + /* While waiting for empty jobs queue condition we post-process some + * finshed job, as I/O threads may be hanging trying to write against + * the io_ready_pipe_write FD but there are so much pending jobs that + * it's blocking. */ + io_processed_len = listLength(server.io_processed); + unlockThreadedIO(); + if (io_processed_len) { + vmThreadedIOCompletedJob(NULL,server.io_ready_pipe_read,NULL,0); + usleep(1000); /* 1 millisecond */ + } else { + usleep(10000); /* 10 milliseconds */ + } + } +} + +static void vmReopenSwapFile(void) { + /* Note: we don't close the old one as we are in the child process + * and don't want to mess at all with the original file object. */ + server.vm_fp = fopen(server.vm_swap_file,"r+b"); + if (server.vm_fp == NULL) { + redisLog(REDIS_WARNING,"Can't re-open the VM swap file: %s. Exiting.", + server.vm_swap_file); + _exit(1); + } + server.vm_fd = fileno(server.vm_fp); +} + +/* This function must be called while with threaded IO locked */ +static void queueIOJob(iojob *j) { + redisLog(REDIS_DEBUG,"Queued IO Job %p type %d about key '%s'\n", + (void*)j, j->type, (char*)j->key->ptr); + listAddNodeTail(server.io_newjobs,j); + if (server.io_active_threads < server.vm_max_threads) + spawnIOThread(); +} + +static int vmSwapObjectThreaded(robj *key, robj *val, redisDb *db) { + iojob *j; + + assert(key->storage == REDIS_VM_MEMORY); + assert(key->refcount == 1); + + j = zmalloc(sizeof(*j)); + j->type = REDIS_IOJOB_PREPARE_SWAP; + j->db = db; + j->key = key; + j->val = val; + incrRefCount(val); + j->canceled = 0; + j->thread = (pthread_t) -1; + key->storage = REDIS_VM_SWAPPING; + + lockThreadedIO(); + queueIOJob(j); + unlockThreadedIO(); + return REDIS_OK; +} + +/* ============ Virtual Memory - Blocking clients on missing keys =========== */ + +/* This function makes the clinet 'c' waiting for the key 'key' to be loaded. + * If there is not already a job loading the key, it is craeted. + * The key is added to the io_keys list in the client structure, and also + * in the hash table mapping swapped keys to waiting clients, that is, + * server.io_waited_keys. */ +static int waitForSwappedKey(redisClient *c, robj *key) { + struct dictEntry *de; + robj *o; + list *l; + + /* If the key does not exist or is already in RAM we don't need to + * block the client at all. */ + de = dictFind(c->db->dict,key); + if (de == NULL) return 0; + o = dictGetEntryKey(de); + if (o->storage == REDIS_VM_MEMORY) { + return 0; + } else if (o->storage == REDIS_VM_SWAPPING) { + /* We were swapping the key, undo it! */ + vmCancelThreadedIOJob(o); + return 0; + } + + /* OK: the key is either swapped, or being loaded just now. */ + + /* Add the key to the list of keys this client is waiting for. + * This maps clients to keys they are waiting for. */ + listAddNodeTail(c->io_keys,key); + incrRefCount(key); + + /* Add the client to the swapped keys => clients waiting map. */ + de = dictFind(c->db->io_keys,key); + if (de == NULL) { + int retval; + + /* For every key we take a list of clients blocked for it */ + l = listCreate(); + retval = dictAdd(c->db->io_keys,key,l); + incrRefCount(key); + assert(retval == DICT_OK); + } else { + l = dictGetEntryVal(de); + } + listAddNodeTail(l,c); + + /* Are we already loading the key from disk? If not create a job */ + if (o->storage == REDIS_VM_SWAPPED) { + iojob *j; + + o->storage = REDIS_VM_LOADING; + j = zmalloc(sizeof(*j)); + j->type = REDIS_IOJOB_LOAD; + j->db = c->db; + j->key = o; + j->key->vtype = o->vtype; + j->page = o->vm.page; + j->val = NULL; + j->canceled = 0; + j->thread = (pthread_t) -1; + lockThreadedIO(); + queueIOJob(j); + unlockThreadedIO(); + } + return 1; +} + +/* Preload keys needed for the ZUNION and ZINTER commands. */ +static void zunionInterBlockClientOnSwappedKeys(redisClient *c) { + int i, num; + num = atoi(c->argv[2]->ptr); + for (i = 0; i < num; i++) { + waitForSwappedKey(c,c->argv[3+i]); + } +} + +/* Is this client attempting to run a command against swapped keys? + * If so, block it ASAP, load the keys in background, then resume it. + * + * The important idea about this function is that it can fail! If keys will + * still be swapped when the client is resumed, this key lookups will + * just block loading keys from disk. In practical terms this should only + * happen with SORT BY command or if there is a bug in this function. + * + * Return 1 if the client is marked as blocked, 0 if the client can + * continue as the keys it is going to access appear to be in memory. */ +static int blockClientOnSwappedKeys(struct redisCommand *cmd, redisClient *c) { + int j, last; + + if (cmd->vm_preload_proc != NULL) { + cmd->vm_preload_proc(c); + } else { + if (cmd->vm_firstkey == 0) return 0; + last = cmd->vm_lastkey; + if (last < 0) last = c->argc+last; + for (j = cmd->vm_firstkey; j <= last; j += cmd->vm_keystep) + waitForSwappedKey(c,c->argv[j]); + } + + /* If the client was blocked for at least one key, mark it as blocked. */ + if (listLength(c->io_keys)) { + c->flags |= REDIS_IO_WAIT; + aeDeleteFileEvent(server.el,c->fd,AE_READABLE); + server.vm_blocked_clients++; + return 1; + } else { + return 0; + } +} + +/* Remove the 'key' from the list of blocked keys for a given client. + * + * The function returns 1 when there are no longer blocking keys after + * the current one was removed (and the client can be unblocked). */ +static int dontWaitForSwappedKey(redisClient *c, robj *key) { + list *l; + listNode *ln; + listIter li; + struct dictEntry *de; + + /* Remove the key from the list of keys this client is waiting for. */ + listRewind(c->io_keys,&li); + while ((ln = listNext(&li)) != NULL) { + if (compareStringObjects(ln->value,key) == 0) { + listDelNode(c->io_keys,ln); + break; + } + } + assert(ln != NULL); + + /* Remove the client form the key => waiting clients map. */ + de = dictFind(c->db->io_keys,key); + assert(de != NULL); + l = dictGetEntryVal(de); + ln = listSearchKey(l,c); + assert(ln != NULL); + listDelNode(l,ln); + if (listLength(l) == 0) + dictDelete(c->db->io_keys,key); + + return listLength(c->io_keys) == 0; +} + +static void handleClientsBlockedOnSwappedKey(redisDb *db, robj *key) { + struct dictEntry *de; + list *l; + listNode *ln; + int len; + + de = dictFind(db->io_keys,key); + if (!de) return; + + l = dictGetEntryVal(de); + len = listLength(l); + /* Note: we can't use something like while(listLength(l)) as the list + * can be freed by the calling function when we remove the last element. */ + while (len--) { + ln = listFirst(l); + redisClient *c = ln->value; + + if (dontWaitForSwappedKey(c,key)) { + /* Put the client in the list of clients ready to go as we + * loaded all the keys about it. */ + listAddNodeTail(server.io_ready_clients,c); + } + } +} + +/* =========================== Remote Configuration ========================= */ + +static void configSetCommand(redisClient *c) { + robj *o = getDecodedObject(c->argv[3]); + if (!strcasecmp(c->argv[2]->ptr,"dbfilename")) { + zfree(server.dbfilename); + server.dbfilename = zstrdup(o->ptr); + } else if (!strcasecmp(c->argv[2]->ptr,"requirepass")) { + zfree(server.requirepass); + server.requirepass = zstrdup(o->ptr); + } else if (!strcasecmp(c->argv[2]->ptr,"masterauth")) { + zfree(server.masterauth); + server.masterauth = zstrdup(o->ptr); + } else if (!strcasecmp(c->argv[2]->ptr,"maxmemory")) { + server.maxmemory = strtoll(o->ptr, NULL, 10); + } else { + addReplySds(c,sdscatprintf(sdsempty(), + "-ERR not supported CONFIG parameter %s\r\n", + (char*)c->argv[2]->ptr)); + decrRefCount(o); + return; + } + decrRefCount(o); + addReply(c,shared.ok); +} + +static void configGetCommand(redisClient *c) { + robj *o = getDecodedObject(c->argv[2]); + robj *lenobj = createObject(REDIS_STRING,NULL); + char *pattern = o->ptr; + int matches = 0; + + addReply(c,lenobj); + decrRefCount(lenobj); + + if (stringmatch(pattern,"dbfilename",0)) { + addReplyBulkCString(c,"dbfilename"); + addReplyBulkCString(c,server.dbfilename); + matches++; + } + if (stringmatch(pattern,"requirepass",0)) { + addReplyBulkCString(c,"requirepass"); + addReplyBulkCString(c,server.requirepass); + matches++; + } + if (stringmatch(pattern,"masterauth",0)) { + addReplyBulkCString(c,"masterauth"); + addReplyBulkCString(c,server.masterauth); + matches++; + } + if (stringmatch(pattern,"maxmemory",0)) { + char buf[128]; + + snprintf(buf,128,"%llu\n",server.maxmemory); + addReplyBulkCString(c,"maxmemory"); + addReplyBulkCString(c,buf); + matches++; + } + decrRefCount(o); + lenobj->ptr = sdscatprintf(sdsempty(),"*%d\r\n",matches*2); +} + +static void configCommand(redisClient *c) { + if (!strcasecmp(c->argv[1]->ptr,"set")) { + if (c->argc != 4) goto badarity; + configSetCommand(c); + } else if (!strcasecmp(c->argv[1]->ptr,"get")) { + if (c->argc != 3) goto badarity; + configGetCommand(c); + } else if (!strcasecmp(c->argv[1]->ptr,"resetstat")) { + if (c->argc != 2) goto badarity; + server.stat_numcommands = 0; + server.stat_numconnections = 0; + server.stat_expiredkeys = 0; + server.stat_starttime = time(NULL); + addReply(c,shared.ok); + } else { + addReplySds(c,sdscatprintf(sdsempty(), + "-ERR CONFIG subcommand must be one of GET, SET, RESETSTAT\r\n")); + } + return; + +badarity: + addReplySds(c,sdscatprintf(sdsempty(), + "-ERR Wrong number of arguments for CONFIG %s\r\n", + (char*) c->argv[1]->ptr)); } -/* Write a long value in bulk format $\r\n\r\n */ -static int fwriteBulkLong(FILE *fp, long l) { - char buf[128], lbuf[128]; +/* =========================== Pubsub implementation ======================== */ - snprintf(lbuf,sizeof(lbuf),"%ld\r\n",l); - snprintf(buf,sizeof(buf),"$%lu\r\n",(unsigned long)strlen(lbuf)-2); - if (fwrite(buf,strlen(buf),1,fp) == 0) return 0; - if (fwrite(lbuf,strlen(lbuf),1,fp) == 0) return 0; - return 1; +static void freePubsubPattern(void *p) { + pubsubPattern *pat = p; + + decrRefCount(pat->pattern); + zfree(pat); } -/* Write a sequence of commands able to fully rebuild the dataset into - * "filename". Used both by REWRITEAOF and BGREWRITEAOF. */ -static int rewriteAppendOnlyFile(char *filename) { - dictIterator *di = NULL; - dictEntry *de; - FILE *fp; - char tmpfile[256]; - int j; - time_t now = time(NULL); +static int listMatchPubsubPattern(void *a, void *b) { + pubsubPattern *pa = a, *pb = b; - /* Note that we have to use a different temp name here compared to the - * one used by rewriteAppendOnlyFileBackground() function. */ - snprintf(tmpfile,256,"temp-rewriteaof-%d.aof", (int) getpid()); - fp = fopen(tmpfile,"w"); - if (!fp) { - redisLog(REDIS_WARNING, "Failed rewriting the append only file: %s", strerror(errno)); - return REDIS_ERR; - } - for (j = 0; j < server.dbnum; j++) { - char selectcmd[] = "*2\r\n$6\r\nSELECT\r\n"; - redisDb *db = server.db+j; - dict *d = db->dict; - if (dictSize(d) == 0) continue; - di = dictGetIterator(d); - if (!di) { - fclose(fp); - return REDIS_ERR; + return (pa->client == pb->client) && + (compareStringObjects(pa->pattern,pb->pattern) == 0); +} + +/* Subscribe a client to a channel. Returns 1 if the operation succeeded, or + * 0 if the client was already subscribed to that channel. */ +static int pubsubSubscribeChannel(redisClient *c, robj *channel) { + struct dictEntry *de; + list *clients = NULL; + int retval = 0; + + /* Add the channel to the client -> channels hash table */ + if (dictAdd(c->pubsub_channels,channel,NULL) == DICT_OK) { + retval = 1; + incrRefCount(channel); + /* Add the client to the channel -> list of clients hash table */ + de = dictFind(server.pubsub_channels,channel); + if (de == NULL) { + clients = listCreate(); + dictAdd(server.pubsub_channels,channel,clients); + incrRefCount(channel); + } else { + clients = dictGetEntryVal(de); } + listAddNodeTail(clients,c); + } + /* Notify the client */ + addReply(c,shared.mbulk3); + addReply(c,shared.subscribebulk); + addReplyBulk(c,channel); + addReplyLong(c,dictSize(c->pubsub_channels)+listLength(c->pubsub_patterns)); + return retval; +} - /* SELECT the new DB */ - if (fwrite(selectcmd,sizeof(selectcmd)-1,1,fp) == 0) goto werr; - if (fwriteBulkLong(fp,j) == 0) goto werr; +/* Unsubscribe a client from a channel. Returns 1 if the operation succeeded, or + * 0 if the client was not subscribed to the specified channel. */ +static int pubsubUnsubscribeChannel(redisClient *c, robj *channel, int notify) { + struct dictEntry *de; + list *clients; + listNode *ln; + int retval = 0; + + /* Remove the channel from the client -> channels hash table */ + incrRefCount(channel); /* channel may be just a pointer to the same object + we have in the hash tables. Protect it... */ + if (dictDelete(c->pubsub_channels,channel) == DICT_OK) { + retval = 1; + /* Remove the client from the channel -> clients list hash table */ + de = dictFind(server.pubsub_channels,channel); + assert(de != NULL); + clients = dictGetEntryVal(de); + ln = listSearchKey(clients,c); + assert(ln != NULL); + listDelNode(clients,ln); + if (listLength(clients) == 0) { + /* Free the list and associated hash entry at all if this was + * the latest client, so that it will be possible to abuse + * Redis PUBSUB creating millions of channels. */ + dictDelete(server.pubsub_channels,channel); + } + } + /* Notify the client */ + if (notify) { + addReply(c,shared.mbulk3); + addReply(c,shared.unsubscribebulk); + addReplyBulk(c,channel); + addReplyLong(c,dictSize(c->pubsub_channels)+ + listLength(c->pubsub_patterns)); + + } + decrRefCount(channel); /* it is finally safe to release it */ + return retval; +} - /* Iterate this DB writing every entry */ - while((de = dictNext(di)) != NULL) { - robj *key = dictGetEntryKey(de); - robj *o = dictGetEntryVal(de); - time_t expiretime = getExpire(db,key); +/* Subscribe a client to a pattern. Returns 1 if the operation succeeded, or 0 if the clinet was already subscribed to that pattern. */ +static int pubsubSubscribePattern(redisClient *c, robj *pattern) { + int retval = 0; + + if (listSearchKey(c->pubsub_patterns,pattern) == NULL) { + retval = 1; + pubsubPattern *pat; + listAddNodeTail(c->pubsub_patterns,pattern); + incrRefCount(pattern); + pat = zmalloc(sizeof(*pat)); + pat->pattern = getDecodedObject(pattern); + pat->client = c; + listAddNodeTail(server.pubsub_patterns,pat); + } + /* Notify the client */ + addReply(c,shared.mbulk3); + addReply(c,shared.psubscribebulk); + addReplyBulk(c,pattern); + addReplyLong(c,dictSize(c->pubsub_channels)+listLength(c->pubsub_patterns)); + return retval; +} - /* Save the key and associated value */ - if (o->type == REDIS_STRING) { - /* Emit a SET command */ - char cmd[]="*3\r\n$3\r\nSET\r\n"; - if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr; - /* Key and value */ - if (fwriteBulk(fp,key) == 0) goto werr; - if (fwriteBulk(fp,o) == 0) goto werr; - } else if (o->type == REDIS_LIST) { - /* Emit the RPUSHes needed to rebuild the list */ - list *list = o->ptr; - listNode *ln; +/* Unsubscribe a client from a channel. Returns 1 if the operation succeeded, or + * 0 if the client was not subscribed to the specified channel. */ +static int pubsubUnsubscribePattern(redisClient *c, robj *pattern, int notify) { + listNode *ln; + pubsubPattern pat; + int retval = 0; + + incrRefCount(pattern); /* Protect the object. May be the same we remove */ + if ((ln = listSearchKey(c->pubsub_patterns,pattern)) != NULL) { + retval = 1; + listDelNode(c->pubsub_patterns,ln); + pat.client = c; + pat.pattern = pattern; + ln = listSearchKey(server.pubsub_patterns,&pat); + listDelNode(server.pubsub_patterns,ln); + } + /* Notify the client */ + if (notify) { + addReply(c,shared.mbulk3); + addReply(c,shared.punsubscribebulk); + addReplyBulk(c,pattern); + addReplyLong(c,dictSize(c->pubsub_channels)+ + listLength(c->pubsub_patterns)); + } + decrRefCount(pattern); + return retval; +} - listRewind(list); - while((ln = listYield(list))) { - char cmd[]="*3\r\n$5\r\nRPUSH\r\n"; - robj *eleobj = listNodeValue(ln); +/* Unsubscribe from all the channels. Return the number of channels the + * client was subscribed from. */ +static int pubsubUnsubscribeAllChannels(redisClient *c, int notify) { + dictIterator *di = dictGetIterator(c->pubsub_channels); + dictEntry *de; + int count = 0; - if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr; - if (fwriteBulk(fp,key) == 0) goto werr; - if (fwriteBulk(fp,eleobj) == 0) goto werr; - } - } else if (o->type == REDIS_SET) { - /* Emit the SADDs needed to rebuild the set */ - dict *set = o->ptr; - dictIterator *di = dictGetIterator(set); - dictEntry *de; + while((de = dictNext(di)) != NULL) { + robj *channel = dictGetEntryKey(de); - while((de = dictNext(di)) != NULL) { - char cmd[]="*3\r\n$4\r\nSADD\r\n"; - robj *eleobj = dictGetEntryKey(de); + count += pubsubUnsubscribeChannel(c,channel,notify); + } + dictReleaseIterator(di); + return count; +} - if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr; - if (fwriteBulk(fp,key) == 0) goto werr; - if (fwriteBulk(fp,eleobj) == 0) goto werr; - } - dictReleaseIterator(di); - } else if (o->type == REDIS_ZSET) { - /* Emit the ZADDs needed to rebuild the sorted set */ - zset *zs = o->ptr; - dictIterator *di = dictGetIterator(zs->dict); - dictEntry *de; +/* Unsubscribe from all the patterns. Return the number of patterns the + * client was subscribed from. */ +static int pubsubUnsubscribeAllPatterns(redisClient *c, int notify) { + listNode *ln; + listIter li; + int count = 0; - while((de = dictNext(di)) != NULL) { - char cmd[]="*4\r\n$4\r\nZADD\r\n"; - robj *eleobj = dictGetEntryKey(de); - double *score = dictGetEntryVal(de); + listRewind(c->pubsub_patterns,&li); + while ((ln = listNext(&li)) != NULL) { + robj *pattern = ln->value; - if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr; - if (fwriteBulk(fp,key) == 0) goto werr; - if (fwriteBulkDouble(fp,*score) == 0) goto werr; - if (fwriteBulk(fp,eleobj) == 0) goto werr; - } - dictReleaseIterator(di); - } else { - redisAssert(0 != 0); - } - /* Save the expire time */ - if (expiretime != -1) { - char cmd[]="*3\r\n$8\r\nEXPIREAT\r\n"; - /* If this key is already expired skip it */ - if (expiretime < now) continue; - if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr; - if (fwriteBulk(fp,key) == 0) goto werr; - if (fwriteBulkLong(fp,expiretime) == 0) goto werr; - } - } - dictReleaseIterator(di); + count += pubsubUnsubscribePattern(c,pattern,notify); } + return count; +} - /* Make sure data will not remain on the OS's output buffers */ - fflush(fp); - fsync(fileno(fp)); - fclose(fp); - - /* Use RENAME to make sure the DB file is changed atomically only - * if the generate DB file is ok. */ - if (rename(tmpfile,filename) == -1) { - redisLog(REDIS_WARNING,"Error moving temp append only file on the final destination: %s", strerror(errno)); - unlink(tmpfile); - return REDIS_ERR; - } - redisLog(REDIS_NOTICE,"SYNC append only file rewrite performed"); - return REDIS_OK; +/* Publish a message */ +static int pubsubPublishMessage(robj *channel, robj *message) { + int receivers = 0; + struct dictEntry *de; + listNode *ln; + listIter li; -werr: - fclose(fp); - unlink(tmpfile); - redisLog(REDIS_WARNING,"Write error writing append only file on disk: %s", strerror(errno)); - if (di) dictReleaseIterator(di); - return REDIS_ERR; + /* Send to clients listening for that channel */ + de = dictFind(server.pubsub_channels,channel); + if (de) { + list *list = dictGetEntryVal(de); + listNode *ln; + listIter li; + + listRewind(list,&li); + while ((ln = listNext(&li)) != NULL) { + redisClient *c = ln->value; + + addReply(c,shared.mbulk3); + addReply(c,shared.messagebulk); + addReplyBulk(c,channel); + addReplyBulk(c,message); + receivers++; + } + } + /* Send to clients listening to matching channels */ + if (listLength(server.pubsub_patterns)) { + listRewind(server.pubsub_patterns,&li); + channel = getDecodedObject(channel); + while ((ln = listNext(&li)) != NULL) { + pubsubPattern *pat = ln->value; + + if (stringmatchlen((char*)pat->pattern->ptr, + sdslen(pat->pattern->ptr), + (char*)channel->ptr, + sdslen(channel->ptr),0)) { + addReply(pat->client,shared.mbulk3); + addReply(pat->client,shared.messagebulk); + addReplyBulk(pat->client,channel); + addReplyBulk(pat->client,message); + receivers++; + } + } + decrRefCount(channel); + } + return receivers; } -/* This is how rewriting of the append only file in background works: - * - * 1) The user calls BGREWRITEAOF - * 2) Redis calls this function, that forks(): - * 2a) the child rewrite the append only file in a temp file. - * 2b) the parent accumulates differences in server.bgrewritebuf. - * 3) When the child finished '2a' exists. - * 4) The parent will trap the exit code, if it's OK, will append the - * data accumulated into server.bgrewritebuf into the temp file, and - * finally will rename(2) the temp file in the actual file name. - * The the new file is reopened as the new append only file. Profit! - */ -static int rewriteAppendOnlyFileBackground(void) { - pid_t childpid; +static void subscribeCommand(redisClient *c) { + int j; - if (server.bgrewritechildpid != -1) return REDIS_ERR; - if ((childpid = fork()) == 0) { - /* Child */ - char tmpfile[256]; - close(server.fd); + for (j = 1; j < c->argc; j++) + pubsubSubscribeChannel(c,c->argv[j]); +} - snprintf(tmpfile,256,"temp-rewriteaof-bg-%d.aof", (int) getpid()); - if (rewriteAppendOnlyFile(tmpfile) == REDIS_OK) { - exit(0); - } else { - exit(1); - } +static void unsubscribeCommand(redisClient *c) { + if (c->argc == 1) { + pubsubUnsubscribeAllChannels(c,1); + return; } else { - /* Parent */ - if (childpid == -1) { - redisLog(REDIS_WARNING, - "Can't rewrite append only file in background: fork: %s", - strerror(errno)); - return REDIS_ERR; - } - redisLog(REDIS_NOTICE, - "Background append only file rewriting started by pid %d",childpid); - server.bgrewritechildpid = childpid; - /* We set appendseldb to -1 in order to force the next call to the - * feedAppendOnlyFile() to issue a SELECT command, so the differences - * accumulated by the parent into server.bgrewritebuf will start - * with a SELECT statement and it will be safe to merge. */ - server.appendseldb = -1; - return REDIS_OK; + int j; + + for (j = 1; j < c->argc; j++) + pubsubUnsubscribeChannel(c,c->argv[j],1); } - return REDIS_OK; /* unreached */ } -static void bgrewriteaofCommand(redisClient *c) { - if (server.bgrewritechildpid != -1) { - addReplySds(c,sdsnew("-ERR background append only file rewriting already in progress\r\n")); +static void psubscribeCommand(redisClient *c) { + int j; + + for (j = 1; j < c->argc; j++) + pubsubSubscribePattern(c,c->argv[j]); +} + +static void punsubscribeCommand(redisClient *c) { + if (c->argc == 1) { + pubsubUnsubscribeAllPatterns(c,1); return; - } - if (rewriteAppendOnlyFileBackground() == REDIS_OK) { - char *status = "+Background append only file rewriting started\r\n"; - addReplySds(c,sdsnew(status)); } else { - addReply(c,shared.err); + int j; + + for (j = 1; j < c->argc; j++) + pubsubUnsubscribePattern(c,c->argv[j],1); } } -static void aofRemoveTempFile(pid_t childpid) { - char tmpfile[256]; - - snprintf(tmpfile,256,"temp-rewriteaof-bg-%d.aof", (int) childpid); - unlink(tmpfile); +static void publishCommand(redisClient *c) { + int receivers = pubsubPublishMessage(c->argv[1],c->argv[2]); + addReplyLong(c,receivers); } /* ================================= Debugging ============================== */ @@ -6476,19 +9837,79 @@ static void debugCommand(redisClient *c) { } key = dictGetEntryKey(de); val = dictGetEntryVal(de); - addReplySds(c,sdscatprintf(sdsempty(), - "+Key at:%p refcount:%d, value at:%p refcount:%d encoding:%d\r\n", + if (!server.vm_enabled || (key->storage == REDIS_VM_MEMORY || + key->storage == REDIS_VM_SWAPPING)) { + char *strenc; + char buf[128]; + + if (val->encoding < (sizeof(strencoding)/sizeof(char*))) { + strenc = strencoding[val->encoding]; + } else { + snprintf(buf,64,"unknown encoding %d\n", val->encoding); + strenc = buf; + } + addReplySds(c,sdscatprintf(sdsempty(), + "+Key at:%p refcount:%d, value at:%p refcount:%d " + "encoding:%s serializedlength:%lld\r\n", (void*)key, key->refcount, (void*)val, val->refcount, - val->encoding)); + strenc, (long long) rdbSavedObjectLen(val,NULL))); + } else { + addReplySds(c,sdscatprintf(sdsempty(), + "+Key at:%p refcount:%d, value swapped at: page %llu " + "using %llu pages\r\n", + (void*)key, key->refcount, (unsigned long long) key->vm.page, + (unsigned long long) key->vm.usedpages)); + } + } else if (!strcasecmp(c->argv[1]->ptr,"swapin") && c->argc == 3) { + lookupKeyRead(c->db,c->argv[2]); + addReply(c,shared.ok); + } else if (!strcasecmp(c->argv[1]->ptr,"swapout") && c->argc == 3) { + dictEntry *de = dictFind(c->db->dict,c->argv[2]); + robj *key, *val; + + if (!server.vm_enabled) { + addReplySds(c,sdsnew("-ERR Virtual Memory is disabled\r\n")); + return; + } + if (!de) { + addReply(c,shared.nokeyerr); + return; + } + key = dictGetEntryKey(de); + val = dictGetEntryVal(de); + /* If the key is shared we want to create a copy */ + if (key->refcount > 1) { + robj *newkey = dupStringObject(key); + decrRefCount(key); + key = dictGetEntryKey(de) = newkey; + } + /* Swap it */ + if (key->storage != REDIS_VM_MEMORY) { + addReplySds(c,sdsnew("-ERR This key is not in memory\r\n")); + } else if (vmSwapObjectBlocking(key,val) == REDIS_OK) { + dictGetEntryVal(de) = NULL; + addReply(c,shared.ok); + } else { + addReply(c,shared.err); + } } else { addReplySds(c,sdsnew( - "-ERR Syntax error, try DEBUG [SEGFAULT|OBJECT |RELOAD]\r\n")); + "-ERR Syntax error, try DEBUG [SEGFAULT|OBJECT |SWAPIN |SWAPOUT |RELOAD]\r\n")); } } -static void _redisAssert(char *estr) { +static void _redisAssert(char *estr, char *file, int line) { redisLog(REDIS_WARNING,"=== ASSERTION FAILED ==="); - redisLog(REDIS_WARNING,"==> %s\n",estr); + redisLog(REDIS_WARNING,"==> %s:%d '%s' is not true\n",file,line,estr); +#ifdef HAVE_BACKTRACE + redisLog(REDIS_WARNING,"(forcing SIGSEGV in order to print the stack trace)"); + *((char*)-1) = 'x'; +#endif +} + +static void _redisPanic(char *msg, char *file, int line) { + redisLog(REDIS_WARNING,"!!! Software Failure. Press left mouse button to continue"); + redisLog(REDIS_WARNING,"Guru Mediation: %s #%s:%d",msg,file,line); #ifdef HAVE_BACKTRACE redisLog(REDIS_WARNING,"(forcing SIGSEGV in order to print the stack trace)"); *((char*)-1) = 'x'; @@ -6524,7 +9945,6 @@ static void daemonize(void) { FILE *fp; if (fork() != 0) exit(0); /* parent exits */ - printf("New pid: %d\n", getpid()); setsid(); /* create a new session */ /* Every output goes to /dev/null. If Redis is daemonized but @@ -6544,14 +9964,29 @@ static void daemonize(void) { } } +static void version() { + printf("Redis server version %s\n", REDIS_VERSION); + exit(0); +} + +static void usage() { + fprintf(stderr,"Usage: ./redis-server [/path/to/redis.conf]\n"); + fprintf(stderr," ./redis-server - (read config from stdin)\n"); + exit(1); +} + int main(int argc, char **argv) { + time_t start; + initServerConfig(); if (argc == 2) { + if (strcmp(argv[1], "-v") == 0 || + strcmp(argv[1], "--version") == 0) version(); + if (strcmp(argv[1], "--help") == 0) usage(); resetServerSaveParams(); loadServerConfig(argv[1]); - } else if (argc > 2) { - fprintf(stderr,"Usage: ./redis-server [/path/to/redis.conf]\n"); - exit(1); + } else if ((argc > 2)) { + usage(); } else { redisLog(REDIS_WARNING,"Warning: no config file specified, using the default config. In order to specify a config file use 'redis-server /path/to/redis.conf'"); } @@ -6561,16 +9996,16 @@ int main(int argc, char **argv) { #ifdef __linux__ linuxOvercommitMemoryWarning(); #endif + start = time(NULL); if (server.appendonly) { if (loadAppendOnlyFile(server.appendfilename) == REDIS_OK) - redisLog(REDIS_NOTICE,"DB loaded from append only file"); + redisLog(REDIS_NOTICE,"DB loaded from append only file: %ld seconds",time(NULL)-start); } else { if (rdbLoad(server.dbfilename) == REDIS_OK) - redisLog(REDIS_NOTICE,"DB loaded from disk"); + redisLog(REDIS_NOTICE,"DB loaded from disk: %ld seconds",time(NULL)-start); } - if (aeCreateFileEvent(server.el, server.fd, AE_READABLE, - acceptHandler, NULL) == AE_ERR) oom("creating file event"); redisLog(REDIS_NOTICE,"The server is now ready to accept connections on port %d", server.port); + aeSetBeforeSleepProc(server.el,beforeSleep); aeMain(server.el); aeDeleteEventLoop(server.el); return 0; @@ -6597,8 +10032,8 @@ static void *getMcontextEip(ucontext_t *uc) { return (void*) uc->uc_mcontext->__ss.__rip; #else return (void*) uc->uc_mcontext->__ss.__eip; - #endif -#elif defined(__i386__) || defined(__X86_64__) || defined(__x86_64__) + #endif +#elif defined(__i386__) || defined(__X86_64__) || defined(__x86_64__) return (void*) uc->uc_mcontext.gregs[REG_EIP]; /* Linux 32/64 bit */ #elif defined(__ia64__) /* Linux IA64 */ return (void*) uc->uc_mcontext.sc_ip; @@ -6622,7 +10057,7 @@ static void segvHandler(int sig, siginfo_t *info, void *secret) { redisLog(REDIS_WARNING, "%s",infostring); /* It's not safe to sdsfree() the returned string under memory * corruption conditions. Let it leak as we are going to abort */ - + trace_size = backtrace(trace, 100); /* overwrite sigaction with caller's address */ if (getMcontextEip(uc) != NULL) { @@ -6640,8 +10075,8 @@ static void segvHandler(int sig, siginfo_t *info, void *secret) { redisLog(REDIS_WARNING,"%d redis-server %p %s + %d", i, trace[i], fn, (unsigned int)offset); } } - // free(messages); Don't call free() with possibly corrupted memory. - exit(0); + /* free(messages); Don't call free() with possibly corrupted memory. */ + _exit(0); } static void setupSigSegvAction(void) {