X-Git-Url: https://git.saurik.com/redis.git/blobdiff_plain/b892cabed4079de8a1c6ac18041e3ea835c80300..ddfaca9d81cdebf4ced1cd65c42983edc052ab57:/redis.c diff --git a/redis.c b/redis.c index df3b4131..c6283e49 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 "1.050" +#define REDIS_VERSION "1.3.4" #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" @@ -152,11 +154,31 @@ #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_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 */ @@ -184,8 +206,9 @@ /* 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) @@ -198,23 +221,70 @@ #define APPENDFSYNC_ALWAYS 1 #define APPENDFSYNC_EVERYSEC 2 +/* We can print the stacktrace, so our assert is defined this way: */ +#define redisAssert(_e) ((_e)?(void)0 : (_redisAssert(#_e,__FILE__,__LINE__),_exit(1))) +static void _redisAssert(char *estr, 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. + * Note that this macro is taken near the structure definition to make sure + * we'll update it when the structure is changed, to avoid bugs like + * bug #85 introduced exactly in this way. */ +#define initStaticStringObject(_var,_ptr) do { \ + _var.refcount = 1; \ + _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; - dict *expires; + 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; +/* Client MULTI/EXEC state */ +typedef struct multiCmd { + robj **argv; + int argc; + struct redisCommand *cmd; +} multiCmd; + +typedef struct multiState { + multiCmd *commands; /* Array of MULTI commands */ + int count; /* Total number of MULTI commands */ +} multiState; + /* With multiplexing we need to take per-clinet state. * Clients are taken in a liked list. */ typedef struct redisClient { @@ -229,13 +299,21 @@ 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 */ + 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 */ int repldbfd; /* replication DB file descriptor */ - long repldboff; /* replication DB file offset */ + long repldboff; /* replication DB file offset */ off_t repldbsize; /* replication DB file size */ + multiState mstate; /* MULTI/EXEC state */ + 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. */ } redisClient; struct saveparam { @@ -248,7 +326,7 @@ struct redisServer { int port; int fd; redisDb *db; - dict *sharingpool; + dict *sharingpool; /* Poll used for object sharing */ unsigned int sharingpoolsize; long long dirty; /* changes to DB from the last save */ list *clients; @@ -258,7 +336,6 @@ 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 */ @@ -275,8 +352,9 @@ struct redisServer { int appendfd; int appendseldb; char *pidfile; - int bgsaveinprogress; pid_t bgsavechildpid; + pid_t bgrewritechildpid; + sds bgrewritebuf; /* buffer taken by parent during oppend only rewrite */ struct saveparam *saveparams; int saveparamslen; char *logfile; @@ -285,6 +363,7 @@ struct redisServer { char *appendfilename; char *requirepass; int shareobjects; + int rdbcompression; /* Replication related */ int isslave; char *masterauth; @@ -293,12 +372,53 @@ 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; + /* 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; + FILE *devnull; }; typedef void redisCommandProc(redisClient *c); @@ -307,6 +427,10 @@ struct redisCommand { redisCommandProc *proc; int arity; int flags; + /* 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 { @@ -332,6 +456,7 @@ typedef struct _redisSortOperation { typedef struct zskiplistNode { struct zskiplistNode **forward; struct zskiplistNode *backward; + unsigned int *span; double score; robj *obj; } zskiplistNode; @@ -351,7 +476,7 @@ typedef struct zset { struct sharedObjectsStruct { robj *crlf, *ok, *err, *emptybulk, *czero, *cone, *pong, *space, - *colon, *nullbulk, *nullmultibulk, + *colon, *nullbulk, *nullmultibulk, *queued, *emptymultibulk, *wrongtypeerr, *nokeyerr, *syntaxerr, *sameobjecterr, *outofrangeerr, *plus, *select0, *select1, *select2, *select3, *select4, @@ -364,6 +489,22 @@ 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 safe 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); @@ -378,15 +519,17 @@ 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 robj *dupStringObject(robj *o); static void replicationFeedSlaves(list *slaves, struct redisCommand *cmd, 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 *getDecodedObject(const 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); @@ -395,12 +538,46 @@ static void freeMemoryIfNeeded(void); static int processCommand(redisClient *c); static void setupSigSegvAction(void); static void rdbRemoveTempFile(pid_t childpid); +static void aofRemoveTempFile(pid_t childpid); static size_t stringObjectLen(robj *o); static void processInputBuffer(redisClient *c); static zskiplist *zslCreate(void); static void zslFree(zskiplist *zsl); static void zslInsert(zskiplist *zsl, double score, robj *obj); static void sendReplyToClientWritev(aeEventLoop *el, int fd, void *privdata, int mask); +static void initClientMultiState(redisClient *c); +static void freeClientMultiState(redisClient *c); +static void queueMultiCommand(redisClient *c, struct redisCommand *cmd); +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 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 authCommand(redisClient *c); static void pingCommand(redisClient *c); @@ -421,6 +598,7 @@ static void dbsizeCommand(redisClient *c); static void lastsaveCommand(redisClient *c); static void saveCommand(redisClient *c); static void bgsaveCommand(redisClient *c); +static void bgrewriteaofCommand(redisClient *c); static void shutdownCommand(redisClient *c); static void moveCommand(redisClient *c); static void renameCommand(redisClient *c); @@ -466,92 +644,113 @@ static void debugCommand(redisClient *c); static void msetCommand(redisClient *c); static void msetnxCommand(redisClient *c); 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); 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); /*================================= 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}, - {"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}, - {"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}, - {"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}, - {"shutdown",shutdownCommand,1,REDIS_CMD_INLINE}, - {"lastsave",lastsaveCommand,1,REDIS_CMD_INLINE}, - {"type",typeCommand,2,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,1,1,1}, + {"set",setCommand,3,REDIS_CMD_BULK|REDIS_CMD_DENYOOM,0,0,0}, + {"setnx",setnxCommand,3,REDIS_CMD_BULK|REDIS_CMD_DENYOOM,0,0,0}, + {"append",appendCommand,3,REDIS_CMD_BULK|REDIS_CMD_DENYOOM,1,1,1}, + {"substr",substrCommand,4,REDIS_CMD_INLINE,1,1,1}, + {"del",delCommand,-2,REDIS_CMD_INLINE,0,0,0}, + {"exists",existsCommand,2,REDIS_CMD_INLINE,1,1,1}, + {"incr",incrCommand,2,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM,1,1,1}, + {"decr",decrCommand,2,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM,1,1,1}, + {"mget",mgetCommand,-2,REDIS_CMD_INLINE,1,-1,1}, + {"rpush",rpushCommand,3,REDIS_CMD_BULK|REDIS_CMD_DENYOOM,1,1,1}, + {"lpush",lpushCommand,3,REDIS_CMD_BULK|REDIS_CMD_DENYOOM,1,1,1}, + {"rpop",rpopCommand,2,REDIS_CMD_INLINE,1,1,1}, + {"lpop",lpopCommand,2,REDIS_CMD_INLINE,1,1,1}, + {"brpop",brpopCommand,-3,REDIS_CMD_INLINE,1,1,1}, + {"blpop",blpopCommand,-3,REDIS_CMD_INLINE,1,1,1}, + {"llen",llenCommand,2,REDIS_CMD_INLINE,1,1,1}, + {"lindex",lindexCommand,3,REDIS_CMD_INLINE,1,1,1}, + {"lset",lsetCommand,4,REDIS_CMD_BULK|REDIS_CMD_DENYOOM,1,1,1}, + {"lrange",lrangeCommand,4,REDIS_CMD_INLINE,1,1,1}, + {"ltrim",ltrimCommand,4,REDIS_CMD_INLINE,1,1,1}, + {"lrem",lremCommand,4,REDIS_CMD_BULK,1,1,1}, + {"rpoplpush",rpoplpushcommand,3,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM,1,2,1}, + {"sadd",saddCommand,3,REDIS_CMD_BULK|REDIS_CMD_DENYOOM,1,1,1}, + {"srem",sremCommand,3,REDIS_CMD_BULK,1,1,1}, + {"smove",smoveCommand,4,REDIS_CMD_BULK,1,2,1}, + {"sismember",sismemberCommand,3,REDIS_CMD_BULK,1,1,1}, + {"scard",scardCommand,2,REDIS_CMD_INLINE,1,1,1}, + {"spop",spopCommand,2,REDIS_CMD_INLINE,1,1,1}, + {"srandmember",srandmemberCommand,2,REDIS_CMD_INLINE,1,1,1}, + {"sinter",sinterCommand,-2,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM,1,-1,1}, + {"sinterstore",sinterstoreCommand,-3,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM,2,-1,1}, + {"sunion",sunionCommand,-2,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM,1,-1,1}, + {"sunionstore",sunionstoreCommand,-3,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM,2,-1,1}, + {"sdiff",sdiffCommand,-2,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM,1,-1,1}, + {"sdiffstore",sdiffstoreCommand,-3,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM,2,-1,1}, + {"smembers",sinterCommand,2,REDIS_CMD_INLINE,1,1,1}, + {"zadd",zaddCommand,4,REDIS_CMD_BULK|REDIS_CMD_DENYOOM,1,1,1}, + {"zincrby",zincrbyCommand,4,REDIS_CMD_BULK|REDIS_CMD_DENYOOM,1,1,1}, + {"zrem",zremCommand,3,REDIS_CMD_BULK,1,1,1}, + {"zremrangebyscore",zremrangebyscoreCommand,4,REDIS_CMD_INLINE,1,1,1}, + {"zrange",zrangeCommand,-4,REDIS_CMD_INLINE,1,1,1}, + {"zrangebyscore",zrangebyscoreCommand,-4,REDIS_CMD_INLINE,1,1,1}, + {"zcount",zcountCommand,4,REDIS_CMD_INLINE,1,1,1}, + {"zrevrange",zrevrangeCommand,-4,REDIS_CMD_INLINE,1,1,1}, + {"zcard",zcardCommand,2,REDIS_CMD_INLINE,1,1,1}, + {"zscore",zscoreCommand,3,REDIS_CMD_BULK|REDIS_CMD_DENYOOM,1,1,1}, + {"zrank",zrankCommand,3,REDIS_CMD_INLINE,1,1,1}, + {"incrby",incrbyCommand,3,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM,1,1,1}, + {"decrby",decrbyCommand,3,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM,1,1,1}, + {"getset",getsetCommand,3,REDIS_CMD_BULK|REDIS_CMD_DENYOOM,1,1,1}, + {"mset",msetCommand,-3,REDIS_CMD_BULK|REDIS_CMD_DENYOOM,1,-1,2}, + {"msetnx",msetnxCommand,-3,REDIS_CMD_BULK|REDIS_CMD_DENYOOM,1,-1,2}, + {"randomkey",randomkeyCommand,1,REDIS_CMD_INLINE,0,0,0}, + {"select",selectCommand,2,REDIS_CMD_INLINE,0,0,0}, + {"move",moveCommand,3,REDIS_CMD_INLINE,1,1,1}, + {"rename",renameCommand,3,REDIS_CMD_INLINE,1,1,1}, + {"renamenx",renamenxCommand,3,REDIS_CMD_INLINE,1,1,1}, + {"expire",expireCommand,3,REDIS_CMD_INLINE,0,0,0}, + {"expireat",expireatCommand,3,REDIS_CMD_INLINE,0,0,0}, + {"keys",keysCommand,2,REDIS_CMD_INLINE,0,0,0}, + {"dbsize",dbsizeCommand,1,REDIS_CMD_INLINE,0,0,0}, + {"auth",authCommand,2,REDIS_CMD_INLINE,0,0,0}, + {"ping",pingCommand,1,REDIS_CMD_INLINE,0,0,0}, + {"echo",echoCommand,2,REDIS_CMD_BULK,0,0,0}, + {"save",saveCommand,1,REDIS_CMD_INLINE,0,0,0}, + {"bgsave",bgsaveCommand,1,REDIS_CMD_INLINE,0,0,0}, + {"bgrewriteaof",bgrewriteaofCommand,1,REDIS_CMD_INLINE,0,0,0}, + {"shutdown",shutdownCommand,1,REDIS_CMD_INLINE,0,0,0}, + {"lastsave",lastsaveCommand,1,REDIS_CMD_INLINE,0,0,0}, + {"type",typeCommand,2,REDIS_CMD_INLINE,1,1,1}, + {"multi",multiCommand,1,REDIS_CMD_INLINE,0,0,0}, + {"exec",execCommand,1,REDIS_CMD_INLINE,0,0,0}, + {"discard",discardCommand,1,REDIS_CMD_INLINE,0,0,0}, + {"sync",syncCommand,1,REDIS_CMD_INLINE,0,0,0}, + {"flushdb",flushdbCommand,1,REDIS_CMD_INLINE,0,0,0}, + {"flushall",flushallCommand,1,REDIS_CMD_INLINE,0,0,0}, + {"sort",sortCommand,-2,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM,1,1,1}, + {"info",infoCommand,1,REDIS_CMD_INLINE,0,0,0}, + {"monitor",monitorCommand,1,REDIS_CMD_INLINE,0,0,0}, + {"ttl",ttlCommand,2,REDIS_CMD_INLINE,1,1,1}, + {"slaveof",slaveofCommand,3,REDIS_CMD_INLINE,0,0,0}, + {"debug",debugCommand,-2,REDIS_CMD_INLINE,0,0,0}, + {NULL,NULL,0,0,0,0,0} }; /*============================ Utility functions ============================ */ @@ -688,13 +887,13 @@ 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); @@ -716,6 +915,12 @@ static void dictVanillaFree(void *privdata, void *val) zfree(val); } +static void dictListDestructor(void *privdata, void *val) +{ + DICT_NOTUSED(privdata); + listRelease((list*)val); +} + static int sdsDictKeyCompare(void *privdata, const void *key1, const void *key2) { @@ -732,6 +937,7 @@ 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); } @@ -750,39 +956,41 @@ static unsigned int dictObjHash(const void *key) { static int dictEncObjKeyCompare(void *privdata, const void *key1, const void *key2) { - const robj *o1 = key1, *o2 = key2; - - if (o1->encoding == REDIS_ENCODING_RAW && - o2->encoding == REDIS_ENCODING_RAW) - return sdsDictKeyCompare(privdata,o1->ptr,o2->ptr); - else { - robj *dec1, *dec2; - int cmp; + robj *o1 = (robj*) key1, *o2 = (robj*) key2; + int cmp; - dec1 = o1->encoding != REDIS_ENCODING_RAW ? - getDecodedObject(o1) : (robj*)o1; - dec2 = o2->encoding != REDIS_ENCODING_RAW ? - getDecodedObject(o2) : (robj*)o2; - cmp = sdsDictKeyCompare(privdata,dec1->ptr,dec2->ptr); - if (dec1 != o1) decrRefCount(dec1); - if (dec2 != o2) decrRefCount(dec2); - return cmp; - } + o1 = getDecodedObject(o1); + o2 = getDecodedObject(o2); + cmp = sdsDictKeyCompare(privdata,o1->ptr,o2->ptr); + decrRefCount(o1); + decrRefCount(o2); + return cmp; } static unsigned int dictEncObjHash(const void *key) { - const robj *o = key; + robj *o = (robj*) key; - if (o->encoding == REDIS_ENCODING_RAW) + if (o->encoding == REDIS_ENCODING_RAW) { return dictGenHashFunction(o->ptr, sdslen((sds)o->ptr)); - else { - robj *dec = getDecodedObject(o); - unsigned int hash = dictGenHashFunction(dec->ptr, sdslen((sds)dec->ptr)); - decrRefCount(dec); - return hash; + } 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 */ @@ -792,15 +1000,17 @@ 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 */ NULL, /* val dup */ dictEncObjKeyCompare, /* key compare */ dictRedisObjectDestructor, /* key destructor */ - dictVanillaFree /* val destructor */ + dictVanillaFree /* val destructor of malloc(sizeof(double)) */ }; +/* Db->dict */ static dictType hashDictType = { dictObjHash, /* hash function */ NULL, /* key dup */ @@ -810,6 +1020,28 @@ 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 */ +}; + +/* Keylist hash table type has unencoded redis objects as keys and + * 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 */ + NULL, /* val dup */ + dictObjKeyCompare, /* key compare */ + dictRedisObjectDestructor, /* key destructor */ + dictListDestructor /* val destructor */ +}; + /* ========================= Random utility functions ======================= */ /* Redis generally does not try to recover from out of memory conditions @@ -818,8 +1050,7 @@ static dictType hashDictType = { * is based on heap allocation for send buffers, so we simply abort. * At least the code will be simpler to read... */ static void oom(const char *msg) { - fprintf(stderr, "%s: Out of memory\n",msg); - fflush(stderr); + redisLog(REDIS_WARNING, "%s: Out of memory\n",msg); sleep(1); abort(); } @@ -829,15 +1060,23 @@ 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"); + (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); + } } } } @@ -858,23 +1097,110 @@ static void tryResizeHashTables(void) { 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); + redisLog(REDIS_VERBOSE,"The hash table %d is too sparse, resize it...",j); dictResize(server.db[j].dict); - redisLog(REDIS_DEBUG,"Hash table %d resized.",j); + redisLog(REDIS_VERBOSE,"Hash table %d resized.",j); } if (htNeedsResize(server.db[j].expires)) dictResize(server.db[j].expires); } } +/* A background saving child (BGSAVE) terminated its work. Handle this. */ +void backgroundSaveDoneHandler(int statloc) { + int exitcode = WEXITSTATUS(statloc); + int bysignal = WIFSIGNALED(statloc); + + if (!bysignal && exitcode == 0) { + redisLog(REDIS_NOTICE, + "Background saving terminated with success"); + server.dirty = 0; + server.lastsave = time(NULL); + } else if (!bysignal && exitcode != 0) { + redisLog(REDIS_WARNING, "Background saving error"); + } else { + redisLog(REDIS_WARNING, + "Background saving terminated by signal"); + rdbRemoveTempFile(server.bgsavechildpid); + } + server.bgsavechildpid = -1; + /* Possibly there are slaves waiting for a BGSAVE in order to be served + * (the first stage of SYNC is a bulk transfer of dump.rdb) */ + updateSlavesWaitingBgsave(exitcode == 0 ? REDIS_OK : REDIS_ERR); +} + +/* A background append only file rewriting (BGREWRITEAOF) terminated its work. + * Handle this. */ +void backgroundRewriteDoneHandler(int statloc) { + int exitcode = WEXITSTATUS(statloc); + int bysignal = WIFSIGNALED(statloc); + + if (!bysignal && exitcode == 0) { + int fd; + char tmpfile[256]; + + redisLog(REDIS_NOTICE, + "Background append only file rewriting terminated with success"); + /* Now it's time to flush the differences accumulated by the parent */ + snprintf(tmpfile,256,"temp-rewriteaof-bg-%d.aof", (int) server.bgrewritechildpid); + fd = open(tmpfile,O_WRONLY|O_APPEND); + if (fd == -1) { + redisLog(REDIS_WARNING, "Not able to open the temp append only file produced by the child: %s", strerror(errno)); + goto cleanup; + } + /* Flush our data... */ + if (write(fd,server.bgrewritebuf,sdslen(server.bgrewritebuf)) != + (signed) sdslen(server.bgrewritebuf)) { + redisLog(REDIS_WARNING, "Error or short write trying to flush the parent diff of the append log file in the child temp file: %s", strerror(errno)); + close(fd); + goto cleanup; + } + redisLog(REDIS_NOTICE,"Parent diff flushed into the new append log file with success (%lu bytes)",sdslen(server.bgrewritebuf)); + /* Now our work is to rename the temp file into the stable file. And + * switch the file descriptor used by the server for append only. */ + if (rename(tmpfile,server.appendfilename) == -1) { + redisLog(REDIS_WARNING,"Can't rename the temp append only file into the stable one: %s", strerror(errno)); + close(fd); + goto cleanup; + } + /* Mission completed... almost */ + redisLog(REDIS_NOTICE,"Append only file successfully rewritten."); + if (server.appendfd != -1) { + /* If append only is actually enabled... */ + close(server.appendfd); + server.appendfd = fd; + fsync(fd); + server.appendseldb = -1; /* Make sure it will issue SELECT */ + redisLog(REDIS_NOTICE,"The new append only file was selected for future appends."); + } else { + /* If append only is disabled we just generate a dump in this + * format. Why not? */ + close(fd); + } + } else if (!bysignal && exitcode != 0) { + redisLog(REDIS_WARNING, "Background append only file rewriting error"); + } else { + redisLog(REDIS_WARNING, + "Background append only file rewriting terminated by signal"); + } +cleanup: + sdsfree(server.bgrewritebuf); + server.bgrewritebuf = sdsempty(); + aofRemoveTempFile(server.bgrewritechildpid); + server.bgrewritechildpid = -1; +} + 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++) { @@ -884,7 +1210,7 @@ static int serverCron(struct aeEventLoop *eventLoop, long long id, void *clientD 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); + redisLog(REDIS_VERBOSE,"DB %d: %lld keys (%lld volatile) in %lld slots HT.",j,used,vkeys,size); /* dictPrintStats(server.dict); */ } } @@ -895,43 +1221,32 @@ 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.bgsaveinprogress) tryResizeHashTables(); + if (server.bgsavechildpid == -1) tryResizeHashTables(); /* Show information about connected clients */ if (!(loops % 5)) { - redisLog(REDIS_DEBUG,"%d clients connected (%d slaves), %zu bytes in use, %d shared objects", + redisLog(REDIS_VERBOSE,"%d clients connected (%d slaves), %zu bytes in use, %d shared objects", listLength(server.clients)-listLength(server.slaves), listLength(server.slaves), - server.usedmemory, + zmalloc_used_memory(), dictSize(server.sharingpool)); } /* Close connections of timedout clients */ - if (server.maxidletime && !(loops % 10)) + if ((server.maxidletime && !(loops % 10)) || server.blpop_blocked_clients) closeTimedoutClients(); - /* Check if a background saving in progress terminated */ - if (server.bgsaveinprogress) { + /* Check if a background saving or AOF rewrite in progress terminated */ + if (server.bgsavechildpid != -1 || server.bgrewritechildpid != -1) { int statloc; - if (wait3(&statloc,WNOHANG,NULL)) { - int exitcode = WEXITSTATUS(statloc); - int bysignal = WIFSIGNALED(statloc); - - if (!bysignal && exitcode == 0) { - redisLog(REDIS_NOTICE, - "Background saving terminated with success"); - server.dirty = 0; - server.lastsave = time(NULL); - } else if (!bysignal && exitcode != 0) { - redisLog(REDIS_WARNING, "Background saving error"); + pid_t pid; + + if ((pid = wait3(&statloc,WNOHANG,NULL)) != 0) { + if (pid == server.bgsavechildpid) { + backgroundSaveDoneHandler(statloc); } else { - redisLog(REDIS_WARNING, - "Background saving terminated by signal"); - rdbRemoveTempFile(server.bgsavechildpid); + backgroundRewriteDoneHandler(statloc); } - server.bgsaveinprogress = 0; - server.bgsavechildpid = -1; - updateSlavesWaitingBgsave(exitcode == 0 ? REDIS_OK : REDIS_ERR); } } else { /* If there is not a background saving in progress check if @@ -961,7 +1276,7 @@ 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; @@ -981,6 +1296,32 @@ static int serverCron(struct aeEventLoop *eventLoop, long long id, void *clientD } 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 % 30) == 0 && + 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) { redisLog(REDIS_NOTICE,"Connecting to MASTER..."); @@ -991,6 +1332,38 @@ static int serverCron(struct aeEventLoop *eventLoop, long long id, void *clientD return 1000; } +/* 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) { shared.crlf = createObject(REDIS_STRING,sdsnew("\r\n")); shared.ok = createObject(REDIS_STRING,sdsnew("+OK\r\n")); @@ -1001,8 +1374,8 @@ static void createSharedObjects(void) { shared.nullbulk = createObject(REDIS_STRING,sdsnew("$-1\r\n")); shared.nullmultibulk = createObject(REDIS_STRING,sdsnew("*-1\r\n")); shared.emptymultibulk = createObject(REDIS_STRING,sdsnew("*0\r\n")); - /* no such key */ shared.pong = createObject(REDIS_STRING,sdsnew("+PONG\r\n")); + shared.queued = createObject(REDIS_STRING,sdsnew("+QUEUED\r\n")); shared.wrongtypeerr = createObject(REDIS_STRING,sdsnew( "-ERR Operation against a key holding the wrong kind of value\r\n")); shared.nokeyerr = createObject(REDIS_STRING,sdsnew( @@ -1044,7 +1417,7 @@ 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 */ @@ -1058,12 +1431,22 @@ static void initServerConfig() { server.appendseldb = -1; /* Make sure the first time will not match */ server.pidfile = "/var/run/redis.pid"; server.dbfilename = "dump.rdb"; - server.appendfilename = "appendonly.log"; + server.appendfilename = "appendonly.aof"; server.requirepass = NULL; server.shareobjects = 0; + server.rdbcompression = 1; server.sharingpoolsize = 1024; 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; + resetServerSaveParams(); appendServerSaveParams(60*60,1); /* save after 1 hour and 1 change */ @@ -1091,6 +1474,11 @@ 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(); @@ -1106,19 +1494,25 @@ static void initServer() { } for (j = 0; j < server.dbnum; j++) { server.db[j].dict = dictCreate(&hashDictType,NULL); - server.db[j].expires = dictCreate(&setDictType,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.cronloops = 0; - server.bgsaveinprogress = 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_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); @@ -1128,6 +1522,8 @@ static void initServer() { exit(1); } } + + if (server.vm_enabled) vmInit(); } /* Empty the whole database */ @@ -1212,6 +1608,7 @@ 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 { @@ -1260,6 +1657,10 @@ static void loadServerConfig(char *filename) { if ((server.shareobjects = yesnotoi(argv[1])) == -1) { err = "argument must be 'yes' or 'no'"; goto loaderr; } + } else if (!strcasecmp(argv[0],"rdbcompression") && argc == 2) { + 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) { @@ -1285,11 +1686,26 @@ 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]); + server.pidfile = zstrdup(argv[1]); } else if (!strcasecmp(argv[0],"dbfilename") && argc == 2) { - server.dbfilename = zstrdup(argv[1]); + 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 { err = "Bad directive or wrong number of arguments"; goto loaderr; } @@ -1323,21 +1739,45 @@ static void freeClientArgv(redisClient *c) { 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 + * 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) + unblockClientWaitingData(c); + aeDeleteFileEvent(server.el,c->fd,AE_READABLE); aeDeleteFileEvent(server.el,c->fd,AE_WRITABLE); - sdsfree(c->querybuf); listRelease(c->reply); freeClientArgv(c); close(c->fd); + /* Remove from the list of clients */ ln = listSearchKey(server.clients,c); - assert(ln != NULL); + 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); + /* Other cleanup */ if (c->flags & REDIS_SLAVE) { if (c->replstate == REDIS_REPL_SEND_BULK && c->repldbfd != -1) close(c->repldbfd); list *l = (c->flags & REDIS_MONITOR) ? server.monitors : server.slaves; ln = listSearchKey(l,c); - assert(ln != NULL); + redisAssert(ln != NULL); listDelNode(l,ln); } if (c->flags & REDIS_MASTER) { @@ -1346,6 +1786,7 @@ static void freeClient(redisClient *c) { } zfree(c->argv); zfree(c->mbargv); + freeClientMultiState(c); zfree(c); } @@ -1354,10 +1795,11 @@ 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; @@ -1383,9 +1825,7 @@ static void sendReplyToClient(aeEventLoop *el, int fd, void *privdata, int mask) REDIS_NOTUSED(el); REDIS_NOTUSED(mask); - /* Use writev() if we have enough buffers to send */ - #if 0 if (!server.glueoutputbuf && listLength(c->reply) > REDIS_WRITEV_THRESHOLD && !(c->flags & REDIS_MASTER)) @@ -1393,7 +1833,6 @@ static void sendReplyToClient(aeEventLoop *el, int fd, void *privdata, int mask) sendReplyToClientWritev(el, fd, privdata, mask); return; } - #endif while(listLength(c->reply)) { if (server.glueoutputbuf && listLength(c->reply) > 1) @@ -1432,7 +1871,7 @@ 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; @@ -1485,7 +1924,7 @@ 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; @@ -1539,6 +1978,21 @@ static void resetClient(redisClient *c) { c->multibulk = 0; } +/* Call() is the core of Redis execution of a command */ +static void call(redisClient *c, struct redisCommand *cmd) { + long long dirty; + + dirty = server.dirty; + cmd->proc(c); + if (server.appendonly && server.dirty-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 (listLength(server.monitors)) + replicationFeedSlaves(server.monitors,cmd,c->db->id,c->argv,c->argc); + server.stat_numcommands++; +} + /* If this function gets called we already read a whole * command, argments are in the client argv/argc fields. * processCommand() execute the command or prepare the @@ -1549,7 +2003,6 @@ static void resetClient(redisClient *c) { * if 0 is returned the client was destroied (i.e. after QUIT). */ static int processCommand(redisClient *c) { struct redisCommand *cmd; - long long dirty; /* Free some memory if needed (maxmemory setting) */ if (server.maxmemory) freeMemoryIfNeeded(); @@ -1628,14 +2081,22 @@ 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,sdsnew("-ERR unknown command\r\n")); + addReplySds(c, + sdscatprintf(sdsempty(), "-ERR unknown command '%s'\r\n", + (char*)c->argv[0]->ptr)); resetClient(c); return 1; } else if ((cmd->arity > 0 && cmd->arity != c->argc) || (c->argc < -cmd->arity)) { - addReplySds(c,sdsnew("-ERR wrong number of arguments\r\n")); + addReplySds(c, + sdscatprintf(sdsempty(), + "-ERR wrong number of arguments for '%s' command\r\n", + cmd->name)); resetClient(c); return 1; } else if (server.maxmemory && cmd->flags & REDIS_CMD_DENYOOM && zmalloc_used_memory() > server.maxmemory) { @@ -1643,6 +2104,7 @@ static int processCommand(redisClient *c) { 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]); @@ -1664,6 +2126,8 @@ 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; } } @@ -1685,27 +2149,23 @@ static int processCommand(redisClient *c) { } /* Exec the command */ - dirty = server.dirty; - cmd->proc(c); - if (server.appendonly && server.dirty-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 (listLength(server.monitors)) - replicationFeedSlaves(server.monitors,cmd,c->db->id,c->argv,c->argc); - server.stat_numcommands++; + 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) { listNode *ln; + listIter li; int outc = 0, j; robj **outv; /* (args*2)+1 is enough room for args, spaces, newlines */ @@ -1723,8 +2183,8 @@ static void replicationFeedSlaves(list *slaves, struct redisCommand *cmd, int di robj *lenobj; lenobj = createObject(REDIS_STRING, - sdscatprintf(sdsempty(),"%d\r\n", - stringObjectLen(argv[j]))); + sdscatprintf(sdsempty(),"%lu\r\n", + (unsigned long) stringObjectLen(argv[j]))); lenobj->refcount = 0; outv[outc++] = lenobj; } @@ -1736,8 +2196,8 @@ static void replicationFeedSlaves(list *slaves, struct redisCommand *cmd, int di * 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 */ @@ -1775,6 +2235,13 @@ static void replicationFeedSlaves(list *slaves, struct redisCommand *cmd, int di static void processInputBuffer(redisClient *c) { again: + /* Before to process the input buffer, make sure the client is not + * waitig for a blocking operation such as BLPOP. Note that the first + * iteration the client is never blocked, otherwise the processInputBuffer + * 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 || c->flags & REDIS_IO_WAIT) return; if (c->bulklen == -1) { /* Read the first line of the query */ char *p = strchr(c->querybuf,'\n'); @@ -1796,11 +2263,6 @@ again: sdsupdatelen(query); /* Now we can split the query in arguments */ - if (sdslen(query) == 0) { - /* Ignore empty query */ - sdsfree(query); - return; - } argv = sdssplitlen(query,sdslen(query)," ",1,&argc); sdsfree(query); @@ -1816,13 +2278,19 @@ again: } } zfree(argv); - /* Execute the command. If the client is still valid - * after processCommand() return and there is something - * on the query buffer try to process the next command. */ - if (c->argc && processCommand(c) && sdslen(c->querybuf)) goto again; + if (c->argc) { + /* Execute the command. If the client is still valid + * after processCommand() return and there is something + * on the query buffer try to process the next command. */ + if (processCommand(c) && sdslen(c->querybuf)) goto again; + } else { + /* Nothing to process, argc == 0. Just process the query + * buffer if it's not empty or return to the caller */ + if (sdslen(c->querybuf)) goto 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; } @@ -1859,12 +2327,12 @@ 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; } @@ -1874,7 +2342,8 @@ static void readQueryFromClient(aeEventLoop *el, int fd, void *privdata, int mas } else { return; } - processInputBuffer(c); + if (!(c->flags & REDIS_BLOCKED)) + processInputBuffer(c); } static int selectDb(redisClient *c, int id) { @@ -1886,7 +2355,7 @@ static int selectDb(redisClient *c, int id) { static void *dupClientReplyValue(void *o) { incrRefCount((robj*)o); - return 0; + return o; } static redisClient *createClient(int fd) { @@ -1912,12 +2381,17 @@ static redisClient *createClient(int fd) { c->reply = listCreate(); listSetFreeMethod(c->reply,decrRefCount); listSetDupMethod(c->reply,dupClientReplyValue); + c->blockingkeys = NULL; + c->blockingkeysnum = 0; + c->io_keys = listCreate(); + listSetFreeMethod(c->io_keys,decrRefCount); if (aeCreateFileEvent(server.el, c->fd, AE_READABLE, - readQueryFromClient, c, NULL) == AE_ERR) { + readQueryFromClient, c) == AE_ERR) { freeClient(c); return NULL; } listAddNodeTail(server.clients,c); + initClientMultiState(c); return c; } @@ -1926,13 +2400,13 @@ static void addReply(redisClient *c, robj *obj) { (c->replstate == REDIS_REPL_NONE || c->replstate == REDIS_REPL_ONLINE) && aeCreateFileEvent(server.el, c->fd, AE_WRITABLE, - sendReplyToClient, c, NULL) == AE_ERR) return; - if (obj->encoding != REDIS_ENCODING_RAW) { - obj = getDecodedObject(obj); - } else { - incrRefCount(obj); + 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,obj); + listAddNodeTail(c->reply,getDecodedObject(obj)); } static void addReplySds(redisClient *c, sds s) { @@ -1941,6 +2415,22 @@ static void addReplySds(redisClient *c, sds s) { decrRefCount(o); } +static void addReplyDouble(redisClient *c, double d) { + char buf[128]; + + snprintf(buf,sizeof(buf),"%.17g",d); + addReplySds(c,sdscatprintf(sdsempty(),"$%lu\r\n%s\r\n", + (unsigned long) strlen(buf),buf)); +} + +static void addReplyLong(redisClient *c, long l) { + char buf[128]; + size_t len; + + len = snprintf(buf,sizeof(buf),":%ld\r\n",l); + addReplySds(c,sdsnewlen(buf,len)); +} + static void addReplyBulkLen(redisClient *c, robj *obj) { size_t len; @@ -1949,6 +2439,7 @@ static void addReplyBulkLen(redisClient *c, robj *obj) { } else { long n = (long)obj->ptr; + /* Compute how many bytes will take this integer as a radix 10 string */ len = 1; if (n < 0) { len++; @@ -1958,7 +2449,7 @@ static void addReplyBulkLen(redisClient *c, robj *obj) { len++; } } - addReplySds(c,sdscatprintf(sdsempty(),"$%d\r\n",len)); + addReplySds(c,sdscatprintf(sdsempty(),"$%lu\r\n",(unsigned long)len)); } static void acceptHandler(aeEventLoop *el, int fd, void *privdata, int mask) { @@ -1971,10 +2462,10 @@ 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 */ @@ -2002,17 +2493,32 @@ 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; } @@ -2020,6 +2526,11 @@ static robj *createStringObject(char *ptr, size_t len) { return createObject(REDIS_STRING,sdsnewlen(ptr,len)); } +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(); @@ -2067,38 +2578,86 @@ static void freeHashObject(robj *o) { } static void incrRefCount(robj *o) { + redisAssert(!server.vm_enabled || o->storage == REDIS_VM_MEMORY); 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 + /* 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_SWAPPED || o->storage == REDIS_VM_LOADING) { + redisAssert(o->refcount == 1); + } + 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: assert(0 != 0); break; + default: redisAssert(0 != 0); 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) { @@ -2133,7 +2692,7 @@ static robj *tryObjectSharing(robj *o) { if (o == NULL || server.shareobjects == 0) return o; - assert(o->type == REDIS_STRING); + redisAssert(o->type == REDIS_STRING); de = dictFind(server.sharingpool,o); if (de) { robj *shared = dictGetEntryKey(de); @@ -2151,7 +2710,7 @@ static robj *tryObjectSharing(robj *o) { if (dictSize(server.sharingpool) >= server.sharingpoolsize) { de = dictGetRandomKey(server.sharingpool); - assert(de != NULL); + redisAssert(de != NULL); c = ((unsigned long) dictGetEntryVal(de))-1; dictGetEntryVal(de) = (void*) c; if (c == 0) { @@ -2164,7 +2723,7 @@ static robj *tryObjectSharing(robj *o) { int retval; retval = dictAdd(server.sharingpool,o,(void*)1); - assert(retval == DICT_OK); + redisAssert(retval == DICT_OK); incrRefCount(o); } return o; @@ -2207,7 +2766,7 @@ static int tryObjectEncoding(robj *o) { if (o->refcount > 1) return REDIS_ERR; /* Currently we try to encode only strings */ - assert(o->type == REDIS_STRING); + 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; @@ -2219,11 +2778,15 @@ static int tryObjectEncoding(robj *o) { return REDIS_OK; } -/* Get a decoded version of an encoded object (returned as a new object) */ -static robj *getDecodedObject(const robj *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; - assert(o->encoding != REDIS_ENCODING_RAW); + if (o->encoding == REDIS_ENCODING_RAW) { + incrRefCount(o); + return o; + } if (o->type == REDIS_STRING && o->encoding == REDIS_ENCODING_INT) { char buf[32]; @@ -2231,16 +2794,20 @@ static robj *getDecodedObject(const robj *o) { dec = createStringObject(buf,strlen(buf)); return dec; } else { - assert(1 != 1); + redisAssert(1 != 1); } } /* Compare two string objects via strcmp() or alike. * Note that the objects may be integer-encoded. In such a case we * use snprintf() to get a string representation of the numbers on the stack - * and compare the strings, it's much faster than calling getDecodedObject(). */ + * and compare the strings, it's much faster than calling getDecodedObject(). + * + * Important note: if objects are not integer encoded, but binary-safe strings, + * sdscmp() from sds.c will apply memcmp() so this function ca be considered + * binary safe. */ static int compareStringObjects(robj *a, robj *b) { - assert(a->type == REDIS_STRING && b->type == REDIS_STRING); + redisAssert(a->type == REDIS_STRING && b->type == REDIS_STRING); char bufa[128], bufb[128], *astr, *bstr; int bothsds = 1; @@ -2263,7 +2830,7 @@ static int compareStringObjects(robj *a, robj *b) { } static size_t stringObjectLen(robj *o) { - assert(o->type == REDIS_STRING); + redisAssert(o->type == REDIS_STRING); if (o->encoding == REDIS_ENCODING_RAW) { return sdslen(o->ptr); } else { @@ -2273,7 +2840,7 @@ static size_t stringObjectLen(robj *o) { } } -/*============================ DB saving/loading ============================ */ +/*============================ RDB saving/loading =========================== */ static int rdbSaveType(FILE *fp, unsigned char type) { if (fwrite(&type,1,1,fp) == 0) return -1; @@ -2394,7 +2961,7 @@ static int rdbSaveStringObjectRaw(FILE *fp, robj *obj) { /* Try LZF compression - under 20 bytes it's unable to compress even * aaaaaaaaaaaaaaaaaa so skip it */ - if (len > 20) { + if (server.rdbcompression && len > 20) { int retval; retval = rdbSaveLzfStringObject(fp,obj); @@ -2412,16 +2979,20 @@ static int rdbSaveStringObjectRaw(FILE *fp, robj *obj) { /* Like rdbSaveStringObjectRaw() but handle encoded objects */ static int rdbSaveStringObject(FILE *fp, robj *obj) { int retval; - robj *dec; + /* 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) { - dec = getDecodedObject(obj); - retval = rdbSaveStringObjectRaw(fp,dec); - decrRefCount(dec); - return retval; + obj = getDecodedObject(obj); + retval = rdbSaveStringObjectRaw(fp,obj); + decrRefCount(obj); } else { - return rdbSaveStringObjectRaw(fp,obj); + retval = rdbSaveStringObjectRaw(fp,obj); } + return retval; } /* Save a double value. Doubles are saved as strings prefixed by an unsigned @@ -2444,13 +3015,83 @@ static int rdbSaveDoubleValue(FILE *fp, double val) { buf[0] = (val < 0) ? 255 : 254; } else { snprintf((char*)buf+1,sizeof(buf)-1,"%.17g",val); - buf[0] = strlen((char*)buf); + buf[0] = strlen((char*)buf+1); len = buf[0]+1; } if (fwrite(buf,len,1,fp) == 0) return -1; 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 { + redisAssert(0 != 0); + } + 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; @@ -2460,6 +3101,12 @@ 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) { @@ -2494,54 +3141,25 @@ 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 { - assert(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); @@ -2577,14 +3195,16 @@ werr: static int rdbSaveBackground(char *filename) { pid_t childpid; - if (server.bgsaveinprogress) return REDIS_ERR; + 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 */ @@ -2594,7 +3214,6 @@ static int rdbSaveBackground(char *filename) { return REDIS_ERR; } redisLog(REDIS_NOTICE,"Background saving started by pid %d",childpid); - server.bgsaveinprogress = 1; server.bgsavechildpid = childpid; return REDIS_OK; } @@ -2625,35 +3244,29 @@ 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); - } } } @@ -2676,18 +3289,18 @@ static robj *rdbLoadIntegerObject(FILE *fp, int enctype) { val = (int32_t)v; } else { val = 0; /* anti-warning */ - assert(0!=0); + redisAssert(0!=0); } 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; @@ -2700,12 +3313,12 @@ 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: @@ -2713,9 +3326,9 @@ static robj *rdbLoadStringObject(FILE*fp, int rdbver) { case REDIS_RDB_ENC_INT32: return tryObjectSharing(rdbLoadIntegerObject(fp,len)); case REDIS_RDB_ENC_LZF: - return tryObjectSharing(rdbLoadLzfStringObject(fp,rdbver)); + return tryObjectSharing(rdbLoadLzfStringObject(fp)); default: - assert(0!=0); + redisAssert(0!=0); } } @@ -2740,11 +3353,69 @@ static int rdbLoadDoubleValue(FILE *fp, double *val) { case 253: *val = R_Nan; return 0; default: if (fread(buf,len,1,fp) == 0) return -1; + buf[len] = '\0'; sscanf(buf, "%lg", val); return 0; } } +/* 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; + + if (type == REDIS_STRING) { + /* Read string value */ + if ((o = rdbLoadStringObject(fp)) == NULL) return NULL; + 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; + 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,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; + 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 { + redisAssert(0 != 0); + } + return o; +} + static int rdbLoad(char *filename) { FILE *fp; robj *keyobj = NULL; @@ -2754,6 +3425,7 @@ 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; @@ -2765,7 +3437,7 @@ 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; @@ -2783,7 +3455,7 @@ 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); @@ -2794,55 +3466,9 @@ 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 { - assert(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) { @@ -2857,6 +3483,13 @@ 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; @@ -2895,9 +3528,16 @@ static void echoCommand(redisClient *c) { static void setGenericCommand(redisClient *c, int nx) { int retval; + if (nx) deleteIfVolatile(c->db,c->argv[1]); 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 { @@ -2921,24 +3561,31 @@ static void setnxCommand(redisClient *c) { setGenericCommand(c,1); } -static void getCommand(redisClient *c) { +static int getGenericCommand(redisClient *c) { robj *o = lookupKeyRead(c->db,c->argv[1]); if (o == NULL) { addReply(c,shared.nullbulk); + return REDIS_OK; } 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; } } } +static void getCommand(redisClient *c) { + getGenericCommand(c); +} + static void getsetCommand(redisClient *c) { - getCommand(c); + if (getGenericCommand(c) == REDIS_ERR) return; if (dictAdd(c->db->dict,c->argv[1],c->argv[2]) == DICT_ERR) { dictReplace(c->db->dict,c->argv[1],c->argv[2]); } else { @@ -2969,6 +3616,53 @@ static void mgetCommand(redisClient *c) { } } +static void msetGenericCommand(redisClient *c, int nx) { + int j, busykeys = 0; + + if ((c->argc % 2) == 0) { + addReplySds(c,sdsnew("-ERR wrong number of arguments for MSET\r\n")); + return; + } + /* Handle the NX flag. The MSETNX semantic is to return zero and don't + * set nothing at all if at least one already key exists. */ + if (nx) { + for (j = 1; j < c->argc; j += 2) { + if (lookupKeyWrite(c->db,c->argv[j]) != NULL) { + busykeys++; + } + } + } + if (busykeys) { + addReply(c, shared.czero); + return; + } + + for (j = 1; j < c->argc; j += 2) { + int retval; + + 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]); + incrRefCount(c->argv[j+1]); + } else { + incrRefCount(c->argv[j]); + incrRefCount(c->argv[j+1]); + } + removeExpire(c->db,c->argv[j]); + } + server.dirty += (c->argc-1)/2; + addReply(c, nx ? shared.cone : shared.ok); +} + +static void msetCommand(redisClient *c) { + msetGenericCommand(c,0); +} + +static void msetnxCommand(redisClient *c) { + msetGenericCommand(c,1); +} + static void incrDecrCommand(redisClient *c, long long incr) { long long value; int retval; @@ -2988,7 +3682,7 @@ static void incrDecrCommand(redisClient *c, long long incr) { else if (o->encoding == REDIS_ENCODING_INT) value = (long)o->ptr; else - assert(1 != 1); + redisAssert(1 != 1); } } @@ -3026,6 +3720,96 @@ static void decrbyCommand(redisClient *c) { 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); + + o = lookupKeyRead(c->db,c->argv[1]); + if (o == NULL) { + addReply(c,shared.nullbulk); + } else { + if (o->type != REDIS_STRING) { + addReply(c,shared.wrongtypeerr); + } else { + size_t rangelen, strlen; + sds range; + + 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(),"$%lu\r\n",rangelen)); + range = sdsnewlen((char*)o->ptr+start,rangelen); + addReplySds(c,range); + addReply(c,shared.crlf); + decrRefCount(o); + } + } +} + /* ========================= Type agnostic commands ========================= */ static void delCommand(redisClient *c) { @@ -3086,7 +3870,7 @@ static void keysCommand(redisClient *c) { dictEntry *de; sds pattern = c->argv[1]->ptr; int plen = sdslen(pattern); - int numkeys = 0, keyslen = 0; + unsigned long numkeys = 0; robj *lenobj = createObject(REDIS_STRING,NULL); di = dictGetIterator(c->db->dict); @@ -3099,17 +3883,15 @@ static void keysCommand(redisClient *c) { if ((pattern[0] == '*' && pattern[1] == '\0') || stringmatchlen(pattern,plen,key,sdslen(key),0)) { if (expireIfNeeded(c->db,keyobj) == 0) { - if (numkeys != 0) - addReply(c,shared.space); + addReplyBulkLen(c,keyobj); addReply(c,keyobj); + addReply(c,shared.crlf); numkeys++; - keyslen += sdslen(key); } } } dictReleaseIterator(di); - lenobj->ptr = sdscatprintf(sdsempty(),"$%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) { @@ -3143,7 +3925,7 @@ static void typeCommand(redisClient *c) { } static void saveCommand(redisClient *c) { - if (server.bgsaveinprogress) { + if (server.bgsavechildpid != -1) { addReplySds(c,sdsnew("-ERR background save in progress\r\n")); return; } @@ -3155,12 +3937,13 @@ static void saveCommand(redisClient *c) { } static void bgsaveCommand(redisClient *c) { - if (server.bgsaveinprogress) { + if (server.bgsavechildpid != -1) { addReplySds(c,sdsnew("-ERR background save already in progress\r\n")); return; } if (rdbSaveBackground(server.dbfilename) == REDIS_OK) { - addReply(c,shared.ok); + char *status = "+Background saving started\r\n"; + addReplySds(c,sdsnew(status)); } else { addReply(c,shared.err); } @@ -3171,25 +3954,33 @@ static void shutdownCommand(redisClient *c) { /* Kill the saving child if there is a background saving in progress. We want to avoid race conditions, for instance our saving child may overwrite the synchronous saving did by SHUTDOWN. */ - if (server.bgsaveinprogress) { + if (server.bgsavechildpid != -1) { redisLog(REDIS_WARNING,"There is a live saving child. Killing it!"); kill(server.bgsavechildpid,SIGKILL); rdbRemoveTempFile(server.bgsavechildpid); } - /* SYNC SAVE */ - if (rdbSave(server.dbfilename) == REDIS_OK) { - if (server.daemonize) - unlink(server.pidfile); - redisLog(REDIS_WARNING,"%zu bytes used at exit",zmalloc_used_memory()); - redisLog(REDIS_WARNING,"Server exit now, bye bye..."); - exit(1); + 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 { - /* 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")); + /* Snapshotting. Perform a SYNC SAVE and exit */ + if (rdbSave(server.dbfilename) == REDIS_OK) { + if (server.daemonize) + 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")); + } } } @@ -3283,6 +4074,10 @@ 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])) { + addReply(c,shared.cone); + return; + } lobj = createListObject(); list = lobj->ptr; if (where == REDIS_HEAD) { @@ -3298,6 +4093,10 @@ static void pushGenericCommand(redisClient *c, int where) { addReply(c,shared.wrongtypeerr); 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]); @@ -3307,7 +4106,7 @@ 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) { @@ -3487,7 +4286,7 @@ static void ltrimCommand(redisClient *c) { o = lookupKeyWrite(c->db,c->argv[1]); if (o == NULL) { - addReply(c,shared.nokeyerr); + addReply(c,shared.ok); } else { if (o->type != REDIS_LIST) { addReply(c,shared.wrongtypeerr); @@ -3602,20 +4401,24 @@ static void rpoplpushcommand(redisClient *c) { 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) { + if (dobj && dobj->type != REDIS_LIST) { addReply(c,shared.wrongtypeerr); return; } - /* Add the element to the target list */ - dstlist = dobj->ptr; - listAddNodeHead(dstlist,ele); - incrRefCount(ele); + + /* 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 */ addReplyBulkLen(c,ele); @@ -3743,7 +4546,7 @@ static void scardCommand(redisClient *c) { addReply(c,shared.wrongtypeerr); } else { s = o->ptr; - addReplySds(c,sdscatprintf(sdsempty(),":%d\r\n", + addReplySds(c,sdscatprintf(sdsempty(),":%lu\r\n", dictSize(s))); } } @@ -3808,12 +4611,12 @@ static int qsortCompareSetsByCardinality(const void *s1, const void *s2) { return dictSize(*d1)-dictSize(*d2); } -static void sinterGenericCommand(redisClient *c, robj **setskeys, int setsnum, robj *dstkey) { +static void sinterGenericCommand(redisClient *c, robj **setskeys, unsigned long setsnum, robj *dstkey) { dict **dv = zmalloc(sizeof(dict*)*setsnum); dictIterator *di; dictEntry *de; robj *lenobj = NULL, *dstset = NULL; - int j, cardinality = 0; + unsigned long j, cardinality = 0; for (j = 0; j < setsnum; j++) { robj *setobj; @@ -3824,8 +4627,9 @@ static void sinterGenericCommand(redisClient *c, robj **setskeys, int setsnum, r if (!setobj) { zfree(dv); if (dstkey) { - deleteKey(c->db,dstkey); - addReply(c,shared.ok); + if (deleteKey(c->db,dstkey)) + server.dirty++; + addReply(c,shared.czero); } else { addReply(c,shared.nullmultibulk); } @@ -3890,9 +4694,9 @@ static void sinterGenericCommand(redisClient *c, robj **setskeys, int setsnum, r } if (!dstkey) { - lenobj->ptr = sdscatprintf(sdsempty(),"*%d\r\n",cardinality); + lenobj->ptr = sdscatprintf(sdsempty(),"*%lu\r\n",cardinality); } else { - addReplySds(c,sdscatprintf(sdsempty(),":%d\r\n", + addReplySds(c,sdscatprintf(sdsempty(),":%lu\r\n", dictSize((dict*)dstset->ptr))); server.dirty++; } @@ -3994,7 +4798,7 @@ static void sunionDiffGenericCommand(redisClient *c, robj **setskeys, int setsnu if (!dstkey) { decrRefCount(dstset); } else { - addReplySds(c,sdscatprintf(sdsempty(),":%d\r\n", + addReplySds(c,sdscatprintf(sdsempty(),":%lu\r\n", dictSize((dict*)dstset->ptr))); server.dirty++; } @@ -4040,6 +4844,8 @@ 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; @@ -4053,8 +4859,10 @@ static zskiplist *zslCreate(void) { 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; + zsl->header->span[j] = 0; + } zsl->header->backward = NULL; zsl->tail = NULL; return zsl; @@ -4063,6 +4871,7 @@ static zskiplist *zslCreate(void) { static void zslFreeNode(zskiplistNode *node) { decrRefCount(node->obj); zfree(node->forward); + zfree(node->span); zfree(node); } @@ -4070,6 +4879,7 @@ 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]; @@ -4088,15 +4898,21 @@ static int zslRandomLevel(void) { static void zslInsert(zskiplist *zsl, double score, robj *obj) { zskiplistNode *update[ZSKIPLIST_MAXLEVEL], *x; + unsigned int rank[ZSKIPLIST_MAXLEVEL]; int i, level; x = zsl->header; 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 @@ -4105,15 +4921,30 @@ 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; @@ -4141,12 +4972,19 @@ static int zslDelete(zskiplist *zsl, double score, robj *obj) { 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 (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 == zsl->header) ? - NULL : x->backward; + x->forward[0]->backward = x->backward; } else { zsl->tail = x->backward; } @@ -4183,12 +5021,19 @@ static unsigned long zslDeleteRange(zskiplist *zsl, double min, double max, dict zskiplistNode *next; for (i = 0; i < zsl->level; i++) { - if (update[i]->forward[i] != x) break; - update[i]->forward[i] = x->forward[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 == zsl->header) ? - NULL : x->backward; + x->forward[0]->backward = x->backward; } else { zsl->tail = x->backward; } @@ -4220,58 +5065,150 @@ static zskiplistNode *zslFirstWithScore(zskiplist *zsl, double score) { 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 */ -static void zaddCommand(redisClient *c) { +/* This generic command implements both ZADD and ZINCRBY. + * scoreval is the score if the operation is a ZADD (doincrement == 0) or + * the increment if the operation is a ZINCRBY (doincrement == 1). */ +static void zaddGenericCommand(redisClient *c, robj *key, robj *ele, double scoreval, int doincrement) { robj *zsetobj; zset *zs; double *score; - zsetobj = lookupKeyWrite(c->db,c->argv[1]); + zsetobj = lookupKeyWrite(c->db,key); if (zsetobj == NULL) { zsetobj = createZsetObject(); - dictAdd(c->db->dict,c->argv[1],zsetobj); - incrRefCount(c->argv[1]); + dictAdd(c->db->dict,key,zsetobj); + incrRefCount(key); } else { if (zsetobj->type != REDIS_ZSET) { addReply(c,shared.wrongtypeerr); return; } } - score = zmalloc(sizeof(double)); - *score = strtod(c->argv[2]->ptr,NULL); zs = zsetobj->ptr; - if (dictAdd(zs->dict,c->argv[3],score) == DICT_OK) { + + /* Ok now since we implement both ZADD and ZINCRBY here the code + * needs to handle the two different conditions. It's all about setting + * '*score', that is, the new score to set, to the right value. */ + score = zmalloc(sizeof(double)); + if (doincrement) { + dictEntry *de; + + /* Read the old score. If the element was not present starts from 0 */ + de = dictFind(zs->dict,ele); + if (de) { + double *oldscore = dictGetEntryVal(de); + *score = *oldscore + scoreval; + } else { + *score = scoreval; + } + } else { + *score = scoreval; + } + + /* What follows is a simple remove and re-insert operation that is common + * to both ZADD and ZINCRBY... */ + if (dictAdd(zs->dict,ele,score) == DICT_OK) { /* case 1: New element */ - incrRefCount(c->argv[3]); /* added to hash */ - zslInsert(zs->zsl,*score,c->argv[3]); - incrRefCount(c->argv[3]); /* added to skiplist */ + incrRefCount(ele); /* added to hash */ + zslInsert(zs->zsl,*score,ele); + incrRefCount(ele); /* added to skiplist */ server.dirty++; - addReply(c,shared.cone); + if (doincrement) + addReplyDouble(c,*score); + else + addReply(c,shared.cone); } else { dictEntry *de; double *oldscore; /* case 2: Score update operation */ - de = dictFind(zs->dict,c->argv[3]); - assert(de != NULL); + de = dictFind(zs->dict,ele); + redisAssert(de != NULL); oldscore = dictGetEntryVal(de); if (*score != *oldscore) { int deleted; - deleted = zslDelete(zs->zsl,*oldscore,c->argv[3]); - assert(deleted != 0); - zslInsert(zs->zsl,*score,c->argv[3]); - incrRefCount(c->argv[3]); - dictReplace(zs->dict,c->argv[3],score); + /* Remove and insert the element in the skip list with new score */ + deleted = zslDelete(zs->zsl,*oldscore,ele); + redisAssert(deleted != 0); + zslInsert(zs->zsl,*score,ele); + incrRefCount(ele); + /* Update the score in the hash table */ + dictReplace(zs->dict,ele,score); server.dirty++; } else { zfree(score); } - addReply(c,shared.czero); + if (doincrement) + addReplyDouble(c,*score); + else + addReply(c,shared.czero); } } +static void zaddCommand(redisClient *c) { + double scoreval; + + scoreval = strtod(c->argv[2]->ptr,NULL); + zaddGenericCommand(c,c->argv[1],c->argv[3],scoreval,0); +} + +static void zincrbyCommand(redisClient *c) { + double scoreval; + + scoreval = strtod(c->argv[2]->ptr,NULL); + zaddGenericCommand(c,c->argv[1],c->argv[3],scoreval,1); +} + static void zremCommand(redisClient *c) { robj *zsetobj; zset *zs; @@ -4297,7 +5234,7 @@ static void zremCommand(redisClient *c) { /* Delete from the skiplist */ oldscore = dictGetEntryVal(de); deleted = zslDelete(zs->zsl,*oldscore,c->argv[2]); - assert(deleted != 0); + redisAssert(deleted != 0); /* Delete from the hash table */ dictDelete(zs->dict,c->argv[2]); @@ -4335,6 +5272,14 @@ static void zrangeGenericCommand(redisClient *c, int reverse) { robj *o; int start = atoi(c->argv[2]->ptr); int end = atoi(c->argv[3]->ptr); + int withscores = 0; + + if (c->argc == 5 && !strcasecmp(c->argv[4]->ptr,"withscores")) { + withscores = 1; + } else if (c->argc >= 5) { + addReply(c,shared.syntaxerr); + return; + } o = lookupKeyRead(c->db,c->argv[1]); if (o == NULL) { @@ -4366,23 +5311,24 @@ static void zrangeGenericCommand(redisClient *c, int reverse) { if (end >= llen) end = llen-1; rangelen = (end-start)+1; - /* Return the result in form of a multi-bulk reply */ + /* check if starting point is trivial, before searching + * the element in log(N) time */ if (reverse) { - ln = zsl->tail; - while (start--) - ln = ln->backward; + ln = start == 0 ? zsl->tail : zslGetElementByRank(zsl, llen - start); } else { - ln = zsl->header->forward[0]; - while (start--) - ln = ln->forward[0]; + ln = start == 0 ? zsl->header->forward[0] : zslGetElementByRank(zsl, start + 1); } - addReplySds(c,sdscatprintf(sdsempty(),"*%d\r\n",rangelen)); + /* 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; addReplyBulkLen(c,ele); addReply(c,ele); addReply(c,shared.crlf); + if (withscores) + addReplyDouble(c,ln->score); ln = reverse ? ln->backward : ln->forward[0]; } } @@ -4397,14 +5343,64 @@ 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); + } + + /* 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; + } + + /* Parse "LIMIT" */ + if (c->argc == (7 + withscores) && strcasecmp(c->argv[4]->ptr,"limit")) { + addReply(c,shared.syntaxerr); + return; + } 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.nullmultibulk); } else { if (o->type != REDIS_ZSET) { addReply(c,shared.wrongtypeerr); @@ -4412,14 +5408,17 @@ 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; } @@ -4427,22 +5426,49 @@ 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); + if (!justcount) { + lenobj = createObject(REDIS_STRING,NULL); + addReply(c,lenobj); + decrRefCount(lenobj); + } - while(ln && ln->score <= max) { - ele = ln->obj; - addReplyBulkLen(c,ele); - addReply(c,ele); - addReply(c,shared.crlf); + while(ln && (maxex ? (ln->score < max) : (ln->score <= max))) { + if (offset) { + offset--; + ln = ln->forward[0]; + continue; + } + if (limit == 0) break; + if (!justcount) { + ele = ln->obj; + addReplyBulkLen(c,ele); + addReply(c,ele); + addReply(c,shared.crlf); + if (withscores) + addReplyDouble(c,ln->score); + } ln = ln->forward[0]; rangelen++; + if (limit > 0) limit--; + } + if (justcount) { + addReplyLong(c,(long)rangelen); + } else { + lenobj->ptr = sdscatprintf(sdsempty(),"*%lu\r\n", + withscores ? (rangelen*2) : rangelen); } - lenobj->ptr = sdscatprintf(sdsempty(),"*%d\r\n",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; @@ -4456,7 +5482,7 @@ static void zcardCommand(redisClient *c) { addReply(c,shared.wrongtypeerr); } else { zs = o->ptr; - addReplySds(c,sdscatprintf(sdsempty(),":%d\r\n",zs->zsl->length)); + addReplySds(c,sdscatprintf(sdsempty(),":%lu\r\n",zs->zsl->length)); } } } @@ -4480,25 +5506,53 @@ static void zscoreCommand(redisClient *c) { if (!de) { addReply(c,shared.nullbulk); } else { - char buf[128]; double *score = dictGetEntryVal(de); - snprintf(buf,sizeof(buf),"%.17g",*score); - addReplySds(c,sdscatprintf(sdsempty(),"$%d\r\n%s\r\n", - strlen(buf),buf)); + addReplyDouble(c,*score); } } } } -/* ========================= 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) { + robj *o; + o = lookupKeyRead(c->db,c->argv[1]); + if (o == NULL) { + addReply(c,shared.nullbulk); + return; + } + if (o->type != REDIS_ZSET) { + addReply(c,shared.wrongtypeerr); + } else { + zset *zs = o->ptr; + zskiplist *zsl = zs->zsl; + dictEntry *de; + unsigned long rank; + + de = dictFind(zs->dict,c->argv[2]); + if (!de) { + addReply(c,shared.nullbulk); + return; + } + + double *score = dictGetEntryVal(de); + rank = zslGetRank(zsl, *score, c->argv[2]); + if (rank) { + 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 flushallCommand(redisClient *c) { server.dirty += emptyDb(); @@ -4528,17 +5582,25 @@ static robj *lookupKeyByPattern(redisDb *db, robj *pattern, robj *subst) { char buf[REDIS_SORTKEY_MAX+1]; } keyname; - if (subst->encoding == REDIS_ENCODING_RAW) - incrRefCount(subst); - else { - subst = getDecodedObject(subst); + /* 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') { + return subst; } - spat = pattern->ptr; + /* The substitution object may be specially encoded. If so we create + * a decoded object on the fly. Otherwise getDecodedObject will just + * increment the ref count, that we'll decrement later. */ + subst = getDecodedObject(subst); + ssub = subst->ptr; if (sdslen(spat)+sdslen(ssub)-1 > REDIS_SORTKEY_MAX) return NULL; p = strchr(spat,'*'); - if (!p) return NULL; + if (!p) { + decrRefCount(subst); + return NULL; + } prefixlen = p-spat; sublen = sdslen(ssub); @@ -4549,10 +5611,7 @@ static robj *lookupKeyByPattern(redisDb *db, robj *pattern, robj *subst) { keyname.buf[prefixlen+sublen+postfixlen] = '\0'; keyname.len = prefixlen+sublen+postfixlen; - keyobj.refcount = 1; - keyobj.type = REDIS_STRING; - keyobj.ptr = ((char*)&keyname)+(sizeof(long)*2); - + initStaticStringObject(keyobj,((char*)&keyname)+(sizeof(long)*2)) decrRefCount(subst); /* printf("lookup '%s' => %p\n", keyname.buf,de); */ @@ -4592,20 +5651,13 @@ static int sortCompare(const void *s1, const void *s2) { } } else { /* Compare elements directly */ - if (so1->obj->encoding == REDIS_ENCODING_RAW && - so2->obj->encoding == REDIS_ENCODING_RAW) { - cmp = strcoll(so1->obj->ptr,so2->obj->ptr); - } else { - robj *dec1, *dec2; - - dec1 = so1->obj->encoding == REDIS_ENCODING_RAW ? - so1->obj : getDecodedObject(so1->obj); - dec2 = so2->obj->encoding == REDIS_ENCODING_RAW ? - so2->obj : getDecodedObject(so2->obj); - cmp = strcoll(dec1->ptr,dec2->ptr); - if (dec1 != so1->obj) decrRefCount(dec1); - if (dec2 != so2->obj) decrRefCount(dec2); - } + robj *dec1, *dec2; + + dec1 = getDecodedObject(so1->obj); + dec2 = getDecodedObject(so2->obj); + cmp = strcoll(dec1->ptr,dec2->ptr); + decrRefCount(dec1); + decrRefCount(dec2); } } return server.sort_desc ? -cmp : cmp; @@ -4626,10 +5678,12 @@ 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.nokeyerr); + addReply(c,shared.nullmultibulk); return; } - if (sortval->type != REDIS_SET && sortval->type != REDIS_LIST) { + if (sortval->type != REDIS_SET && sortval->type != REDIS_LIST && + sortval->type != REDIS_ZSET) + { addReply(c,shared.wrongtypeerr); return; } @@ -4682,17 +5736,22 @@ static void sortCommand(redisClient *c) { } /* Load the sorting vector with all the objects to sort */ - vectorlen = (sortval->type == REDIS_LIST) ? - listLength((list*)sortval->ptr) : - dictSize((dict*)sortval->ptr); + switch(sortval->type) { + 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 */ + } vector = zmalloc(sizeof(redisSortObject)*vectorlen); j = 0; + 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; @@ -4700,10 +5759,17 @@ static void sortCommand(redisClient *c) { j++; } } else { - dict *set = sortval->ptr; + dict *set; dictIterator *di; dictEntry *setele; + if (sortval->type == REDIS_SET) { + set = sortval->ptr; + } else { + zset *zs = sortval->ptr; + set = zs->dict; + } + di = dictGetIterator(set); while((setele = dictNext(di)) != NULL) { vector[j].obj = dictGetEntryKey(setele); @@ -4713,7 +5779,7 @@ static void sortCommand(redisClient *c) { } dictReleaseIterator(di); } - assert(j == vectorlen); + redisAssert(j == vectorlen); /* Now it's time to load the right scores in the sorting vector */ if (dontsort == 0) { @@ -4724,20 +5790,18 @@ static void sortCommand(redisClient *c) { byval = lookupKeyByPattern(c->db,sortby,vector[j].obj); if (!byval || byval->type != REDIS_STRING) continue; if (alpha) { - if (byval->encoding == REDIS_ENCODING_RAW) { - vector[j].u.cmpobj = byval; - incrRefCount(byval); - } else { - vector[j].u.cmpobj = getDecodedObject(byval); - } + 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 - assert(1 != 1); + redisAssert(1 != 1); } } } else { @@ -4748,7 +5812,7 @@ static void sortCommand(redisClient *c) { if (vector[j].obj->encoding == REDIS_ENCODING_INT) vector[j].u.score = (long) vector[j].obj->ptr; else - assert(1 != 1); + redisAssert(1 != 1); } } } @@ -4783,13 +5847,15 @@ static void sortCommand(redisClient *c) { addReplySds(c,sdscatprintf(sdsempty(),"*%d\r\n",outputlen)); for (j = start; j <= end; j++) { listNode *ln; + listIter li; + if (!getop) { addReplyBulkLen(c,vector[j].obj); addReply(c,vector[j].obj); addReply(c,shared.crlf); } - 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); @@ -4803,7 +5869,7 @@ static void sortCommand(redisClient *c) { addReply(c,shared.crlf); } } else { - assert(sop->type == REDIS_SORT_GET); /* always fails */ + redisAssert(sop->type == REDIS_SORT_GET); /* always fails */ } } } @@ -4814,12 +5880,14 @@ 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); @@ -4832,7 +5900,7 @@ static void sortCommand(redisClient *c) { incrRefCount(val); } } else { - assert(sop->type == REDIS_SORT_GET); /* always fails */ + redisAssert(sop->type == REDIS_SORT_GET); /* always fails */ } } } @@ -4856,37 +5924,75 @@ static void sortCommand(redisClient *c) { zfree(vector); } -static void infoCommand(redisClient *c) { +/* 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. */ +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" - "uptime_in_seconds:%d\r\n" - "uptime_in_days:%d\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:%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" + "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.bgsaveinprogress, + server.bgsavechildpid != -1, server.lastsave, + server.bgrewritechildpid != -1, server.stat_numconnections, server.stat_numcommands, + server.vm_enabled != 0, server.masterhost == NULL ? "master" : "slave" ); if (server.masterhost) { @@ -4902,17 +6008,53 @@ static void infoCommand(redisClient *c) { 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; keys = dictSize(server.db[j].dict); vkeys = dictSize(server.db[j].expires); if (keys || vkeys) { - info = sdscatprintf(info, "db%d: keys=%lld,expires=%lld\r\n", + info = sdscatprintf(info, "db%d:keys=%lld,expires=%lld\r\n", j, keys, vkeys); } } - addReplySds(c,sdscatprintf(sdsempty(),"$%d\r\n",sdslen(info))); + return info; +} + +static void infoCommand(redisClient *c) { + sds info = genRedisInfoString(); + addReplySds(c,sdscatprintf(sdsempty(),"$%lu\r\n", + (unsigned long)sdslen(info))); addReplySds(c,info); addReply(c,shared.crlf); } @@ -5031,47 +6173,279 @@ static void ttlCommand(redisClient *c) { addReplySds(c,sdscatprintf(sdsempty(),":%d\r\n",ttl)); } -static void msetGenericCommand(redisClient *c, int nx) { +/* ================================ MULTI/EXEC ============================== */ + +/* Client state initialization for MULTI/EXEC */ +static void initClientMultiState(redisClient *c) { + c->mstate.commands = NULL; + c->mstate.count = 0; +} + +/* Release all the resources associated with MULTI/EXEC state */ +static void freeClientMultiState(redisClient *c) { int j; - if ((c->argc % 2) == 0) { - addReplySds(c,sdsnew("-ERR wrong number of arguments\r\n")); + for (j = 0; j < c->mstate.count; j++) { + int i; + multiCmd *mc = c->mstate.commands+j; + + for (i = 0; i < mc->argc; i++) + decrRefCount(mc->argv[i]); + zfree(mc->argv); + } + zfree(c->mstate.commands); +} + +/* Add a new command into the MULTI commands queue */ +static void queueMultiCommand(redisClient *c, struct redisCommand *cmd) { + multiCmd *mc; + int j; + + c->mstate.commands = zrealloc(c->mstate.commands, + sizeof(multiCmd)*(c->mstate.count+1)); + mc = c->mstate.commands+c->mstate.count; + mc->cmd = cmd; + mc->argc = c->argc; + mc->argv = zmalloc(sizeof(robj*)*c->argc); + memcpy(mc->argv,c->argv,sizeof(robj*)*c->argc); + for (j = 0; j < c->argc; j++) + incrRefCount(mc->argv[j]); + c->mstate.count++; +} + +static void multiCommand(redisClient *c) { + c->flags |= REDIS_MULTI; + addReply(c,shared.ok); +} + +static void discardCommand(redisClient *c) { + if (!(c->flags & REDIS_MULTI)) { + addReplySds(c,sdsnew("-ERR DISCARD without MULTI\r\n")); return; } - /* Handle the NX flag. The MSETNX semantic is to return zero and don't - * set nothing at all if at least one already key exists. */ - if (nx) { - for (j = 1; j < c->argc; j += 2) { - if (dictFind(c->db->dict,c->argv[j]) != NULL) { - addReply(c, shared.czero); - return; - } - } + + freeClientMultiState(c); + initClientMultiState(c); + c->flags &= (~REDIS_MULTI); + addReply(c,shared.ok); +} + +static void execCommand(redisClient *c) { + int j; + robj **orig_argv; + int orig_argc; + + if (!(c->flags & REDIS_MULTI)) { + addReplySds(c,sdsnew("-ERR EXEC without MULTI\r\n")); + return; } - for (j = 1; j < c->argc; j += 2) { - int retval; + orig_argv = c->argv; + orig_argc = c->argc; + addReplySds(c,sdscatprintf(sdsempty(),"*%d\r\n",c->mstate.count)); + for (j = 0; j < c->mstate.count; j++) { + c->argc = c->mstate.commands[j].argc; + c->argv = c->mstate.commands[j].argv; + call(c,c->mstate.commands[j].cmd); + } + c->argv = orig_argv; + c->argc = orig_argc; + freeClientMultiState(c); + initClientMultiState(c); + c->flags &= (~REDIS_MULTI); +} - 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]); - incrRefCount(c->argv[j+1]); +/* =========================== Blocking Operations ========================= */ + +/* Currently Redis blocking operations support is limited to list POP ops, + * so the current implementation is not fully generic, but it is also not + * completely specific so it will not require a rewrite to support new + * kind of blocking operations in the future. + * + * Still it's important to note that list blocking operations can be already + * used as a notification mechanism in order to implement other blocking + * operations at application level, so there must be a very strong evidence + * of usefulness and generality before new blocking operations are implemented. + * + * This is how the current blocking POP works, we use BLPOP as example: + * - If the user calls BLPOP and the key exists and contains a non empty list + * then LPOP is called instead. So BLPOP is semantically the same as LPOP + * if there is not to block. + * - If instead BLPOP is called and the key does not exists or the list is + * 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 (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 + * the new element inside the list we return it to the (first / oldest) + * blocking client, unblock the client, and remove it form the list. + * + * The above comment and the source code should be enough in order to understand + * the implementation and modify / fix it later. + */ + +/* Set a client in blocking mode for the specified key, with the specified + * timeout */ +static void blockForKeys(redisClient *c, robj **keys, int numkeys, time_t timeout) { + dictEntry *de; + list *l; + int j; + + c->blockingkeys = zmalloc(sizeof(robj*)*numkeys); + c->blockingkeysnum = numkeys; + c->blockingto = timeout; + 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]); + + /* 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 { - incrRefCount(c->argv[j]); - incrRefCount(c->argv[j+1]); + l = dictGetEntryVal(de); } - removeExpire(c->db,c->argv[j]); + listAddNodeTail(l,c); } - server.dirty += (c->argc-1)/2; - addReply(c, nx ? shared.cone : shared.ok); + /* Mark the client as a blocked client */ + c->flags |= REDIS_BLOCKED; + server.blpop_blocked_clients++; } -static void msetCommand(redisClient *c) { - msetGenericCommand(c,0); +/* Unblock a client that's waiting in a blocking operation such as BLPOP */ +static void unblockClientWaitingData(redisClient *c) { + dictEntry *de; + list *l; + int j; + + 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); + 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); +} + +/* This should be called from any function PUSHing into lists. + * 'c' is the "pushing client", 'key' is the key it is pushing data against, + * 'ele' is the element pushed. + * + * If the function returns 0 there was no client waiting for a list push + * against this key. + * + * If the function returns 1 there was a client waiting for a list push + * against this key, the element was passed to this client thus it's not + * needed to actually add it to the list and the caller should return asap. */ +static int handleClientsWaitingListPush(redisClient *c, robj *key, robj *ele) { + struct dictEntry *de; + redisClient *receiver; + list *l; + listNode *ln; + + de = dictFind(c->db->blockingkeys,key); + if (de == NULL) return 0; + l = dictGetEntryVal(de); + ln = listFirst(l); + assert(ln != NULL); + receiver = ln->value; + + addReplySds(receiver,sdsnew("*2\r\n")); + addReplyBulkLen(receiver,key); + addReply(receiver,key); + addReply(receiver,shared.crlf); + addReplyBulkLen(receiver,ele); + addReply(receiver,ele); + addReply(receiver,shared.crlf); + unblockClientWaitingData(receiver); + return 1; } -static void msetnxCommand(redisClient *c) { - msetGenericCommand(c,1); +/* Blocking RPOP/LPOP */ +static void blockingPopGenericCommand(redisClient *c, int where) { + robj *o; + time_t timeout; + int j; + + 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")); + addReplyBulkLen(c,argv[1]); + addReply(c,argv[1]); + addReply(c,shared.crlf); + 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[c->argc-1]->ptr,NULL,10); + if (timeout > 0) timeout += time(NULL); + blockForKeys(c,c->argv+1,c->argc-2,timeout); +} + +static void blpopCommand(redisClient *c) { + blockingPopGenericCommand(c,REDIS_HEAD); +} + +static void brpopCommand(redisClient *c) { + blockingPopGenericCommand(c,REDIS_TAIL); } /* =============================== Replication ============================= */ @@ -5154,15 +6528,16 @@ static void syncCommand(redisClient *c) { redisLog(REDIS_NOTICE,"Slave ask for synchronization"); /* Here we need to check if there is a background saving operation * in progress, or if it is required to start one */ - if (server.bgsaveinprogress) { + if (server.bgsavechildpid != -1) { /* Ok a background save is in progress. Let's check if it is a good * one for replication, i.e. if there is another slave that is * 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; } @@ -5229,7 +6604,7 @@ 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; @@ -5241,7 +6616,7 @@ static void sendBulkToSlave(aeEventLoop *el, int fd, void *privdata, int mask) { aeDeleteFileEvent(server.el,slave->fd,AE_WRITABLE); slave->replstate = REDIS_REPL_ONLINE; if (aeCreateFileEvent(server.el, slave->fd, AE_WRITABLE, - sendReplyToClient, slave, NULL) == AE_ERR) { + sendReplyToClient, slave) == AE_ERR) { freeClient(slave); return; } @@ -5259,9 +6634,10 @@ 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) { @@ -5285,7 +6661,7 @@ static void updateSlavesWaitingBgsave(int bgsaveerr) { slave->repldbsize = buf.st_size; slave->replstate = REDIS_REPL_SEND_BULK; aeDeleteFileEvent(server.el,slave->fd,AE_WRITABLE); - if (aeCreateFileEvent(server.el, slave->fd, AE_WRITABLE, sendBulkToSlave, slave, NULL) == AE_ERR) { + if (aeCreateFileEvent(server.el, slave->fd, AE_WRITABLE, sendBulkToSlave, slave) == AE_ERR) { freeClient(slave); continue; } @@ -5293,9 +6669,11 @@ 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) @@ -5404,6 +6782,7 @@ static int syncWithMaster(void) { } server.master = createClient(fd); server.master->flags |= REDIS_MASTER; + server.master->authenticated = 1; server.replstate = REDIS_REPL_CONNECTED; return REDIS_OK; } @@ -5432,6 +6811,27 @@ 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: @@ -5445,40 +6845,32 @@ 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... */ } } @@ -5497,8 +6889,8 @@ static void feedAppendOnlyFile(struct redisCommand *cmd, int dictid, robj **argv char seldb[64]; snprintf(seldb,sizeof(seldb),"%d",dictid); - buf = sdscatprintf(buf,"*2\r\n$6\r\nSELECT\r\n$%d\r\n%s\r\n", - strlen(seldb),seldb); + buf = sdscatprintf(buf,"*2\r\n$6\r\nSELECT\r\n$%lu\r\n%s\r\n", + (unsigned long)strlen(seldb),seldb); server.appendseldb = dictid; } @@ -5521,13 +6913,11 @@ static void feedAppendOnlyFile(struct redisCommand *cmd, int dictid, robj **argv for (j = 0; j < argc; j++) { robj *o = argv[j]; - if (o->encoding != REDIS_ENCODING_RAW) - o = getDecodedObject(o); - buf = sdscatprintf(buf,"$%d\r\n",sdslen(o->ptr)); + o = getDecodedObject(o); + buf = sdscatprintf(buf,"$%lu\r\n",(unsigned long)sdslen(o->ptr)); buf = sdscatlen(buf,o->ptr,sdslen(o->ptr)); buf = sdscatlen(buf,"\r\n",2); - if (o != argv[j]) - decrRefCount(o); + decrRefCount(o); } /* Free the objects from the modified argv for EXPIREAT */ @@ -5553,6 +6943,14 @@ static void feedAppendOnlyFile(struct redisCommand *cmd, int dictid, robj **argv } exit(1); } + /* If a background append only file rewriting is in progress we want to + * accumulate the differences between the child DB and the current one + * in a buffer, so that when the child process will do its work we + * can append the differences to the new append only file. */ + if (server.bgrewritechildpid != -1) + server.bgrewritebuf = sdscatlen(server.bgrewritebuf,buf,sdslen(buf)); + + sdsfree(buf); now = time(NULL); if (server.appendfsync == APPENDFSYNC_ALWAYS || (server.appendfsync == APPENDFSYNC_EVERYSEC && @@ -5596,6 +6994,7 @@ 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; @@ -5628,7 +7027,7 @@ int loadAppendOnlyFile(char *filename) { if (buf[0] != '$') goto fmterr; len = strtol(buf+1,NULL,10); argsds = sdsnewlen(NULL,len); - if (fread(argsds,len,1,fp) == 0) goto fmterr; + if (len && fread(argsds,len,1,fp) == 0) goto fmterr; argv[j] = createObject(REDIS_STRING,argsds); if (fread(buf,2,1,fp) == 0) goto fmterr; /* discard CRLF */ } @@ -5657,6 +7056,13 @@ 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); @@ -5674,69 +7080,1436 @@ fmterr: exit(1); } -/* ================================= Debugging ============================== */ - -static void debugCommand(redisClient *c) { - if (!strcasecmp(c->argv[1]->ptr,"segfault")) { - *((char*)-1) = 'x'; - } else if (!strcasecmp(c->argv[1]->ptr,"object") && c->argc == 3) { - dictEntry *de = dictFind(c->db->dict,c->argv[2]); - robj *key, *val; +/* Write an object into a file in the bulk format $\r\n\r\n */ +static int fwriteBulk(FILE *fp, robj *obj) { + char buf[128]; + int decrrc = 0; - if (!de) { - addReply(c,shared.nokeyerr); - return; - } - key = dictGetEntryKey(de); - val = dictGetEntryVal(de); - addReplySds(c,sdscatprintf(sdsempty(), - "+Key at:%p refcount:%d, value at:%p refcount:%d encoding:%d\r\n", - key, key->refcount, val, val->refcount, val->encoding)); - } else { - addReplySds(c,sdsnew( - "-ERR Syntax error, try DEBUG [SEGFAULT|OBJECT ]\r\n")); - } + /* Avoid the incr/decr ref count business if possible to help + * copy-on-write (we are often in a child process when this function + * is called). + * Also makes sure that key objects don't get incrRefCount-ed when VM + * is enabled */ + if (obj->encoding != REDIS_ENCODING_RAW) { + obj = getDecodedObject(obj); + decrrc = 1; + } + snprintf(buf,sizeof(buf),"$%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; + if (decrrc) decrRefCount(obj); + return 1; +err: + if (decrrc) decrRefCount(obj); + return 0; } -/* =================================== Main! ================================ */ +/* Write a double value in bulk format $\r\n\r\n */ +static int fwriteBulkDouble(FILE *fp, double d) { + char buf[128], dbuf[128]; -#ifdef __linux__ -int linuxOvercommitMemoryValue(void) { - FILE *fp = fopen("/proc/sys/vm/overcommit_memory","r"); - char buf[64]; + snprintf(dbuf,sizeof(dbuf),"%.17g\r\n",d); + snprintf(buf,sizeof(buf),"$%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; +} - if (!fp) return -1; - if (fgets(buf,64,fp) == NULL) { - fclose(fp); - return -1; - } - fclose(fp); +/* Write a long value in bulk format $\r\n\r\n */ +static int fwriteBulkLong(FILE *fp, long l) { + char buf[128], lbuf[128]; - return atoi(buf); + snprintf(lbuf,sizeof(lbuf),"%ld\r\n",l); + snprintf(buf,sizeof(buf),"$%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; } -void linuxOvercommitMemoryWarning(void) { - if (linuxOvercommitMemoryValue() == 0) { - redisLog(REDIS_WARNING,"WARNING overcommit_memory is set to 0! Background save may fail under low condition memory. To fix this issue add 'vm.overcommit_memory = 1' to /etc/sysctl.conf and then reboot or run the command 'sysctl vm.overcommit_memory=1' for this to take effect."); +/* 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; } -} -#endif /* __linux__ */ + 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; + } -static void daemonize(void) { - int fd; - FILE *fp; + /* SELECT the new DB */ + if (fwrite(selectcmd,sizeof(selectcmd)-1,1,fp) == 0) goto werr; + if (fwriteBulkLong(fp,j) == 0) goto werr; - if (fork() != 0) exit(0); /* parent exits */ - setsid(); /* create a new session */ + /* 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); - /* Every output goes to /dev/null. If Redis is daemonized but - * the 'logfile' is set to 'stdout' in the configuration file - * it will not log at all. */ - if ((fd = open("/dev/null", O_RDWR, 0)) != -1) { - dup2(fd, STDIN_FILENO); - dup2(fd, STDOUT_FILENO); - dup2(fd, STDERR_FILENO); - if (fd > STDERR_FILENO) close(fd); + /* 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; + 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 (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) { + 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 (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; + + 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 (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; + } + 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; + /* 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; + } + 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); + decrRefCount(j->key); + 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 (compareStringObjects(job->key,o) == 0) { + 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; + + sigemptyset(&mask); + sigaddset(&mask,SIGCHLD); + sigaddset(&mask,SIGHUP); + sigaddset(&mask,SIGPIPE); + pthread_sigmask(SIG_SETMASK, &mask, &omask); + pthread_create(&thread,&server.io_threads_attr,IOThreadEntryPoint,NULL); + 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 = dupStringObject(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 = dupStringObject(key); + 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; +} + +/* 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_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); + } + } +} + +/* ================================= Debugging ============================== */ + +static void debugCommand(redisClient *c) { + if (!strcasecmp(c->argv[1]->ptr,"segfault")) { + *((char*)-1) = 'x'; + } else if (!strcasecmp(c->argv[1]->ptr,"reload")) { + if (rdbSave(server.dbfilename) != REDIS_OK) { + addReply(c,shared.err); + return; + } + emptyDb(); + if (rdbLoad(server.dbfilename) != REDIS_OK) { + addReply(c,shared.err); + return; + } + redisLog(REDIS_WARNING,"DB reloaded by DEBUG RELOAD"); + addReply(c,shared.ok); + } else if (!strcasecmp(c->argv[1]->ptr,"loadaof")) { + emptyDb(); + if (loadAppendOnlyFile(server.appendfilename) != REDIS_OK) { + addReply(c,shared.err); + return; + } + redisLog(REDIS_WARNING,"Append Only File loaded by DEBUG LOADAOF"); + addReply(c,shared.ok); + } else if (!strcasecmp(c->argv[1]->ptr,"object") && c->argc == 3) { + dictEntry *de = dictFind(c->db->dict,c->argv[2]); + robj *key, *val; + + if (!de) { + addReply(c,shared.nokeyerr); + return; + } + key = dictGetEntryKey(de); + val = dictGetEntryVal(de); + if (!server.vm_enabled || (key->storage == REDIS_VM_MEMORY || + key->storage == REDIS_VM_SWAPPING)) { + addReplySds(c,sdscatprintf(sdsempty(), + "+Key at:%p refcount:%d, value at:%p refcount:%d " + "encoding:%d serializedlength:%lld\r\n", + (void*)key, key->refcount, (void*)val, val->refcount, + val->encoding, (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,"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 |SWAPOUT |RELOAD]\r\n")); + } +} + +static void _redisAssert(char *estr, char *file, int line) { + redisLog(REDIS_WARNING,"=== ASSERTION FAILED ==="); + 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 +} + +/* =================================== Main! ================================ */ + +#ifdef __linux__ +int linuxOvercommitMemoryValue(void) { + FILE *fp = fopen("/proc/sys/vm/overcommit_memory","r"); + char buf[64]; + + if (!fp) return -1; + if (fgets(buf,64,fp) == NULL) { + fclose(fp); + return -1; + } + fclose(fp); + + return atoi(buf); +} + +void linuxOvercommitMemoryWarning(void) { + if (linuxOvercommitMemoryValue() == 0) { + redisLog(REDIS_WARNING,"WARNING overcommit_memory is set to 0! Background save may fail under low condition memory. To fix this issue add 'vm.overcommit_memory = 1' to /etc/sysctl.conf and then reboot or run the command 'sysctl vm.overcommit_memory=1' for this to take effect."); + } +} +#endif /* __linux__ */ + +static void daemonize(void) { + int fd; + FILE *fp; + + if (fork() != 0) exit(0); /* parent exits */ + setsid(); /* create a new session */ + + /* Every output goes to /dev/null. If Redis is daemonized but + * the 'logfile' is set to 'stdout' in the configuration file + * it will not log at all. */ + if ((fd = open("/dev/null", O_RDWR, 0)) != -1) { + dup2(fd, STDIN_FILENO); + dup2(fd, STDOUT_FILENO); + dup2(fd, STDERR_FILENO); + if (fd > STDERR_FILENO) close(fd); } /* Try to write the pid file */ fp = fopen(server.pidfile,"w"); @@ -5747,6 +8520,8 @@ static void daemonize(void) { } int main(int argc, char **argv) { + time_t start; + initServerConfig(); if (argc == 2) { resetServerSaveParams(); @@ -5757,22 +8532,22 @@ int main(int argc, char **argv) { } 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'"); } - initServer(); if (server.daemonize) daemonize(); + initServer(); redisLog(REDIS_NOTICE,"Server started, Redis version " REDIS_VERSION); #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, 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; @@ -5789,15 +8564,19 @@ static void *getMcontextEip(ucontext_t *uc) { #elif defined(__dietlibc__) return (void*) uc->uc_mcontext.eip; #elif defined(__APPLE__) && !defined(MAC_OS_X_VERSION_10_6) + #if __x86_64__ + return (void*) uc->uc_mcontext->__ss.__rip; + #else return (void*) uc->uc_mcontext->__ss.__eip; + #endif #elif defined(__APPLE__) && defined(MAC_OS_X_VERSION_10_6) #if defined(_STRUCT_X86_THREAD_STATE64) && !defined(__i386__) return (void*) uc->uc_mcontext->__ss.__rip; #else return (void*) uc->uc_mcontext->__ss.__eip; #endif -#elif defined(__i386__) || defined(__X86_64__) /* Linux x86 */ - return (void*) uc->uc_mcontext.gregs[REG_EIP]; +#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; #else @@ -5810,36 +8589,16 @@ static void segvHandler(int sig, siginfo_t *info, void *secret) { char **messages = NULL; int i, trace_size = 0; unsigned long offset=0; - time_t uptime = time(NULL)-server.stat_starttime; ucontext_t *uc = (ucontext_t*) secret; + sds infostring; REDIS_NOTUSED(info); redisLog(REDIS_WARNING, "======= Ooops! Redis %s got signal: -%d- =======", REDIS_VERSION, sig); - redisLog(REDIS_WARNING, "%s", sdscatprintf(sdsempty(), - "redis_version:%s; " - "uptime_in_seconds:%d; " - "connected_clients:%d; " - "connected_slaves:%d; " - "used_memory:%zu; " - "changes_since_last_save:%lld; " - "bgsave_in_progress:%d; " - "last_save_time:%d; " - "total_connections_received:%lld; " - "total_commands_processed:%lld; " - "role:%s;" - ,REDIS_VERSION, - uptime, - listLength(server.clients)-listLength(server.slaves), - listLength(server.slaves), - server.usedmemory, - server.dirty, - server.bgsaveinprogress, - server.lastsave, - server.stat_numconnections, - server.stat_numcommands, - server.masterhost == NULL ? "master" : "slave" - )); + infostring = genRedisInfoString(); + 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 */ @@ -5858,8 +8617,8 @@ 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); - exit(0); + /* free(messages); Don't call free() with possibly corrupted memory. */ + _exit(0); } static void setupSigSegvAction(void) {