X-Git-Url: https://git.saurik.com/redis.git/blobdiff_plain/996cb5f704b2b37e764d63d0889463fd8ef2b70e..ed7451548a0e874ac3b2e029268655afab96b506:/redis.c diff --git a/redis.c b/redis.c index 36bee82c..dce4eb2d 100644 --- a/redis.c +++ b/redis.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2006-2009, Salvatore Sanfilippo + * Copyright (c) 2009-2010, Salvatore Sanfilippo * All rights reserved. * * Redistribution and use in source and binary forms, with or without @@ -27,7 +27,7 @@ * POSSIBILITY OF SUCH DAMAGE. */ -#define REDIS_VERSION "1.3.2" +#define REDIS_VERSION "1.3.4" #include "fmacros.h" #include "config.h" @@ -164,15 +164,20 @@ #define REDIS_VM_MAX_NEAR_PAGES 65536 #define REDIS_VM_MAX_RANDOM_JUMP 4096 #define REDIS_VM_MAX_THREADS 32 +#define REDIS_THREAD_STACK_SIZE (1024*1024*4) +/* The following is the *percentage* of completed I/O jobs to process when the + * handelr is called. While Virtual Memory I/O operations are performed by + * threads, this operations must be processed by the main thread when completed + * in order to take effect. */ +#define REDIS_MAX_COMPLETED_JOBS_PROCESSED 1 /* Client flags */ -#define REDIS_CLOSE 1 /* This client connection should be closed ASAP */ -#define REDIS_SLAVE 2 /* This client is a slave server */ -#define REDIS_MASTER 4 /* This client is a master server */ -#define REDIS_MONITOR 8 /* This client is a slave monitor, see MONITOR */ -#define REDIS_MULTI 16 /* This client is in a MULTI context */ -#define REDIS_BLOCKED 32 /* The client is waiting in a blocking operation */ -#define REDIS_IO_WAIT 64 /* The client is waiting for Virtual Memory I/O */ +#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 */ @@ -216,8 +221,8 @@ #define APPENDFSYNC_EVERYSEC 2 /* We can print the stacktrace, so our assert is defined this way: */ -#define redisAssert(_e) ((_e)?(void)0 : (_redisAssert(#_e),exit(1))) -static void _redisAssert(char *estr); +#define redisAssert(_e) ((_e)?(void)0 : (_redisAssert(#_e,__FILE__,__LINE__),_exit(1))) +static void _redisAssert(char *estr, char *file, int line); /*================================= Data types ============================== */ @@ -263,6 +268,7 @@ typedef struct redisDb { dict *dict; /* The keyspace for this DB */ dict *expires; /* Timeout of keys with a timeout set */ dict *blockingkeys; /* Keys with clients waiting for data (BLPOP) */ + dict *io_keys; /* Keys with clients waiting for VM I/O */ int id; } redisDb; @@ -292,8 +298,7 @@ typedef struct redisClient { list *reply; int sentlen; time_t lastinteraction; /* time of the last interaction, used for timeout */ - int flags; /* REDIS_CLOSE | REDIS_SLAVE | REDIS_MONITOR */ - /* REDIS_MULTI */ + int flags; /* REDIS_SLAVE | REDIS_MONITOR | REDIS_MULTI ... */ int slaveseldb; /* slave selected db, if this client is a slave */ int authenticated; /* when requirepass is non-NULL */ int replstate; /* replication state if this is a slave */ @@ -301,7 +306,7 @@ typedef struct redisClient { long repldboff; /* replication DB file offset */ off_t repldbsize; /* replication DB file size */ multiState mstate; /* MULTI/EXEC state */ - robj **blockingkeys; /* The key we waiting to terminate a blocking + robj **blockingkeys; /* The key we are waiting to terminate a blocking * operation such as BLPOP. Otherwise NULL. */ int blockingkeysnum; /* Number of blocking keys */ time_t blockingto; /* Blocking operation timeout. If UNIX current time @@ -330,7 +335,6 @@ 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 */ @@ -368,7 +372,8 @@ struct redisServer { int replstate; unsigned int maxclients; unsigned long long maxmemory; - unsigned int blockedclients; + 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; @@ -376,6 +381,7 @@ struct redisServer { 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; @@ -393,8 +399,11 @@ struct redisServer { 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_clients; /* All the clients waiting for SWAP I/O operations */ + 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 @@ -408,6 +417,7 @@ struct redisServer { 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); @@ -416,6 +426,10 @@ 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 { @@ -474,15 +488,17 @@ struct sharedObjectsStruct { static double R_Zero, R_PosInf, R_NegInf, R_Nan; /* VM threaded I/O request message */ -#define REDIS_IOJOB_LOAD 0 -#define REDIS_IOJOB_SWAP 1 -typedef struct iojon { +#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_* */ - int dbid; /* Redis database ID */ + 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 + 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; @@ -530,18 +546,36 @@ static void sendReplyToClientWritev(aeEventLoop *el, int fd, void *privdata, int static void initClientMultiState(redisClient *c); static void freeClientMultiState(redisClient *c); static void queueMultiCommand(redisClient *c, struct redisCommand *cmd); -static void unblockClient(redisClient *c); +static void unblockClientWaitingData(redisClient *c); static int handleClientsWaitingListPush(redisClient *c, robj *key, robj *ele); static void vmInit(void); static void vmMarkPagesFree(off_t page, off_t count); static robj *vmLoadObject(robj *key); static robj *vmPreviewObject(robj *key); -static int vmSwapOneObject(void); +static int vmSwapOneObjectBlocking(void); +static int vmSwapOneObjectThreaded(void); static int vmCanSwapOut(void); -static void freeOneObjectFromFreelist(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); @@ -611,6 +645,7 @@ static void zaddCommand(redisClient *c); static void zincrbyCommand(redisClient *c); static void zrangeCommand(redisClient *c); static void zrangebyscoreCommand(redisClient *c); +static void zcountCommand(redisClient *c); static void zrevrangeCommand(redisClient *c); static void zcardCommand(redisClient *c); static void zremCommand(redisClient *c); @@ -618,93 +653,98 @@ 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); /*================================= Globals ================================= */ /* Global vars */ static struct redisServer server; /* server global state */ static struct redisCommand cmdTable[] = { - {"get",getCommand,2,REDIS_CMD_INLINE}, - {"set",setCommand,3,REDIS_CMD_BULK|REDIS_CMD_DENYOOM}, - {"setnx",setnxCommand,3,REDIS_CMD_BULK|REDIS_CMD_DENYOOM}, - {"del",delCommand,-2,REDIS_CMD_INLINE}, - {"exists",existsCommand,2,REDIS_CMD_INLINE}, - {"incr",incrCommand,2,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM}, - {"decr",decrCommand,2,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM}, - {"mget",mgetCommand,-2,REDIS_CMD_INLINE}, - {"rpush",rpushCommand,3,REDIS_CMD_BULK|REDIS_CMD_DENYOOM}, - {"lpush",lpushCommand,3,REDIS_CMD_BULK|REDIS_CMD_DENYOOM}, - {"rpop",rpopCommand,2,REDIS_CMD_INLINE}, - {"lpop",lpopCommand,2,REDIS_CMD_INLINE}, - {"brpop",brpopCommand,-3,REDIS_CMD_INLINE}, - {"blpop",blpopCommand,-3,REDIS_CMD_INLINE}, - {"llen",llenCommand,2,REDIS_CMD_INLINE}, - {"lindex",lindexCommand,3,REDIS_CMD_INLINE}, - {"lset",lsetCommand,4,REDIS_CMD_BULK|REDIS_CMD_DENYOOM}, - {"lrange",lrangeCommand,4,REDIS_CMD_INLINE}, - {"ltrim",ltrimCommand,4,REDIS_CMD_INLINE}, - {"lrem",lremCommand,4,REDIS_CMD_BULK}, - {"rpoplpush",rpoplpushcommand,3,REDIS_CMD_BULK|REDIS_CMD_DENYOOM}, - {"sadd",saddCommand,3,REDIS_CMD_BULK|REDIS_CMD_DENYOOM}, - {"srem",sremCommand,3,REDIS_CMD_BULK}, - {"smove",smoveCommand,4,REDIS_CMD_BULK}, - {"sismember",sismemberCommand,3,REDIS_CMD_BULK}, - {"scard",scardCommand,2,REDIS_CMD_INLINE}, - {"spop",spopCommand,2,REDIS_CMD_INLINE}, - {"srandmember",srandmemberCommand,2,REDIS_CMD_INLINE}, - {"sinter",sinterCommand,-2,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM}, - {"sinterstore",sinterstoreCommand,-3,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM}, - {"sunion",sunionCommand,-2,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM}, - {"sunionstore",sunionstoreCommand,-3,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM}, - {"sdiff",sdiffCommand,-2,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM}, - {"sdiffstore",sdiffstoreCommand,-3,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM}, - {"smembers",sinterCommand,2,REDIS_CMD_INLINE}, - {"zadd",zaddCommand,4,REDIS_CMD_BULK|REDIS_CMD_DENYOOM}, - {"zincrby",zincrbyCommand,4,REDIS_CMD_BULK|REDIS_CMD_DENYOOM}, - {"zrem",zremCommand,3,REDIS_CMD_BULK}, - {"zremrangebyscore",zremrangebyscoreCommand,4,REDIS_CMD_INLINE}, - {"zrange",zrangeCommand,-4,REDIS_CMD_INLINE}, - {"zrangebyscore",zrangebyscoreCommand,-4,REDIS_CMD_INLINE}, - {"zrevrange",zrevrangeCommand,-4,REDIS_CMD_INLINE}, - {"zcard",zcardCommand,2,REDIS_CMD_INLINE}, - {"zscore",zscoreCommand,3,REDIS_CMD_BULK|REDIS_CMD_DENYOOM}, - {"incrby",incrbyCommand,3,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM}, - {"decrby",decrbyCommand,3,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM}, - {"getset",getsetCommand,3,REDIS_CMD_BULK|REDIS_CMD_DENYOOM}, - {"mset",msetCommand,-3,REDIS_CMD_BULK|REDIS_CMD_DENYOOM}, - {"msetnx",msetnxCommand,-3,REDIS_CMD_BULK|REDIS_CMD_DENYOOM}, - {"randomkey",randomkeyCommand,1,REDIS_CMD_INLINE}, - {"select",selectCommand,2,REDIS_CMD_INLINE}, - {"move",moveCommand,3,REDIS_CMD_INLINE}, - {"rename",renameCommand,3,REDIS_CMD_INLINE}, - {"renamenx",renamenxCommand,3,REDIS_CMD_INLINE}, - {"expire",expireCommand,3,REDIS_CMD_INLINE}, - {"expireat",expireatCommand,3,REDIS_CMD_INLINE}, - {"keys",keysCommand,2,REDIS_CMD_INLINE}, - {"dbsize",dbsizeCommand,1,REDIS_CMD_INLINE}, - {"auth",authCommand,2,REDIS_CMD_INLINE}, - {"ping",pingCommand,1,REDIS_CMD_INLINE}, - {"echo",echoCommand,2,REDIS_CMD_BULK}, - {"save",saveCommand,1,REDIS_CMD_INLINE}, - {"bgsave",bgsaveCommand,1,REDIS_CMD_INLINE}, - {"bgrewriteaof",bgrewriteaofCommand,1,REDIS_CMD_INLINE}, - {"shutdown",shutdownCommand,1,REDIS_CMD_INLINE}, - {"lastsave",lastsaveCommand,1,REDIS_CMD_INLINE}, - {"type",typeCommand,2,REDIS_CMD_INLINE}, - {"multi",multiCommand,1,REDIS_CMD_INLINE}, - {"exec",execCommand,1,REDIS_CMD_INLINE}, - {"sync",syncCommand,1,REDIS_CMD_INLINE}, - {"flushdb",flushdbCommand,1,REDIS_CMD_INLINE}, - {"flushall",flushallCommand,1,REDIS_CMD_INLINE}, - {"sort",sortCommand,-2,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM}, - {"info",infoCommand,1,REDIS_CMD_INLINE}, - {"monitor",monitorCommand,1,REDIS_CMD_INLINE}, - {"ttl",ttlCommand,2,REDIS_CMD_INLINE}, - {"slaveof",slaveofCommand,3,REDIS_CMD_INLINE}, - {"debug",debugCommand,-2,REDIS_CMD_INLINE}, - {NULL,NULL,0,0} + {"get",getCommand,2,REDIS_CMD_INLINE,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}, + {"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}, + {"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 ============================ */ @@ -841,13 +881,13 @@ 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); @@ -924,12 +964,27 @@ static int dictEncObjKeyCompare(void *privdata, const void *key1, static unsigned int dictEncObjHash(const void *key) { robj *o = (robj*) key; - o = getDecodedObject(o); - unsigned int hash = dictGenHashFunction(o->ptr, sdslen((sds)o->ptr)); - decrRefCount(o); - return hash; + if (o->encoding == REDIS_ENCODING_RAW) { + return dictGenHashFunction(o->ptr, sdslen((sds)o->ptr)); + } else { + if (o->encoding == REDIS_ENCODING_INT) { + char buf[32]; + int len; + + len = snprintf(buf,32,"%ld",(long)o->ptr); + return dictGenHashFunction((unsigned char*)buf, len); + } else { + unsigned int hash; + + o = getDecodedObject(o); + hash = dictGenHashFunction(o->ptr, sdslen((sds)o->ptr)); + decrRefCount(o); + return hash; + } + } } +/* Sets type and expires */ static dictType setDictType = { dictEncObjHash, /* hash function */ NULL, /* key dup */ @@ -939,6 +994,7 @@ static dictType setDictType = { NULL /* val destructor */ }; +/* Sorted sets hash (note: a skiplist is used in addition to the hash table) */ static dictType zsetDictType = { dictEncObjHash, /* hash function */ NULL, /* key dup */ @@ -948,6 +1004,7 @@ static dictType zsetDictType = { dictVanillaFree /* val destructor of malloc(sizeof(double)) */ }; +/* Db->dict */ static dictType hashDictType = { dictObjHash, /* hash function */ NULL, /* key dup */ @@ -957,8 +1014,19 @@ 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) */ + * 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 */ @@ -986,9 +1054,10 @@ 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 (server.maxidletime && !(c->flags & REDIS_SLAVE) && /* no timeout for slaves */ @@ -1000,7 +1069,7 @@ static void closeTimedoutClients(void) { } else if (c->flags & REDIS_BLOCKED) { if (c->blockingto != 0 && c->blockingto < now) { addReply(c,shared.nullmultibulk); - unblockClient(c); + unblockClientWaitingData(c); } } } @@ -1127,9 +1196,6 @@ static int serverCron(struct aeEventLoop *eventLoop, long long id, void *clientD * To access a global var is faster than calling time(NULL) */ server.unixtime = time(NULL); - /* Update the global state with the amount of used memory */ - server.usedmemory = zmalloc_used_memory(); - /* Show some info about non-empty databases */ for (j = 0; j < server.dbnum; j++) { long long size, used, vkeys; @@ -1156,12 +1222,12 @@ static int serverCron(struct aeEventLoop *eventLoop, long long id, void *clientD 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)) || server.blockedclients) + if ((server.maxidletime && !(loops % 10)) || server.blpop_blocked_clients) closeTimedoutClients(); /* Check if a background saving or AOF rewrite in progress terminated */ @@ -1230,15 +1296,23 @@ static int serverCron(struct aeEventLoop *eventLoop, long long id, void *clientD while (server.vm_enabled && zmalloc_used_memory() > server.vm_max_memory) { - if (listLength(server.objfreelist)) { - freeOneObjectFromFreelist(); - } else if (vmSwapOneObject() == REDIS_ERR) { - if ((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!"); - } - break; + 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; } } @@ -1252,6 +1326,38 @@ 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")); @@ -1325,13 +1431,15 @@ static void initServerConfig() { server.rdbcompression = 1; server.sharingpoolsize = 1024; server.maxclients = 0; - server.blockedclients = 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(); @@ -1360,6 +1468,11 @@ 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(); @@ -1375,8 +1488,10 @@ 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; @@ -1385,7 +1500,6 @@ static void initServer() { 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); @@ -1393,12 +1507,6 @@ static void initServer() { 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.vm_enabled) { - /* 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"); - } if (server.appendonly) { server.appendfd = open(server.appendfilename,O_WRONLY|O_APPEND|O_CREAT,0644); @@ -1572,15 +1680,18 @@ 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) { @@ -1623,14 +1734,14 @@ static void freeClient(redisClient *c) { listNode *ln; /* Note that if the client we are freeing is blocked into a blocking - * call, we have to set querybuf to NULL *before* to call unblockClient() - * to avoid processInputBuffer() will get called. Also it is important - * to remove the file events after this, because this call adds - * the READABLE event. */ + * call, we have to set querybuf to NULL *before* to call + * unblockClientWaitingData() to avoid processInputBuffer() will get + * called. Also it is important to remove the file events after + * this, because this call adds the READABLE event. */ sdsfree(c->querybuf); c->querybuf = NULL; if (c->flags & REDIS_BLOCKED) - unblockClient(c); + unblockClientWaitingData(c); aeDeleteFileEvent(server.el,c->fd,AE_READABLE); aeDeleteFileEvent(server.el,c->fd,AE_WRITABLE); @@ -1641,12 +1752,19 @@ static void freeClient(redisClient *c) { ln = listSearchKey(server.clients,c); redisAssert(ln != NULL); listDelNode(server.clients,ln); - /* Remove from the list of clients waiting for VM operations */ - if (server.vm_enabled && listLength(c->io_keys)) { - ln = listSearchKey(server.io_clients,c); - if (ln) listDelNode(server.io_clients,ln); - listRelease(c->io_keys); + /* 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) @@ -1671,10 +1789,11 @@ 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; @@ -1956,6 +2075,9 @@ static int processCommand(redisClient *c) { freeClient(c); return 0; } + + /* Now lookup the command and check ASAP about trivial error conditions + * such wrong arity, bad command name and so forth. */ cmd = lookupCommand(c->argv[0]->ptr); if (!cmd) { addReplySds(c, @@ -1976,6 +2098,7 @@ 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]); @@ -1997,6 +2120,8 @@ 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; } } @@ -2018,24 +2143,23 @@ static int processCommand(redisClient *c) { } /* Exec the command */ - if (c->flags & REDIS_MULTI && cmd->proc != execCommand) { + if (c->flags & REDIS_MULTI && cmd->proc != execCommand && cmd->proc != discardCommand) { queueMultiCommand(c,cmd); addReply(c,shared.queued); } else { + if (server.vm_enabled && server.vm_max_threads > 0 && + blockClientOnSwappedKeys(cmd,c)) return 1; call(c,cmd); } /* Prepare the client for the next command */ - if (c->flags & REDIS_CLOSE) { - freeClient(c); - return 0; - } resetClient(c); return 1; } static void replicationFeedSlaves(list *slaves, struct redisCommand *cmd, int dictid, robj **argv, int argc) { listNode *ln; + listIter li; int outc = 0, j; robj **outv; /* (args*2)+1 is enough room for args, spaces, newlines */ @@ -2066,8 +2190,8 @@ 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 */ @@ -2212,7 +2336,8 @@ 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) { @@ -2224,7 +2349,7 @@ static int selectDb(redisClient *c, int id) { static void *dupClientReplyValue(void *o) { incrRefCount((robj*)o); - return 0; + return o; } static redisClient *createClient(int fd) { @@ -2292,6 +2417,14 @@ static void addReplyDouble(redisClient *c, double d) { (unsigned long) strlen(buf),buf)); } +static void addReplyLong(redisClient *c, long l) { + char buf[128]; + size_t len; + + len = snprintf(buf,sizeof(buf),":%ld\r\n",l); + addReplySds(c,sdsnewlen(buf,len)); +} + static void addReplyBulkLen(redisClient *c, robj *obj) { size_t len; @@ -2354,12 +2487,15 @@ 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 { if (server.vm_enabled) { + pthread_mutex_unlock(&server.obj_freelist_mutex); o = zmalloc(sizeof(*o)); } else { o = zmalloc(sizeof(*o)-sizeof(struct redisObjectVM)); @@ -2370,6 +2506,10 @@ static robj *createObject(int type, void *ptr) { 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; } @@ -2381,6 +2521,7 @@ static robj *createStringObject(char *ptr, size_t len) { } static robj *dupStringObject(robj *o) { + assert(o->encoding == REDIS_ENCODING_RAW); return createStringObject(o->ptr,sdslen(o->ptr)); } @@ -2438,7 +2579,8 @@ static void incrRefCount(robj *o) { static void decrRefCount(void *obj) { robj *o = obj; - /* Object is swapped out, or in the process of being loaded. */ + /* 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)) { @@ -2449,9 +2591,11 @@ static void decrRefCount(void *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; } @@ -2467,9 +2611,11 @@ static void decrRefCount(void *obj) { case REDIS_HASH: freeHashObject(o); 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); } } @@ -2490,10 +2636,16 @@ static robj *lookupKey(redisDb *db, robj *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; @@ -2822,21 +2974,16 @@ static int rdbSaveStringObjectRaw(FILE *fp, robj *obj) { static int rdbSaveStringObject(FILE *fp, robj *obj) { int retval; - if (obj->storage == REDIS_VM_MEMORY && - obj->encoding != REDIS_ENCODING_RAW) - { + /* Avoid incr/decr ref count business when possible. + * This plays well with copy-on-write given that we are probably + * in a child process (BGSAVE). Also this makes sure key objects + * of swapped objects are not incRefCount-ed (an assert does not allow + * this in order to avoid bugs) */ + if (obj->encoding != REDIS_ENCODING_RAW) { obj = getDecodedObject(obj); retval = rdbSaveStringObjectRaw(fp,obj); decrRefCount(obj); } else { - /* This is a fast path when we are sure the object is not encoded. - * Note that's any *faster* actually as we needed to add the conditional - * but because this may happen in a background process we don't want - * to touch the object fields with incr/decrRefCount in order to - * preveny copy on write of pages. - * - * Also incrRefCount() will have a failing assert() if we try to call - * it against an object with storage != REDIS_VM_MEMORY. */ retval = rdbSaveStringObjectRaw(fp,obj); } return retval; @@ -2877,11 +3024,12 @@ static int rdbSaveObject(FILE *fp, robj *o) { } else if (o->type == REDIS_LIST) { /* Save a list value */ list *list = o->ptr; + listIter li; listNode *ln; - listRewind(list); if (rdbSaveLen(fp,listLength(list)) == -1) return -1; - while((ln = listYield(list))) { + listRewind(list,&li); + while((ln = listNext(&li))) { robj *eleobj = listNodeValue(ln); if (rdbSaveStringObject(fp,eleobj) == -1) return -1; @@ -2924,20 +3072,16 @@ static int rdbSaveObject(FILE *fp, robj *o) { * 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) { - static FILE *fp = NULL; - - if (fp == NULL) fp = fopen("/dev/null","w"); - assert(fp != NULL); - +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) { - off_t bytes = rdbSavedObjectLen(o); +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; } @@ -2951,6 +3095,12 @@ 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) { @@ -2995,19 +3145,15 @@ static int rdbSave(char *filename) { if (rdbSaveObject(fp,o) == -1) goto werr; } else { /* REDIS_VM_SWAPPED or REDIS_VM_LOADING */ - robj *po, *newkey; + robj *po; /* Get a preview of the object in memory */ po = vmPreviewObject(key); - /* Also duplicate the key object, to pass around a standard - * string object. */ - newkey = dupStringObject(key); /* Save type, key, value */ if (rdbSaveType(fp,key->vtype) == -1) goto werr; - if (rdbSaveStringObject(fp,newkey) == -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); - decrRefCount(newkey); } } dictReleaseIterator(di); @@ -3044,13 +3190,15 @@ static int rdbSaveBackground(char *filename) { pid_t childpid; if (server.bgsavechildpid != -1) return REDIS_ERR; + if (server.vm_enabled) waitEmptyIOJobsQueue(); if ((childpid = fork()) == 0) { /* Child */ + if (server.vm_enabled) vmReopenSwapFile(); close(server.fd); if (rdbSave(filename) == REDIS_OK) { - exit(0); + _exit(0); } else { - exit(1); + _exit(1); } } else { /* Parent */ @@ -3220,6 +3368,10 @@ static robj *rdbLoadObject(int type, FILE *fp) { 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; @@ -3329,7 +3481,7 @@ static int rdbLoad(char *filename) { loadedkeys++; if (server.vm_enabled && (loadedkeys % 5000) == 0) { while (zmalloc_used_memory() > server.vm_max_memory) { - if (vmSwapOneObject() == REDIS_ERR) break; + if (vmSwapOneObjectBlocking() == REDIS_ERR) break; } } } @@ -3562,6 +3714,52 @@ 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)); +} + /* ========================= Type agnostic commands ========================= */ static void delCommand(redisClient *c) { @@ -3622,7 +3820,7 @@ static void keysCommand(redisClient *c) { dictEntry *de; sds pattern = c->argv[1]->ptr; int plen = sdslen(pattern); - unsigned long numkeys = 0, keyslen = 0; + unsigned long numkeys = 0; robj *lenobj = createObject(REDIS_STRING,NULL); di = dictGetIterator(c->db->dict); @@ -3635,17 +3833,15 @@ 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) { @@ -3716,6 +3912,7 @@ static void shutdownCommand(redisClient *c) { if (server.appendonly) { /* Append only file: fsync() the AOF and exit */ fsync(server.appendfd); + if (server.vm_enabled) unlink(server.vm_swap_file); exit(0); } else { /* Snapshotting. Perform a SYNC SAVE and exit */ @@ -3724,6 +3921,7 @@ static void shutdownCommand(redisClient *c) { unlink(server.pidfile); redisLog(REDIS_WARNING,"%zu bytes used at exit",zmalloc_used_memory()); redisLog(REDIS_WARNING,"Server exit now, bye bye..."); + if (server.vm_enabled) unlink(server.vm_swap_file); exit(0); } else { /* Ooops.. error saving! The best we can do is to continue operating. @@ -5009,28 +5207,64 @@ static void zrevrangeCommand(redisClient *c) { zrangeGenericCommand(c,1); } -static void zrangebyscoreCommand(redisClient *c) { +/* This command implements both ZRANGEBYSCORE and ZCOUNT. + * If justcount is non-zero, just the count is returned. */ +static void genericZrangebyscoreCommand(redisClient *c, int justcount) { robj *o; - double min = strtod(c->argv[2]->ptr,NULL); - double max = strtod(c->argv[3]->ptr,NULL); + double min, max; + int minex = 0, maxex = 0; /* are min or max exclusive? */ int offset = 0, limit = -1; + int withscores = 0; + int badsyntax = 0; + + /* Parse the min-max interval. If one of the values is prefixed + * by the "(" character, it's considered "open". For instance + * ZRANGEBYSCORE zset (1.5 (2.5 will match min < x < max + * ZRANGEBYSCORE zset 1.5 2.5 will instead match min <= x <= max */ + if (((char*)c->argv[2]->ptr)[0] == '(') { + min = strtod((char*)c->argv[2]->ptr+1,NULL); + minex = 1; + } else { + min = strtod(c->argv[2]->ptr,NULL); + } + if (((char*)c->argv[3]->ptr)[0] == '(') { + max = strtod((char*)c->argv[3]->ptr+1,NULL); + maxex = 1; + } else { + max = strtod(c->argv[3]->ptr,NULL); + } - if (c->argc != 4 && c->argc != 7) { + /* Parse "WITHSCORES": note that if the command was called with + * the name ZCOUNT then we are sure that c->argc == 4, so we'll never + * enter the following paths to parse WITHSCORES and LIMIT. */ + if (c->argc == 5 || c->argc == 8) { + if (strcasecmp(c->argv[c->argc-1]->ptr,"withscores") == 0) + withscores = 1; + else + badsyntax = 1; + } + if (c->argc != (4 + withscores) && c->argc != (7 + withscores)) + badsyntax = 1; + if (badsyntax) { addReplySds(c, sdsnew("-ERR wrong number of arguments for ZRANGEBYSCORE\r\n")); return; - } else if (c->argc == 7 && strcasecmp(c->argv[4]->ptr,"limit")) { + } + + /* Parse "LIMIT" */ + if (c->argc == (7 + withscores) && strcasecmp(c->argv[4]->ptr,"limit")) { addReply(c,shared.syntaxerr); return; - } else if (c->argc == 7) { + } else if (c->argc == (7 + withscores)) { offset = atoi(c->argv[5]->ptr); limit = atoi(c->argv[6]->ptr); if (offset < 0) offset = 0; } + /* Ok, lookup the key and get the range */ o = lookupKeyRead(c->db,c->argv[1]); if (o == NULL) { - addReply(c,shared.nullmultibulk); + addReply(c,justcount ? shared.czero : shared.nullmultibulk); } else { if (o->type != REDIS_ZSET) { addReply(c,shared.wrongtypeerr); @@ -5038,14 +5272,17 @@ 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; } @@ -5053,30 +5290,49 @@ 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); - decrRefCount(lenobj); + if (!justcount) { + lenobj = createObject(REDIS_STRING,NULL); + addReply(c,lenobj); + decrRefCount(lenobj); + } - while(ln && ln->score <= max) { + while(ln && (maxex ? (ln->score < max) : (ln->score <= max))) { if (offset) { offset--; ln = ln->forward[0]; continue; } if (limit == 0) break; - ele = ln->obj; - addReplyBulkLen(c,ele); - addReply(c,ele); - addReply(c,shared.crlf); + if (!justcount) { + ele = ln->obj; + 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--; } - lenobj->ptr = sdscatprintf(sdsempty(),"*%d\r\n",rangelen); + if (justcount) { + addReplyLong(c,(long)rangelen); + } else { + lenobj->ptr = sdscatprintf(sdsempty(),"*%lu\r\n", + withscores ? (rangelen*2) : rangelen); + } } } } +static void zrangebyscoreCommand(redisClient *c) { + genericZrangebyscoreCommand(c,0); +} + +static void zcountCommand(redisClient *c) { + genericZrangebyscoreCommand(c,1); +} + static void zcardCommand(redisClient *c) { robj *o; zset *zs; @@ -5325,9 +5581,10 @@ static void sortCommand(redisClient *c) { if (sortval->type == REDIS_LIST) { list *list = sortval->ptr; listNode *ln; + listIter li; - listRewind(list); - while((ln = listYield(list))) { + listRewind(list,&li); + while((ln = listNext(&li))) { robj *ele = ln->value; vector[j].obj = ele; vector[j].u.score = 0; @@ -5423,13 +5680,15 @@ 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); @@ -5454,12 +5713,14 @@ 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); @@ -5513,7 +5774,7 @@ static void bytesToHuman(char *s, unsigned long long n) { sprintf(s,"%.2fM",d); } else if (n < (1024LL*1024*1024*1024)) { d = (double)n/(1024LL*1024*1024); - sprintf(s,"%.2fM",d); + sprintf(s,"%.2fG",d); } } @@ -5526,7 +5787,7 @@ static sds genRedisInfoString(void) { int j; char hmem[64]; - bytesToHuman(hmem,server.usedmemory); + bytesToHuman(hmem,zmalloc_used_memory()); info = sdscatprintf(sdsempty(), "redis_version:%s\r\n" "arch_bits:%s\r\n" @@ -5555,8 +5816,8 @@ static sds genRedisInfoString(void) { uptime/(3600*24), listLength(server.clients)-listLength(server.slaves), listLength(server.slaves), - server.blockedclients, - server.usedmemory, + server.blpop_blocked_clients, + zmalloc_used_memory(), hmem, server.dirty, server.bgsavechildpid != -1, @@ -5581,6 +5842,7 @@ static sds genRedisInfoString(void) { ); } if (server.vm_enabled) { + lockThreadedIO(); info = sdscatprintf(info, "vm_conf_max_memory:%llu\r\n" "vm_conf_page_size:%llu\r\n" @@ -5589,14 +5851,25 @@ static sds genRedisInfoString(void) { "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 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; @@ -5778,6 +6051,18 @@ static void multiCommand(redisClient *c) { addReply(c,shared.ok); } +static void discardCommand(redisClient *c) { + if (!(c->flags & REDIS_MULTI)) { + addReplySds(c,sdsnew("-ERR DISCARD without MULTI\r\n")); + return; + } + + freeClientMultiState(c); + initClientMultiState(c); + c->flags &= (~REDIS_MULTI); + addReply(c,shared.ok); +} + static void execCommand(redisClient *c) { int j; robj **orig_argv; @@ -5866,12 +6151,11 @@ static void blockForKeys(redisClient *c, robj **keys, int numkeys, time_t timeou } /* Mark the client as a blocked client */ c->flags |= REDIS_BLOCKED; - aeDeleteFileEvent(server.el,c->fd,AE_READABLE); - server.blockedclients++; + server.blpop_blocked_clients++; } /* Unblock a client that's waiting in a blocking operation such as BLPOP */ -static void unblockClient(redisClient *c) { +static void unblockClientWaitingData(redisClient *c) { dictEntry *de; list *l; int j; @@ -5893,18 +6177,12 @@ static void unblockClient(redisClient *c) { zfree(c->blockingkeys); c->blockingkeys = NULL; c->flags &= (~REDIS_BLOCKED); - server.blockedclients--; - /* Ok now we are ready to get read events from socket, note that we - * can't trap errors here as it's possible that unblockClients() is - * called from freeClient() itself, and the only thing we can do - * if we failed to register the READABLE event is to kill the client. - * Still the following function should never fail in the real world as - * we are sure the file descriptor is sane, and we exit on out of mem. */ - aeCreateFileEvent(server.el, c->fd, AE_READABLE, readQueryFromClient, c); - /* As a final step we want to process data if there is some command waiting - * in the input buffer. Note that this is safe even if unblockClient() - * gets called from freeClient() because freeClient() will be smart - * enough to call this function *after* c->querybuf was set to NULL. */ + server.blpop_blocked_clients--; + /* We want to process data if there is some command waiting + * in the input buffer. Note that this is safe even if + * unblockClientWaitingData() gets called from freeClient() because + * freeClient() will be smart enough to call this function + * *after* c->querybuf was set to NULL. */ if (c->querybuf && sdslen(c->querybuf) > 0) processInputBuffer(c); } @@ -5938,7 +6216,7 @@ static int handleClientsWaitingListPush(redisClient *c, robj *key, robj *ele) { addReplyBulkLen(receiver,ele); addReply(receiver,ele); addReply(receiver,shared.crlf); - unblockClient(receiver); + unblockClientWaitingData(receiver); return 1; } @@ -6089,9 +6367,10 @@ static void syncCommand(redisClient *c) { * registering differences since the server forked to save */ redisClient *slave; listNode *ln; + listIter li; - listRewind(server.slaves); - while((ln = listYield(server.slaves))) { + listRewind(server.slaves,&li); + while((ln = listNext(&li))) { slave = ln->value; if (slave->replstate == REDIS_REPL_WAIT_BGSAVE_END) break; } @@ -6188,9 +6467,10 @@ 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) { @@ -6222,9 +6502,11 @@ 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) @@ -6362,16 +6644,25 @@ static void slaveofCommand(redisClient *c) { /* ============================ Maxmemory directive ======================== */ -/* 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. */ -static void freeOneObjectFromFreelist(void) { +/* 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; - listNode *head = listFirst(server.objfreelist); - o = listNodeValue(head); - listDelNode(server.objfreelist,head); - zfree(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 @@ -6387,35 +6678,32 @@ static void freeOneObjectFromFreelist(void) { */ static void freeMemoryIfNeeded(void) { while (server.maxmemory && zmalloc_used_memory() > server.maxmemory) { - if (listLength(server.objfreelist)) { - freeOneObjectFromFreelist(); - } 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... */ } } @@ -6605,7 +6893,7 @@ int loadAppendOnlyFile(char *filename) { loadedkeys++; if (server.vm_enabled && (loadedkeys % 5000) == 0) { while (zmalloc_used_memory() > server.vm_max_memory) { - if (vmSwapOneObject() == REDIS_ERR) break; + if (vmSwapOneObjectBlocking() == REDIS_ERR) break; } } } @@ -6628,16 +6916,26 @@ fmterr: /* Write an object into a file in the bulk format $\r\n\r\n */ static int fwriteBulk(FILE *fp, robj *obj) { char buf[128]; - obj = getDecodedObject(obj); + int decrrc = 0; + + /* Avoid the incr/decr ref count business if possible to help + * copy-on-write (we are often in a child process when this function + * is called). + * Also makes sure that key objects don't get incrRefCount-ed when VM + * is enabled */ + if (obj->encoding != REDIS_ENCODING_RAW) { + obj = getDecodedObject(obj); + decrrc = 1; + } snprintf(buf,sizeof(buf),"$%ld\r\n",(long)sdslen(obj->ptr)); if (fwrite(buf,strlen(buf),1,fp) == 0) goto err; if (sdslen(obj->ptr) && fwrite(obj->ptr,sdslen(obj->ptr),1,fp) == 0) goto err; if (fwrite("\r\n",2,1,fp) == 0) goto err; - decrRefCount(obj); + if (decrrc) decrRefCount(obj); return 1; err: - decrRefCount(obj); + if (decrrc) decrRefCount(obj); return 0; } @@ -6703,13 +7001,16 @@ static int rewriteAppendOnlyFile(char *filename) { 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); - key = dupStringObject(key); swapped = 1; } expiretime = getExpire(db,key); @@ -6726,9 +7027,10 @@ static int rewriteAppendOnlyFile(char *filename) { /* Emit the RPUSHes needed to rebuild the list */ list *list = o->ptr; listNode *ln; + listIter li; - listRewind(list); - while((ln = listYield(list))) { + listRewind(list,&li); + while((ln = listNext(&li))) { char cmd[]="*3\r\n$5\r\nRPUSH\r\n"; robj *eleobj = listNodeValue(ln); @@ -6780,12 +7082,7 @@ static int rewriteAppendOnlyFile(char *filename) { if (fwriteBulk(fp,key) == 0) goto werr; if (fwriteBulkLong(fp,expiretime) == 0) goto werr; } - /* We created a few temp objects if the key->value pair - * was about a swapped out object. Free both. */ - if (swapped) { - decrRefCount(key); - decrRefCount(o); - } + if (swapped) decrRefCount(o); } dictReleaseIterator(di); } @@ -6829,16 +7126,18 @@ 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]; - close(server.fd); + 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); + _exit(0); } else { - exit(1); + _exit(1); } } else { /* Parent */ @@ -6903,13 +7202,40 @@ static void aofRemoveTempFile(pid_t childpid) { */ /* =================== 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() */ - server.vm_fp = fopen("/tmp/redisvm","w+b"); + 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. Exiting."); + redisLog(REDIS_WARNING, + "Impossible to open the swap file: %s. Exiting.", + strerror(errno)); exit(1); } server.vm_fd = fileno(server.vm_fp); @@ -6932,16 +7258,15 @@ static void vmInit(void) { 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); - /* Try to remove the swap file, so the OS will really delete it from the - * file system when Redis exists. */ - unlink("/tmp/redisvm"); /* Initialize threaded I/O (used by Virtual Memory) */ server.io_newjobs = listCreate(); server.io_processing = listCreate(); server.io_processed = listCreate(); - server.io_clients = 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." @@ -6951,15 +7276,23 @@ static void vmInit(void) { 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; @@ -7043,6 +7380,7 @@ static int vmFindContiguousPages(off_t *first, int n) { 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 { @@ -7067,24 +7405,34 @@ static int vmFindContiguousPages(off_t *first, int n) { 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 vmSwapObject(robj *key, robj *val) { - off_t pages = rdbSavedObjectPages(val); +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 (fseeko(server.vm_fp,page*server.vm_page_size,SEEK_SET) == -1) { - redisLog(REDIS_WARNING, - "Critical VM problem in vmSwapObject(): can't seek: %s", - strerror(errno)); - return REDIS_ERR; - } - rdbSaveObject(server.vm_fp,val); + if (vmWriteObjectOnSwap(val,page) == REDIS_ERR) return REDIS_ERR; key->vm.page = page; key->vm.usedpages = pages; key->storage = REDIS_VM_SWAPPED; @@ -7096,10 +7444,28 @@ static int vmSwapObject(robj *key, robj *val) { (unsigned long long) page, (unsigned long long) pages); server.vm_stats_swapped_objects++; server.vm_stats_swapouts++; - fflush(server.vm_fp); 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. * @@ -7108,18 +7474,8 @@ static int vmSwapObject(robj *key, robj *val) { static robj *vmGenericLoadObject(robj *key, int preview) { robj *val; - redisAssert(key->storage == REDIS_VM_SWAPPED); - if (fseeko(server.vm_fp,key->vm.page*server.vm_page_size,SEEK_SET) == -1) { - redisLog(REDIS_WARNING, - "Unrecoverable VM problem in vmLoadObject(): can't seek: %s", - strerror(errno)); - exit(1); - } - val = rdbLoadObject(key->vtype,server.vm_fp); - if (val == NULL) { - redisLog(REDIS_WARNING, "Unrecoverable VM problem in vmLoadObject(): can't load object from swap file: %s", strerror(errno)); - exit(1); - } + 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; @@ -7216,21 +7572,28 @@ static double computeObjectSwappability(robj *o) { } break; } - return (double)asize*log(1+asize); + 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. */ -static int vmSwapOneObject(void) { + * 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; - int maxtries = 1000; + /* 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++) { @@ -7241,7 +7604,14 @@ static int vmSwapOneObject(void) { de = dictGetRandomKey(db->dict); key = dictGetEntryKey(de); val = dictGetEntryVal(de); - if (key->storage != REDIS_VM_MEMORY) { + /* 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; } @@ -7249,13 +7619,11 @@ static int vmSwapOneObject(void) { if (!best || swappability > best_swappability) { best = de; best_swappability = swappability; + best_db = db; } } } - if (best == NULL) { - redisLog(REDIS_DEBUG,"No swappable key found!"); - return REDIS_ERR; - } + if (best == NULL) return REDIS_ERR; key = dictGetEntryKey(best); val = dictGetEntryVal(best); @@ -7269,14 +7637,27 @@ static int vmSwapOneObject(void) { key = dictGetEntryKey(best) = newkey; } /* Swap it */ - if (vmSwapObject(key,val) == REDIS_OK) { - dictGetEntryVal(best) = NULL; + if (usethreads) { + vmSwapObjectThreaded(key,val,best_db); return REDIS_OK; } else { - return REDIS_ERR; + 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. */ @@ -7299,6 +7680,15 @@ static int deleteIfSwapped(redisDb *db, robj *key) { /* =================== 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. */ @@ -7306,7 +7696,7 @@ static void vmThreadedIOCompletedJob(aeEventLoop *el, int fd, void *privdata, int mask) { char buf[1]; - int retval; + int retval, processed = 0, toprocess = -1, trytoswap = 1; REDIS_NOTUSED(el); REDIS_NOTUSED(mask); REDIS_NOTUSED(privdata); @@ -7314,7 +7704,121 @@ static void vmThreadedIOCompletedJob(aeEventLoop *el, int fd, void *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, @@ -7335,35 +7839,70 @@ static void unlockThreadedIO(void) { * processed, otherwise make sure to flag it as canceled. */ static void vmCancelThreadedIOJob(robj *o) { list *lists[3] = { - server.io_newjobs, server.io_processing, server.io_processed + 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]); - while ((ln = listYield(lists[i])) != NULL) { + 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 not yet processed the best thing to do + /* If the job was yet not processed the best thing to do * is to remove it from the queue at all */ - decrRefCount(job->key); - if (job->type == REDIS_IOJOB_SWAP) - decrRefCount(job->val); + 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) @@ -7377,6 +7916,301 @@ static void vmCancelThreadedIOJob(robj *o) { 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) { @@ -7412,12 +8246,13 @@ static void debugCommand(redisClient *c) { } key = dictGetEntryKey(de); val = dictGetEntryVal(de); - if (server.vm_enabled && key->storage == REDIS_VM_MEMORY) { + 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, rdbSavedObjectLen(val))); + val->encoding, (long long) rdbSavedObjectLen(val,NULL))); } else { addReplySds(c,sdscatprintf(sdsempty(), "+Key at:%p refcount:%d, value swapped at: page %llu " @@ -7448,7 +8283,7 @@ static void debugCommand(redisClient *c) { /* Swap it */ if (key->storage != REDIS_VM_MEMORY) { addReplySds(c,sdsnew("-ERR This key is not in memory\r\n")); - } else if (vmSwapObject(key,val) == REDIS_OK) { + } else if (vmSwapObjectBlocking(key,val) == REDIS_OK) { dictGetEntryVal(de) = NULL; addReply(c,shared.ok); } else { @@ -7460,9 +8295,9 @@ static void debugCommand(redisClient *c) { } } -static void _redisAssert(char *estr) { +static void _redisAssert(char *estr, char *file, int line) { redisLog(REDIS_WARNING,"=== ASSERTION FAILED ==="); - redisLog(REDIS_WARNING,"==> %s\n",estr); + redisLog(REDIS_WARNING,"==> %s:%d '%s' is not true\n",file,line,estr); #ifdef HAVE_BACKTRACE redisLog(REDIS_WARNING,"(forcing SIGSEGV in order to print the stack trace)"); *((char*)-1) = 'x'; @@ -7498,7 +8333,6 @@ static void daemonize(void) { FILE *fp; if (fork() != 0) exit(0); /* parent exits */ - printf("New pid: %d\n", getpid()); setsid(); /* create a new session */ /* Every output goes to /dev/null. If Redis is daemonized but @@ -7519,6 +8353,8 @@ static void daemonize(void) { } int main(int argc, char **argv) { + time_t start; + initServerConfig(); if (argc == 2) { resetServerSaveParams(); @@ -7535,14 +8371,16 @@ int main(int argc, char **argv) { #ifdef __linux__ linuxOvercommitMemoryWarning(); #endif + start = time(NULL); if (server.appendonly) { if (loadAppendOnlyFile(server.appendfilename) == REDIS_OK) - redisLog(REDIS_NOTICE,"DB loaded from append only file"); + redisLog(REDIS_NOTICE,"DB loaded from append only file: %ld seconds",time(NULL)-start); } else { if (rdbLoad(server.dbfilename) == REDIS_OK) - redisLog(REDIS_NOTICE,"DB loaded from disk"); + redisLog(REDIS_NOTICE,"DB loaded from disk: %ld seconds",time(NULL)-start); } 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; @@ -7613,7 +8451,7 @@ static void segvHandler(int sig, siginfo_t *info, void *secret) { } } /* free(messages); Don't call free() with possibly corrupted memory. */ - exit(0); + _exit(0); } static void setupSigSegvAction(void) {