X-Git-Url: https://git.saurik.com/redis.git/blobdiff_plain/52d23ddbea757a8a544b2e9042e008025e54c93b..78409a0f84f63630ad49ba4dba4467b523bc6c55:/redis.c diff --git a/redis.c b/redis.c index 1ead01f8..e5f2350d 100644 --- a/redis.c +++ b/redis.c @@ -75,6 +75,7 @@ #include "zmalloc.h" /* total memory usage aware version of malloc/free */ #include "lzf.h" /* LZF compression library */ #include "pqsort.h" /* Partial qsort for SORT+LIMIT */ +#include "zipmap.h" /* Error codes */ #define REDIS_OK 0 @@ -118,9 +119,17 @@ #define REDIS_ZSET 3 #define REDIS_HASH 4 -/* Objects encoding */ +/* Objects encoding. Some kind of objects like Strings and Hashes can be + * internally represented in multiple ways. The 'encoding' field of the object + * is set to one of this fields for this object. */ #define REDIS_ENCODING_RAW 0 /* Raw representation */ #define REDIS_ENCODING_INT 1 /* Encoded as integer */ +#define REDIS_ENCODING_ZIPMAP 2 /* Encoded as zipmap */ +#define REDIS_ENCODING_HT 3 /* Encoded as an hash table */ + +static char* strencoding[] = { + "raw", "int", "zipmap", "hashtable" +}; /* Object types only used for dumping to disk */ #define REDIS_EXPIRETIME 253 @@ -221,6 +230,10 @@ #define APPENDFSYNC_ALWAYS 1 #define APPENDFSYNC_EVERYSEC 2 +/* Hashes related defaults */ +#define REDIS_HASH_MAX_ZIPMAP_ENTRIES 64 +#define REDIS_HASH_MAX_ZIPMAP_VALUE 512 + /* We can print the stacktrace, so our assert is defined this way: */ #define redisAssert(_e) ((_e)?(void)0 : (_redisAssert(#_e,__FILE__,__LINE__),_exit(1))) static void _redisAssert(char *estr, char *file, int line); @@ -386,6 +399,9 @@ struct redisServer { off_t vm_page_size; off_t vm_pages; unsigned long long vm_max_memory; + /* Hashes config */ + size_t hash_max_zipmap_entries; + size_t hash_max_zipmap_value; /* Virtual memory state */ FILE *vm_fp; int vm_fd; @@ -456,6 +472,7 @@ typedef struct _redisSortOperation { typedef struct zskiplistNode { struct zskiplistNode **forward; struct zskiplistNode *backward; + unsigned int *span; double score; robj *obj; } zskiplistNode; @@ -577,6 +594,7 @@ static void readQueryFromClient(aeEventLoop *el, int fd, void *privdata, int mas static struct redisCommand *lookupCommand(char *name); static void call(redisClient *c, struct redisCommand *cmd); static void resetClient(redisClient *c); +static void convertToRealHash(robj *o); static void authCommand(redisClient *c); static void pingCommand(redisClient *c); @@ -658,6 +676,19 @@ static void discardCommand(redisClient *c); static void blpopCommand(redisClient *c); static void brpopCommand(redisClient *c); static void appendCommand(redisClient *c); +static void substrCommand(redisClient *c); +static void zrankCommand(redisClient *c); +static void zrevrankCommand(redisClient *c); +static void hsetCommand(redisClient *c); +static void hgetCommand(redisClient *c); +static void hdelCommand(redisClient *c); +static void hlenCommand(redisClient *c); +static void zremrangebyrankCommand(redisClient *c); +static void zunionCommand(redisClient *c); +static void zinterCommand(redisClient *c); +static void hkeysCommand(redisClient *c); +static void hvalsCommand(redisClient *c); +static void hgetallCommand(redisClient *c); /*================================= Globals ================================= */ @@ -668,6 +699,7 @@ static struct redisCommand cmdTable[] = { {"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}, @@ -704,12 +736,24 @@ static struct redisCommand cmdTable[] = { {"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}, + {"zremrangebyrank",zremrangebyrankCommand,4,REDIS_CMD_INLINE,1,1,1}, + {"zunion",zunionCommand,-4,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM,0,0,0}, + {"zinter",zinterCommand,-4,REDIS_CMD_INLINE|REDIS_CMD_DENYOOM,0,0,0}, {"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_BULK,1,1,1}, + {"zrevrank",zrevrankCommand,3,REDIS_CMD_BULK,1,1,1}, + {"hset",hsetCommand,4,REDIS_CMD_BULK|REDIS_CMD_DENYOOM,1,1,1}, + {"hget",hgetCommand,3,REDIS_CMD_BULK,1,1,1}, + {"hdel",hdelCommand,3,REDIS_CMD_BULK,1,1,1}, + {"hlen",hlenCommand,2,REDIS_CMD_INLINE,1,1,1}, + {"hkeys",hkeysCommand,2,REDIS_CMD_INLINE,1,1,1}, + {"hvals",hvalsCommand,2,REDIS_CMD_INLINE,1,1,1}, + {"hgetall",hgetallCommand,2,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}, @@ -1006,7 +1050,7 @@ static dictType zsetDictType = { }; /* Db->dict */ -static dictType hashDictType = { +static dictType dbDictType = { dictObjHash, /* hash function */ NULL, /* key dup */ NULL, /* val dup */ @@ -1025,6 +1069,16 @@ static dictType keyptrDictType = { NULL /* val destructor */ }; +/* Hash type hash table (note that small hashes are represented with zimpaps) */ +static dictType hashDictType = { + dictEncObjHash, /* hash function */ + NULL, /* key dup */ + NULL, /* val dup */ + dictEncObjKeyCompare, /* key compare */ + dictRedisObjectDestructor, /* key destructor */ + dictRedisObjectDestructor /* val destructor */ +}; + /* Keylist hash table type has unencoded redis objects as keys and * lists as values. It's used for blocking operations (BLPOP) and to * map swapped keys to a list of clients waiting for this keys to be loaded. */ @@ -1441,6 +1495,8 @@ static void initServerConfig() { server.vm_max_memory = 1024LL*1024*1024*1; /* 1 GB of RAM */ server.vm_max_threads = 4; server.vm_blocked_clients = 0; + server.hash_max_zipmap_entries = REDIS_HASH_MAX_ZIPMAP_ENTRIES; + server.hash_max_zipmap_value = REDIS_HASH_MAX_ZIPMAP_VALUE; resetServerSaveParams(); @@ -1488,7 +1544,7 @@ static void initServer() { exit(1); } for (j = 0; j < server.dbnum; j++) { - server.db[j].dict = dictCreate(&hashDictType,NULL); + server.db[j].dict = dictCreate(&dbDictType,NULL); server.db[j].expires = dictCreate(&keyptrDictType,NULL); server.db[j].blockingkeys = dictCreate(&keylistDictType,NULL); if (server.vm_enabled) @@ -1701,6 +1757,12 @@ static void loadServerConfig(char *filename) { server.vm_pages = strtoll(argv[1], NULL, 10); } else if (!strcasecmp(argv[0],"vm-max-threads") && argc == 2) { server.vm_max_threads = strtoll(argv[1], NULL, 10); + } else if (!strcasecmp(argv[0],"hash-max-zipmap-entries") && argc == 2){ + server.hash_max_zipmap_entries = strtol(argv[1], NULL, 10); + } else if (!strcasecmp(argv[0],"hash-max-zipmap-value") && argc == 2){ + server.hash_max_zipmap_value = strtol(argv[1], NULL, 10); + } else if (!strcasecmp(argv[0],"vm-max-threads") && argc == 2) { + server.vm_max_threads = strtoll(argv[1], NULL, 10); } else { err = "Bad directive or wrong number of arguments"; goto loaderr; } @@ -2422,10 +2484,32 @@ static void addReplyLong(redisClient *c, long l) { char buf[128]; size_t len; + if (l == 0) { + addReply(c,shared.czero); + return; + } else if (l == 1) { + addReply(c,shared.cone); + return; + } len = snprintf(buf,sizeof(buf),":%ld\r\n",l); addReplySds(c,sdsnewlen(buf,len)); } +static void addReplyUlong(redisClient *c, unsigned long ul) { + char buf[128]; + size_t len; + + if (ul == 0) { + addReply(c,shared.czero); + return; + } else if (ul == 1) { + addReply(c,shared.cone); + return; + } + len = snprintf(buf,sizeof(buf),":%lu\r\n",ul); + addReplySds(c,sdsnewlen(buf,len)); +} + static void addReplyBulkLen(redisClient *c, robj *obj) { size_t len; @@ -2447,6 +2531,12 @@ static void addReplyBulkLen(redisClient *c, robj *obj) { addReplySds(c,sdscatprintf(sdsempty(),"$%lu\r\n",(unsigned long)len)); } +static void addReplyBulk(redisClient *c, robj *obj) { + addReplyBulkLen(c,obj); + addReply(c,obj); + addReply(c,shared.crlf); +} + static void acceptHandler(aeEventLoop *el, int fd, void *privdata, int mask) { int cport, cfd; char cip[128]; @@ -2538,6 +2628,16 @@ static robj *createSetObject(void) { return createObject(REDIS_SET,d); } +static robj *createHashObject(void) { + /* All the Hashes start as zipmaps. Will be automatically converted + * into hash tables if there are enough elements or big elements + * inside. */ + unsigned char *zm = zipmapNew(); + robj *o = createObject(REDIS_HASH,zm); + o->encoding = REDIS_ENCODING_ZIPMAP; + return o; +} + static robj *createZsetObject(void) { zset *zs = zmalloc(sizeof(*zs)); @@ -2569,7 +2669,17 @@ static void freeZsetObject(robj *o) { } static void freeHashObject(robj *o) { - dictRelease((dict*) o->ptr); + switch (o->encoding) { + case REDIS_ENCODING_HT: + dictRelease((dict*) o->ptr); + break; + case REDIS_ENCODING_ZIPMAP: + zfree(o->ptr); + break; + default: + redisAssert(0); + break; + } } static void incrRefCount(robj *o) { @@ -2610,7 +2720,7 @@ static void decrRefCount(void *obj) { case REDIS_SET: freeSetObject(o); break; case REDIS_ZSET: freeZsetObject(o); break; case REDIS_HASH: freeHashObject(o); break; - default: redisAssert(0 != 0); break; + default: redisAssert(0); break; } if (server.vm_enabled) pthread_mutex_lock(&server.obj_freelist_mutex); if (listLength(server.objfreelist) > REDIS_OBJFREELIST_MAX || @@ -2665,6 +2775,26 @@ static robj *lookupKeyWrite(redisDb *db, robj *key) { return lookupKey(db,key); } +static robj *lookupKeyReadOrReply(redisClient *c, robj *key, robj *reply) { + robj *o = lookupKeyRead(c->db, key); + if (!o) addReply(c,reply); + return o; +} + +static robj *lookupKeyWriteOrReply(redisClient *c, robj *key, robj *reply) { + robj *o = lookupKeyWrite(c->db, key); + if (!o) addReply(c,reply); + return o; +} + +static int checkType(redisClient *c, robj *o, int type) { + if (o->type != type) { + addReply(c,shared.wrongtypeerr); + return 1; + } + return 0; +} + static int deleteKey(redisDb *db, robj *key) { int retval; @@ -2874,7 +3004,7 @@ static int rdbSaveLen(FILE *fp, uint32_t len) { /* String objects in the form "2391" "-100" without any space and with a * range of values that can fit in an 8, 16 or 32 bit signed value can be * encoded as integers to save space */ -static int rdbTryIntegerEncoding(sds s, unsigned char *enc) { +static int rdbTryIntegerEncoding(char *s, size_t len, unsigned char *enc) { long long value; char *endptr, buf[32]; @@ -2885,7 +3015,7 @@ static int rdbTryIntegerEncoding(sds s, unsigned char *enc) { /* If the number converted back into a string is not identical * then it's not possible to encode the string as integer */ - if (strlen(buf) != sdslen(s) || memcmp(buf,s,sdslen(s))) return 0; + if (strlen(buf) != len || memcmp(buf,s,len)) return 0; /* Finally check if it fits in our ranges */ if (value >= -(1<<7) && value <= (1<<7)-1) { @@ -2909,16 +3039,16 @@ static int rdbTryIntegerEncoding(sds s, unsigned char *enc) { } } -static int rdbSaveLzfStringObject(FILE *fp, robj *obj) { - unsigned int comprlen, outlen; +static int rdbSaveLzfStringObject(FILE *fp, unsigned char *s, size_t len) { + size_t comprlen, outlen; unsigned char byte; void *out; /* We require at least four bytes compression for this to be worth it */ - outlen = sdslen(obj->ptr)-4; - if (outlen <= 0) return 0; + if (len <= 4) return 0; + outlen = len-4; if ((out = zmalloc(outlen+1)) == NULL) return 0; - comprlen = lzf_compress(obj->ptr, sdslen(obj->ptr), out, outlen); + comprlen = lzf_compress(s, len, out, outlen); if (comprlen == 0) { zfree(out); return 0; @@ -2927,7 +3057,7 @@ static int rdbSaveLzfStringObject(FILE *fp, robj *obj) { byte = (REDIS_RDB_ENCVAL<<6)|REDIS_RDB_ENC_LZF; if (fwrite(&byte,1,1,fp) == 0) goto writeerr; if (rdbSaveLen(fp,comprlen) == -1) goto writeerr; - if (rdbSaveLen(fp,sdslen(obj->ptr)) == -1) goto writeerr; + if (rdbSaveLen(fp,len) == -1) goto writeerr; if (fwrite(out,comprlen,1,fp) == 0) goto writeerr; zfree(out); return comprlen; @@ -2939,16 +3069,13 @@ writeerr: /* Save a string objet as [len][data] on disk. If the object is a string * representation of an integer value we try to safe it in a special form */ -static int rdbSaveStringObjectRaw(FILE *fp, robj *obj) { - size_t len; +static int rdbSaveRawString(FILE *fp, unsigned char *s, size_t len) { int enclen; - len = sdslen(obj->ptr); - /* Try integer encoding */ if (len <= 11) { unsigned char buf[5]; - if ((enclen = rdbTryIntegerEncoding(obj->ptr,buf)) > 0) { + if ((enclen = rdbTryIntegerEncoding((char*)s,len,buf)) > 0) { if (fwrite(buf,enclen,1,fp) == 0) return -1; return 0; } @@ -2959,7 +3086,7 @@ static int rdbSaveStringObjectRaw(FILE *fp, robj *obj) { if (server.rdbcompression && len > 20) { int retval; - retval = rdbSaveLzfStringObject(fp,obj); + retval = rdbSaveLzfStringObject(fp,s,len); if (retval == -1) return -1; if (retval > 0) return 0; /* retval == 0 means data can't be compressed, save the old way */ @@ -2967,7 +3094,7 @@ static int rdbSaveStringObjectRaw(FILE *fp, robj *obj) { /* Store verbatim */ if (rdbSaveLen(fp,len) == -1) return -1; - if (len && fwrite(obj->ptr,len,1,fp) == 0) return -1; + if (len && fwrite(s,len,1,fp) == 0) return -1; return 0; } @@ -2982,10 +3109,10 @@ static int rdbSaveStringObject(FILE *fp, robj *obj) { * this in order to avoid bugs) */ if (obj->encoding != REDIS_ENCODING_RAW) { obj = getDecodedObject(obj); - retval = rdbSaveStringObjectRaw(fp,obj); + retval = rdbSaveRawString(fp,obj->ptr,sdslen(obj->ptr)); decrRefCount(obj); } else { - retval = rdbSaveStringObjectRaw(fp,obj); + retval = rdbSaveRawString(fp,obj->ptr,sdslen(obj->ptr)); } return retval; } @@ -3063,8 +3190,35 @@ static int rdbSaveObject(FILE *fp, robj *o) { if (rdbSaveDoubleValue(fp,*score) == -1) return -1; } dictReleaseIterator(di); + } else if (o->type == REDIS_HASH) { + /* Save a hash value */ + if (o->encoding == REDIS_ENCODING_ZIPMAP) { + unsigned char *p = zipmapRewind(o->ptr); + unsigned int count = zipmapLen(o->ptr); + unsigned char *key, *val; + unsigned int klen, vlen; + + if (rdbSaveLen(fp,count) == -1) return -1; + while((p = zipmapNext(p,&key,&klen,&val,&vlen)) != NULL) { + if (rdbSaveRawString(fp,key,klen) == -1) return -1; + if (rdbSaveRawString(fp,val,vlen) == -1) return -1; + } + } else { + dictIterator *di = dictGetIterator(o->ptr); + dictEntry *de; + + if (rdbSaveLen(fp,dictSize((dict*)o->ptr)) == -1) return -1; + while((de = dictNext(di)) != NULL) { + robj *key = dictGetEntryKey(de); + robj *val = dictGetEntryVal(de); + + if (rdbSaveStringObject(fp,key) == -1) return -1; + if (rdbSaveStringObject(fp,val) == -1) return -1; + } + dictReleaseIterator(di); + } } else { - redisAssert(0 != 0); + redisAssert(0); } return 0; } @@ -3284,7 +3438,7 @@ static robj *rdbLoadIntegerObject(FILE *fp, int enctype) { val = (int32_t)v; } else { val = 0; /* anti-warning */ - redisAssert(0!=0); + redisAssert(0); } return createObject(REDIS_STRING,sdscatprintf(sdsempty(),"%lld",val)); } @@ -3323,7 +3477,7 @@ static robj *rdbLoadStringObject(FILE*fp) { case REDIS_RDB_ENC_LZF: return tryObjectSharing(rdbLoadLzfStringObject(fp)); default: - redisAssert(0!=0); + redisAssert(0); } } @@ -3359,6 +3513,7 @@ static int rdbLoadDoubleValue(FILE *fp, double *val) { static robj *rdbLoadObject(int type, FILE *fp) { robj *o; + redisLog(REDIS_DEBUG,"LOADING OBJECT %d (at %d)\n",type,ftell(fp)); if (type == REDIS_STRING) { /* Read string value */ if ((o = rdbLoadStringObject(fp)) == NULL) return NULL; @@ -3387,7 +3542,7 @@ static robj *rdbLoadObject(int type, FILE *fp) { } } else if (type == REDIS_ZSET) { /* Read list/set value */ - uint32_t zsetlen; + size_t zsetlen; zset *zs; if ((zsetlen = rdbLoadLen(fp,NULL)) == REDIS_RDB_LENERR) return NULL; @@ -3405,8 +3560,46 @@ static robj *rdbLoadObject(int type, FILE *fp) { zslInsert(zs->zsl,*score,ele); incrRefCount(ele); /* added to skiplist */ } + } else if (type == REDIS_HASH) { + size_t hashlen; + + if ((hashlen = rdbLoadLen(fp,NULL)) == REDIS_RDB_LENERR) return NULL; + o = createHashObject(); + /* Too many entries? Use an hash table. */ + if (hashlen > server.hash_max_zipmap_entries) + convertToRealHash(o); + /* Load every key/value, then set it into the zipmap or hash + * table, as needed. */ + while(hashlen--) { + robj *key, *val; + + if ((key = rdbLoadStringObject(fp)) == NULL) return NULL; + if ((val = rdbLoadStringObject(fp)) == NULL) return NULL; + /* If we are using a zipmap and there are too big values + * the object is converted to real hash table encoding. */ + if (o->encoding != REDIS_ENCODING_HT && + (sdslen(key->ptr) > server.hash_max_zipmap_value || + sdslen(val->ptr) > server.hash_max_zipmap_value)) + { + convertToRealHash(o); + } + + if (o->encoding == REDIS_ENCODING_ZIPMAP) { + unsigned char *zm = o->ptr; + + zm = zipmapSet(zm,key->ptr,sdslen(key->ptr), + val->ptr,sdslen(val->ptr),NULL); + o->ptr = zm; + decrRefCount(key); + decrRefCount(val); + } else { + tryObjectEncoding(key); + tryObjectEncoding(val); + dictAdd((dict*)o->ptr,key,val); + } + } } else { - redisAssert(0 != 0); + redisAssert(0); } return o; } @@ -3513,9 +3706,7 @@ static void pingCommand(redisClient *c) { } static void echoCommand(redisClient *c) { - addReplyBulkLen(c,c->argv[1]); - addReply(c,c->argv[1]); - addReply(c,shared.crlf); + addReplyBulk(c,c->argv[1]); } /*=================================== Strings =============================== */ @@ -3531,7 +3722,7 @@ static void setGenericCommand(redisClient *c, int nx) { * 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 (deleteIfSwapped(c->db,c->argv[1])) + 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]); @@ -3557,21 +3748,17 @@ static void setnxCommand(redisClient *c) { } static int getGenericCommand(redisClient *c) { - robj *o = lookupKeyRead(c->db,c->argv[1]); - - if (o == NULL) { - addReply(c,shared.nullbulk); + robj *o; + + if ((o = lookupKeyReadOrReply(c,c->argv[1],shared.nullbulk)) == NULL) return REDIS_OK; + + if (o->type != REDIS_STRING) { + addReply(c,shared.wrongtypeerr); + return REDIS_ERR; } else { - if (o->type != REDIS_STRING) { - addReply(c,shared.wrongtypeerr); - return REDIS_ERR; - } else { - addReplyBulkLen(c,o); - addReply(c,o); - addReply(c,shared.crlf); - return REDIS_OK; - } + addReplyBulk(c,o); + return REDIS_OK; } } @@ -3603,9 +3790,7 @@ static void mgetCommand(redisClient *c) { if (o->type != REDIS_STRING) { addReply(c,shared.nullbulk); } else { - addReplyBulkLen(c,o); - addReply(c,o); - addReply(c,shared.crlf); + addReplyBulk(c,o); } } } @@ -3761,6 +3946,43 @@ static void appendCommand(redisClient *c) { addReplySds(c,sdscatprintf(sdsempty(),":%lu\r\n",(unsigned long)totlen)); } +static void substrCommand(redisClient *c) { + robj *o; + long start = atoi(c->argv[2]->ptr); + long end = atoi(c->argv[3]->ptr); + size_t rangelen, strlen; + sds range; + + if ((o = lookupKeyReadOrReply(c,c->argv[1],shared.nullbulk)) == NULL || + checkType(c,o,REDIS_STRING)) return; + + o = getDecodedObject(o); + strlen = sdslen(o->ptr); + + /* convert negative indexes */ + if (start < 0) start = strlen+start; + if (end < 0) end = strlen+end; + if (start < 0) start = 0; + if (end < 0) end = 0; + + /* indexes sanity checks */ + if (start > end || (size_t)start >= strlen) { + /* Out of range start or start > end result in null reply */ + addReply(c,shared.nullbulk); + decrRefCount(o); + return; + } + if ((size_t)end >= strlen) end = strlen-1; + rangelen = (end-start)+1; + + /* Return the result */ + addReplySds(c,sdscatprintf(sdsempty(),"$%zu\r\n",rangelen)); + range = sdsnewlen((char*)o->ptr+start,rangelen); + addReplySds(c,range); + addReply(c,shared.crlf); + decrRefCount(o); +} + /* ========================= Type agnostic commands ========================= */ static void delCommand(redisClient *c) { @@ -3772,17 +3994,7 @@ static void delCommand(redisClient *c) { deleted++; } } - switch(deleted) { - case 0: - addReply(c,shared.czero); - break; - case 1: - addReply(c,shared.cone); - break; - default: - addReplySds(c,sdscatprintf(sdsempty(),":%d\r\n",deleted)); - break; - } + addReplyLong(c,deleted); } static void existsCommand(redisClient *c) { @@ -3834,9 +4046,7 @@ static void keysCommand(redisClient *c) { if ((pattern[0] == '*' && pattern[1] == '\0') || stringmatchlen(pattern,plen,key,sdslen(key),0)) { if (expireIfNeeded(c->db,keyobj) == 0) { - addReplyBulkLen(c,keyobj); - addReply(c,keyobj); - addReply(c,shared.crlf); + addReplyBulk(c,keyobj); numkeys++; } } @@ -3868,7 +4078,8 @@ static void typeCommand(redisClient *c) { case REDIS_LIST: type = "+list"; break; case REDIS_SET: type = "+set"; break; case REDIS_ZSET: type = "+zset"; break; - default: type = "unknown"; break; + case REDIS_HASH: type = "+hash"; break; + default: type = "+unknown"; break; } } addReplySds(c,sdsnew(type)); @@ -3925,12 +4136,14 @@ static void shutdownCommand(redisClient *c) { 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... */ + /* 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")); + addReplySds(c, + sdsnew("-ERR can't quit, problems saving the DB\r\n")); } } } @@ -3944,11 +4157,9 @@ static void renameGenericCommand(redisClient *c, int nx) { return; } - o = lookupKeyWrite(c->db,c->argv[1]); - if (o == NULL) { - addReply(c,shared.nokeyerr); + if ((o = lookupKeyWriteOrReply(c,c->argv[1],shared.nokeyerr)) == NULL) return; - } + incrRefCount(o); deleteIfVolatile(c->db,c->argv[2]); if (dictAdd(c->db->dict,c->argv[2],o) == DICT_ERR) { @@ -4026,7 +4237,7 @@ 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.ok); + addReply(c,shared.cone); return; } lobj = createListObject(); @@ -4045,7 +4256,7 @@ static void pushGenericCommand(redisClient *c, int where) { return; } if (handleClientsWaitingListPush(c,c->argv[1],c->argv[2])) { - addReply(c,shared.ok); + addReply(c,shared.cone); return; } list = lobj->ptr; @@ -4057,7 +4268,7 @@ 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) { @@ -4071,107 +4282,78 @@ static void rpushCommand(redisClient *c) { static void llenCommand(redisClient *c) { robj *o; list *l; + + if ((o = lookupKeyReadOrReply(c,c->argv[1],shared.czero)) == NULL || + checkType(c,o,REDIS_LIST)) return; - o = lookupKeyRead(c->db,c->argv[1]); - if (o == NULL) { - addReply(c,shared.czero); - return; - } else { - if (o->type != REDIS_LIST) { - addReply(c,shared.wrongtypeerr); - } else { - l = o->ptr; - addReplySds(c,sdscatprintf(sdsempty(),":%d\r\n",listLength(l))); - } - } + l = o->ptr; + addReplyUlong(c,listLength(l)); } static void lindexCommand(redisClient *c) { robj *o; int index = atoi(c->argv[2]->ptr); - - o = lookupKeyRead(c->db,c->argv[1]); - if (o == NULL) { + list *list; + listNode *ln; + + if ((o = lookupKeyReadOrReply(c,c->argv[1],shared.nullbulk)) == NULL || + checkType(c,o,REDIS_LIST)) return; + list = o->ptr; + + ln = listIndex(list, index); + if (ln == NULL) { addReply(c,shared.nullbulk); } else { - if (o->type != REDIS_LIST) { - addReply(c,shared.wrongtypeerr); - } else { - list *list = o->ptr; - listNode *ln; - - ln = listIndex(list, index); - if (ln == NULL) { - addReply(c,shared.nullbulk); - } else { - robj *ele = listNodeValue(ln); - addReplyBulkLen(c,ele); - addReply(c,ele); - addReply(c,shared.crlf); - } - } + robj *ele = listNodeValue(ln); + addReplyBulk(c,ele); } } static void lsetCommand(redisClient *c) { robj *o; int index = atoi(c->argv[2]->ptr); - - o = lookupKeyWrite(c->db,c->argv[1]); - if (o == NULL) { - addReply(c,shared.nokeyerr); + list *list; + listNode *ln; + + if ((o = lookupKeyWriteOrReply(c,c->argv[1],shared.nokeyerr)) == NULL || + checkType(c,o,REDIS_LIST)) return; + list = o->ptr; + + ln = listIndex(list, index); + if (ln == NULL) { + addReply(c,shared.outofrangeerr); } else { - if (o->type != REDIS_LIST) { - addReply(c,shared.wrongtypeerr); - } else { - list *list = o->ptr; - listNode *ln; - - ln = listIndex(list, index); - if (ln == NULL) { - addReply(c,shared.outofrangeerr); - } else { - robj *ele = listNodeValue(ln); + robj *ele = listNodeValue(ln); - decrRefCount(ele); - listNodeValue(ln) = c->argv[3]; - incrRefCount(c->argv[3]); - addReply(c,shared.ok); - server.dirty++; - } - } + decrRefCount(ele); + listNodeValue(ln) = c->argv[3]; + incrRefCount(c->argv[3]); + addReply(c,shared.ok); + server.dirty++; } } static void popGenericCommand(redisClient *c, int where) { robj *o; + list *list; + listNode *ln; - o = lookupKeyWrite(c->db,c->argv[1]); - if (o == NULL) { - addReply(c,shared.nullbulk); - } else { - if (o->type != REDIS_LIST) { - addReply(c,shared.wrongtypeerr); - } else { - list *list = o->ptr; - listNode *ln; + if ((o = lookupKeyWriteOrReply(c,c->argv[1],shared.nullbulk)) == NULL || + checkType(c,o,REDIS_LIST)) return; + list = o->ptr; - if (where == REDIS_HEAD) - ln = listFirst(list); - else - ln = listLast(list); + if (where == REDIS_HEAD) + ln = listFirst(list); + else + ln = listLast(list); - if (ln == NULL) { - addReply(c,shared.nullbulk); - } else { - robj *ele = listNodeValue(ln); - addReplyBulkLen(c,ele); - addReply(c,ele); - addReply(c,shared.crlf); - listDelNode(list,ln); - server.dirty++; - } - } + if (ln == NULL) { + addReply(c,shared.nullbulk); + } else { + robj *ele = listNodeValue(ln); + addReplyBulk(c,ele); + listDelNode(list,ln); + server.dirty++; } } @@ -4187,46 +4369,39 @@ static void lrangeCommand(redisClient *c) { robj *o; int start = atoi(c->argv[2]->ptr); int end = atoi(c->argv[3]->ptr); + int llen; + int rangelen, j; + list *list; + listNode *ln; + robj *ele; + + if ((o = lookupKeyReadOrReply(c,c->argv[1],shared.nullmultibulk)) == NULL || + checkType(c,o,REDIS_LIST)) return; + list = o->ptr; + llen = listLength(list); + + /* convert negative indexes */ + if (start < 0) start = llen+start; + if (end < 0) end = llen+end; + if (start < 0) start = 0; + if (end < 0) end = 0; + + /* indexes sanity checks */ + if (start > end || start >= llen) { + /* Out of range start or start > end result in empty list */ + addReply(c,shared.emptymultibulk); + return; + } + if (end >= llen) end = llen-1; + rangelen = (end-start)+1; - o = lookupKeyRead(c->db,c->argv[1]); - if (o == NULL) { - addReply(c,shared.nullmultibulk); - } else { - if (o->type != REDIS_LIST) { - addReply(c,shared.wrongtypeerr); - } else { - list *list = o->ptr; - listNode *ln; - int llen = listLength(list); - int rangelen, j; - robj *ele; - - /* convert negative indexes */ - if (start < 0) start = llen+start; - if (end < 0) end = llen+end; - if (start < 0) start = 0; - if (end < 0) end = 0; - - /* indexes sanity checks */ - if (start > end || start >= llen) { - /* Out of range start or start > end result in empty list */ - addReply(c,shared.emptymultibulk); - return; - } - if (end >= llen) end = llen-1; - rangelen = (end-start)+1; - - /* Return the result in form of a multi-bulk reply */ - ln = listIndex(list, start); - addReplySds(c,sdscatprintf(sdsempty(),"*%d\r\n",rangelen)); - for (j = 0; j < rangelen; j++) { - ele = listNodeValue(ln); - addReplyBulkLen(c,ele); - addReply(c,ele); - addReply(c,shared.crlf); - ln = ln->next; - } - } + /* Return the result in form of a multi-bulk reply */ + ln = listIndex(list, start); + addReplySds(c,sdscatprintf(sdsempty(),"*%d\r\n",rangelen)); + for (j = 0; j < rangelen; j++) { + ele = listNodeValue(ln); + addReplyBulk(c,ele); + ln = ln->next; } } @@ -4234,87 +4409,76 @@ static void ltrimCommand(redisClient *c) { robj *o; int start = atoi(c->argv[2]->ptr); int end = atoi(c->argv[3]->ptr); - - o = lookupKeyWrite(c->db,c->argv[1]); - if (o == NULL) { - addReply(c,shared.ok); + int llen; + int j, ltrim, rtrim; + list *list; + listNode *ln; + + if ((o = lookupKeyWriteOrReply(c,c->argv[1],shared.ok)) == NULL || + checkType(c,o,REDIS_LIST)) return; + list = o->ptr; + llen = listLength(list); + + /* convert negative indexes */ + if (start < 0) start = llen+start; + if (end < 0) end = llen+end; + if (start < 0) start = 0; + if (end < 0) end = 0; + + /* indexes sanity checks */ + if (start > end || start >= llen) { + /* Out of range start or start > end result in empty list */ + ltrim = llen; + rtrim = 0; } else { - if (o->type != REDIS_LIST) { - addReply(c,shared.wrongtypeerr); - } else { - list *list = o->ptr; - listNode *ln; - int llen = listLength(list); - int j, ltrim, rtrim; - - /* convert negative indexes */ - if (start < 0) start = llen+start; - if (end < 0) end = llen+end; - if (start < 0) start = 0; - if (end < 0) end = 0; - - /* indexes sanity checks */ - if (start > end || start >= llen) { - /* Out of range start or start > end result in empty list */ - ltrim = llen; - rtrim = 0; - } else { - if (end >= llen) end = llen-1; - ltrim = start; - rtrim = llen-end-1; - } + if (end >= llen) end = llen-1; + ltrim = start; + rtrim = llen-end-1; + } - /* Remove list elements to perform the trim */ - for (j = 0; j < ltrim; j++) { - ln = listFirst(list); - listDelNode(list,ln); - } - for (j = 0; j < rtrim; j++) { - ln = listLast(list); - listDelNode(list,ln); - } - server.dirty++; - addReply(c,shared.ok); - } + /* Remove list elements to perform the trim */ + for (j = 0; j < ltrim; j++) { + ln = listFirst(list); + listDelNode(list,ln); } + for (j = 0; j < rtrim; j++) { + ln = listLast(list); + listDelNode(list,ln); + } + server.dirty++; + addReply(c,shared.ok); } static void lremCommand(redisClient *c) { robj *o; - - o = lookupKeyWrite(c->db,c->argv[1]); - if (o == NULL) { - addReply(c,shared.czero); - } else { - if (o->type != REDIS_LIST) { - addReply(c,shared.wrongtypeerr); - } else { - list *list = o->ptr; - listNode *ln, *next; - int toremove = atoi(c->argv[2]->ptr); - int removed = 0; - int fromtail = 0; - - if (toremove < 0) { - toremove = -toremove; - fromtail = 1; - } - ln = fromtail ? list->tail : list->head; - while (ln) { - robj *ele = listNodeValue(ln); - - next = fromtail ? ln->prev : ln->next; - if (compareStringObjects(ele,c->argv[3]) == 0) { - listDelNode(list,ln); - server.dirty++; - removed++; - if (toremove && removed == toremove) break; - } - ln = next; - } - addReplySds(c,sdscatprintf(sdsempty(),":%d\r\n",removed)); + list *list; + listNode *ln, *next; + int toremove = atoi(c->argv[2]->ptr); + int removed = 0; + int fromtail = 0; + + if ((o = lookupKeyWriteOrReply(c,c->argv[1],shared.czero)) == NULL || + checkType(c,o,REDIS_LIST)) return; + list = o->ptr; + + if (toremove < 0) { + toremove = -toremove; + fromtail = 1; + } + ln = fromtail ? list->tail : list->head; + while (ln) { + robj *ele = listNodeValue(ln); + + next = fromtail ? ln->prev : ln->next; + if (compareStringObjects(ele,c->argv[3]) == 0) { + listDelNode(list,ln); + server.dirty++; + removed++; + if (toremove && removed == toremove) break; } + ln = next; } + addReplySds(c,sdscatprintf(sdsempty(),":%d\r\n",removed)); } /* This is the semantic of this command: @@ -4334,57 +4498,49 @@ static void lremCommand(redisClient *c) { */ static void rpoplpushcommand(redisClient *c) { robj *sobj; + list *srclist; + listNode *ln; - sobj = lookupKeyWrite(c->db,c->argv[1]); - if (sobj == NULL) { + if ((sobj = lookupKeyWriteOrReply(c,c->argv[1],shared.nullbulk)) == NULL || + checkType(c,sobj,REDIS_LIST)) return; + srclist = sobj->ptr; + ln = listLast(srclist); + + if (ln == NULL) { addReply(c,shared.nullbulk); } else { - if (sobj->type != REDIS_LIST) { - addReply(c,shared.wrongtypeerr); - } else { - list *srclist = sobj->ptr; - listNode *ln = listLast(srclist); - - if (ln == NULL) { - addReply(c,shared.nullbulk); - } else { - robj *dobj = lookupKeyWrite(c->db,c->argv[2]); - robj *ele = listNodeValue(ln); - list *dstlist; - - if (dobj && dobj->type != REDIS_LIST) { - addReply(c,shared.wrongtypeerr); - return; - } + robj *dobj = lookupKeyWrite(c->db,c->argv[2]); + robj *ele = listNodeValue(ln); + list *dstlist; - /* 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); - addReply(c,ele); - addReply(c,shared.crlf); + if (dobj && dobj->type != REDIS_LIST) { + addReply(c,shared.wrongtypeerr); + return; + } - /* Finally remove the element from the source list */ - listDelNode(srclist,ln); - server.dirty++; + /* Add the element to the target list (unless it's directly + * passed to some BLPOP-ing client */ + if (!handleClientsWaitingListPush(c,c->argv[2],ele)) { + if (dobj == NULL) { + /* Create the list if the key does not exist */ + dobj = createListObject(); + dictAdd(c->db->dict,c->argv[2],dobj); + incrRefCount(c->argv[2]); } + dstlist = dobj->ptr; + listAddNodeHead(dstlist,ele); + incrRefCount(ele); } + + /* Send the element to the client as reply as well */ + addReplyBulk(c,ele); + + /* Finally remove the element from the source list */ + listDelNode(srclist,ln); + server.dirty++; } } - /* ==================================== Sets ================================ */ static void saddCommand(redisClient *c) { @@ -4413,21 +4569,15 @@ static void saddCommand(redisClient *c) { static void sremCommand(redisClient *c) { robj *set; - set = lookupKeyWrite(c->db,c->argv[1]); - if (set == NULL) { - addReply(c,shared.czero); - } else { - if (set->type != REDIS_SET) { - addReply(c,shared.wrongtypeerr); - return; - } - if (dictDelete(set->ptr,c->argv[2]) == DICT_OK) { - server.dirty++; - if (htNeedsResize(set->ptr)) dictResize(set->ptr); - addReply(c,shared.cone); - } else { - addReply(c,shared.czero); - } + if ((set = lookupKeyWriteOrReply(c,c->argv[1],shared.czero)) == NULL || + checkType(c,set,REDIS_SET)) return; + + if (dictDelete(set->ptr,c->argv[2]) == DICT_OK) { + server.dirty++; + if (htNeedsResize(set->ptr)) dictResize(set->ptr); + addReply(c,shared.cone); + } else { + addReply(c,shared.czero); } } @@ -4469,65 +4619,43 @@ static void smoveCommand(redisClient *c) { static void sismemberCommand(redisClient *c) { robj *set; - set = lookupKeyRead(c->db,c->argv[1]); - if (set == NULL) { + if ((set = lookupKeyReadOrReply(c,c->argv[1],shared.czero)) == NULL || + checkType(c,set,REDIS_SET)) return; + + if (dictFind(set->ptr,c->argv[2])) + addReply(c,shared.cone); + else addReply(c,shared.czero); - } else { - if (set->type != REDIS_SET) { - addReply(c,shared.wrongtypeerr); - return; - } - if (dictFind(set->ptr,c->argv[2])) - addReply(c,shared.cone); - else - addReply(c,shared.czero); - } } static void scardCommand(redisClient *c) { robj *o; dict *s; + + if ((o = lookupKeyReadOrReply(c,c->argv[1],shared.czero)) == NULL || + checkType(c,o,REDIS_SET)) return; - o = lookupKeyRead(c->db,c->argv[1]); - if (o == NULL) { - addReply(c,shared.czero); - return; - } else { - if (o->type != REDIS_SET) { - addReply(c,shared.wrongtypeerr); - } else { - s = o->ptr; - addReplySds(c,sdscatprintf(sdsempty(),":%lu\r\n", - dictSize(s))); - } - } + s = o->ptr; + addReplyUlong(c,dictSize(s)); } static void spopCommand(redisClient *c) { robj *set; dictEntry *de; - set = lookupKeyWrite(c->db,c->argv[1]); - if (set == NULL) { + if ((set = lookupKeyWriteOrReply(c,c->argv[1],shared.nullbulk)) == NULL || + checkType(c,set,REDIS_SET)) return; + + de = dictGetRandomKey(set->ptr); + if (de == NULL) { addReply(c,shared.nullbulk); } else { - if (set->type != REDIS_SET) { - addReply(c,shared.wrongtypeerr); - return; - } - de = dictGetRandomKey(set->ptr); - if (de == NULL) { - addReply(c,shared.nullbulk); - } else { - robj *ele = dictGetEntryKey(de); + robj *ele = dictGetEntryKey(de); - addReplyBulkLen(c,ele); - addReply(c,ele); - addReply(c,shared.crlf); - dictDelete(set->ptr,ele); - if (htNeedsResize(set->ptr)) dictResize(set->ptr); - server.dirty++; - } + addReplyBulk(c,ele); + dictDelete(set->ptr,ele); + if (htNeedsResize(set->ptr)) dictResize(set->ptr); + server.dirty++; } } @@ -4535,24 +4663,16 @@ static void srandmemberCommand(redisClient *c) { robj *set; dictEntry *de; - set = lookupKeyRead(c->db,c->argv[1]); - if (set == NULL) { + if ((set = lookupKeyReadOrReply(c,c->argv[1],shared.nullbulk)) == NULL || + checkType(c,set,REDIS_SET)) return; + + de = dictGetRandomKey(set->ptr); + if (de == NULL) { addReply(c,shared.nullbulk); } else { - if (set->type != REDIS_SET) { - addReply(c,shared.wrongtypeerr); - return; - } - de = dictGetRandomKey(set->ptr); - if (de == NULL) { - addReply(c,shared.nullbulk); - } else { - robj *ele = dictGetEntryKey(de); + robj *ele = dictGetEntryKey(de); - addReplyBulkLen(c,ele); - addReply(c,ele); - addReply(c,shared.crlf); - } + addReplyBulk(c,ele); } } @@ -4626,9 +4746,7 @@ static void sinterGenericCommand(redisClient *c, robj **setskeys, unsigned long continue; /* at least one set does not contain the member */ ele = dictGetEntryKey(de); if (!dstkey) { - addReplyBulkLen(c,ele); - addReply(c,ele); - addReply(c,shared.crlf); + addReplyBulk(c,ele); cardinality++; } else { dictAdd(dstset->ptr,ele,NULL); @@ -4664,6 +4782,7 @@ static void sinterstoreCommand(redisClient *c) { #define REDIS_OP_UNION 0 #define REDIS_OP_DIFF 1 +#define REDIS_OP_INTER 2 static void sunionDiffGenericCommand(redisClient *c, robj **setskeys, int setsnum, robj *dstkey, int op) { dict **dv = zmalloc(sizeof(dict*)*setsnum); @@ -4732,9 +4851,7 @@ static void sunionDiffGenericCommand(redisClient *c, robj **setskeys, int setsnu robj *ele; ele = dictGetEntryKey(de); - addReplyBulkLen(c,ele); - addReply(c,ele); - addReply(c,shared.crlf); + addReplyBulk(c,ele); } dictReleaseIterator(di); } else { @@ -4795,6 +4912,8 @@ 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; @@ -4808,8 +4927,13 @@ 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; + + /* span has space for ZSKIPLIST_MAXLEVEL-1 elements */ + if (j < ZSKIPLIST_MAXLEVEL-1) + zsl->header->span[j] = 0; + } zsl->header->backward = NULL; zsl->tail = NULL; return zsl; @@ -4818,6 +4942,7 @@ static zskiplist *zslCreate(void) { static void zslFreeNode(zskiplistNode *node) { decrRefCount(node->obj); zfree(node->forward); + zfree(node->span); zfree(node); } @@ -4825,6 +4950,7 @@ 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]; @@ -4843,15 +4969,21 @@ 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 @@ -4860,15 +4992,30 @@ 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; @@ -4877,6 +5024,31 @@ static void zslInsert(zskiplist *zsl, double score, robj *obj) { zsl->length++; } +/* Internal function used by zslDelete, zslDeleteByScore and zslDeleteByRank */ +void zslDeleteNode(zskiplist *zsl, zskiplistNode *x, zskiplistNode **update) { + int i; + for (i = 0; i < zsl->level; i++) { + if (update[i]->forward[i] == x) { + if (i > 0) { + update[i]->span[i-1] += x->span[i-1] - 1; + } + update[i]->forward[i] = x->forward[i]; + } else { + /* invariant: i > 0, because update[0]->forward[0] + * is always equal to x */ + update[i]->span[i-1] -= 1; + } + } + if (x->forward[0]) { + x->forward[0]->backward = x->backward; + } else { + zsl->tail = x->backward; + } + while(zsl->level > 1 && zsl->header->forward[zsl->level-1] == NULL) + zsl->level--; + zsl->length--; +} + /* Delete an element with matching score/object from the skiplist. */ static int zslDelete(zskiplist *zsl, double score, robj *obj) { zskiplistNode *update[ZSKIPLIST_MAXLEVEL], *x; @@ -4895,20 +5067,8 @@ static int zslDelete(zskiplist *zsl, double score, robj *obj) { * is to find the element with both the right score and object. */ x = x->forward[0]; if (x && score == x->score && compareStringObjects(x->obj,obj) == 0) { - for (i = 0; i < zsl->level; i++) { - if (update[i]->forward[i] != x) break; - update[i]->forward[i] = x->forward[i]; - } - if (x->forward[0]) { - x->forward[0]->backward = (x->backward == zsl->header) ? - NULL : x->backward; - } else { - zsl->tail = x->backward; - } + zslDeleteNode(zsl, x, update); zslFreeNode(x); - while(zsl->level > 1 && zsl->header->forward[zsl->level-1] == NULL) - zsl->level--; - zsl->length--; return 1; } else { return 0; /* not found */ @@ -4920,7 +5080,7 @@ static int zslDelete(zskiplist *zsl, double score, robj *obj) { * Min and mx are inclusive, so a score >= min || score <= max is deleted. * Note that this function takes the reference to the hash table view of the * sorted set, in order to remove the elements from the hash table too. */ -static unsigned long zslDeleteRange(zskiplist *zsl, double min, double max, dict *dict) { +static unsigned long zslDeleteRangeByScore(zskiplist *zsl, double min, double max, dict *dict) { zskiplistNode *update[ZSKIPLIST_MAXLEVEL], *x; unsigned long removed = 0; int i; @@ -4935,28 +5095,44 @@ static unsigned long zslDeleteRange(zskiplist *zsl, double min, double max, dict * is to find the element with both the right score and object. */ x = x->forward[0]; while (x && x->score <= max) { - zskiplistNode *next; + zskiplistNode *next = x->forward[0]; + zslDeleteNode(zsl, x, update); + dictDelete(dict,x->obj); + zslFreeNode(x); + removed++; + x = next; + } + return removed; /* not found */ +} - for (i = 0; i < zsl->level; i++) { - if (update[i]->forward[i] != x) break; - update[i]->forward[i] = x->forward[i]; - } - if (x->forward[0]) { - x->forward[0]->backward = (x->backward == zsl->header) ? - NULL : x->backward; - } else { - zsl->tail = x->backward; +/* Delete all the elements with rank between start and end from the skiplist. + * Start and end are inclusive. Note that start and end need to be 1-based */ +static unsigned long zslDeleteRangeByRank(zskiplist *zsl, unsigned int start, unsigned int end, dict *dict) { + zskiplistNode *update[ZSKIPLIST_MAXLEVEL], *x; + unsigned long traversed = 0, removed = 0; + int i; + + x = zsl->header; + for (i = zsl->level-1; i >= 0; i--) { + while (x->forward[i] && (traversed + (i > 0 ? x->span[i-1] : 1)) < start) { + traversed += i > 0 ? x->span[i-1] : 1; + x = x->forward[i]; } - next = x->forward[0]; + update[i] = x; + } + + traversed++; + x = x->forward[0]; + while (x && traversed <= end) { + zskiplistNode *next = x->forward[0]; + zslDeleteNode(zsl, x, update); dictDelete(dict,x->obj); zslFreeNode(x); - while(zsl->level > 1 && zsl->header->forward[zsl->level-1] == NULL) - zsl->level--; - zsl->length--; removed++; + traversed++; x = next; } - return removed; /* not found */ + return removed; } /* Find the first node having a score equal or greater than the specified one. @@ -4975,6 +5151,53 @@ 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 */ /* This generic command implements both ZADD and ZINCRBY. @@ -5075,60 +5298,245 @@ static void zincrbyCommand(redisClient *c) { static void zremCommand(redisClient *c) { robj *zsetobj; zset *zs; + dictEntry *de; + double *oldscore; + int deleted; - zsetobj = lookupKeyWrite(c->db,c->argv[1]); - if (zsetobj == NULL) { - addReply(c,shared.czero); - } else { - dictEntry *de; - double *oldscore; - int deleted; + if ((zsetobj = lookupKeyWriteOrReply(c,c->argv[1],shared.czero)) == NULL || + checkType(c,zsetobj,REDIS_ZSET)) return; - if (zsetobj->type != REDIS_ZSET) { - addReply(c,shared.wrongtypeerr); - return; - } - zs = zsetobj->ptr; - de = dictFind(zs->dict,c->argv[2]); - if (de == NULL) { - addReply(c,shared.czero); - return; - } - /* Delete from the skiplist */ - oldscore = dictGetEntryVal(de); - deleted = zslDelete(zs->zsl,*oldscore,c->argv[2]); - redisAssert(deleted != 0); - - /* Delete from the hash table */ - dictDelete(zs->dict,c->argv[2]); - if (htNeedsResize(zs->dict)) dictResize(zs->dict); - server.dirty++; - addReply(c,shared.cone); + zs = zsetobj->ptr; + de = dictFind(zs->dict,c->argv[2]); + if (de == NULL) { + addReply(c,shared.czero); + return; } + /* Delete from the skiplist */ + oldscore = dictGetEntryVal(de); + deleted = zslDelete(zs->zsl,*oldscore,c->argv[2]); + redisAssert(deleted != 0); + + /* Delete from the hash table */ + dictDelete(zs->dict,c->argv[2]); + if (htNeedsResize(zs->dict)) dictResize(zs->dict); + server.dirty++; + addReply(c,shared.cone); } static void zremrangebyscoreCommand(redisClient *c) { double min = strtod(c->argv[2]->ptr,NULL); double max = strtod(c->argv[3]->ptr,NULL); + long deleted; robj *zsetobj; zset *zs; - zsetobj = lookupKeyWrite(c->db,c->argv[1]); - if (zsetobj == NULL) { + if ((zsetobj = lookupKeyWriteOrReply(c,c->argv[1],shared.czero)) == NULL || + checkType(c,zsetobj,REDIS_ZSET)) return; + + zs = zsetobj->ptr; + deleted = zslDeleteRangeByScore(zs->zsl,min,max,zs->dict); + if (htNeedsResize(zs->dict)) dictResize(zs->dict); + server.dirty += deleted; + addReplyLong(c,deleted); +} + +static void zremrangebyrankCommand(redisClient *c) { + int start = atoi(c->argv[2]->ptr); + int end = atoi(c->argv[3]->ptr); + int llen; + long deleted; + robj *zsetobj; + zset *zs; + + if ((zsetobj = lookupKeyWriteOrReply(c,c->argv[1],shared.czero)) == NULL || + checkType(c,zsetobj,REDIS_ZSET)) return; + zs = zsetobj->ptr; + llen = zs->zsl->length; + + /* convert negative indexes */ + if (start < 0) start = llen+start; + if (end < 0) end = llen+end; + if (start < 0) start = 0; + if (end < 0) end = 0; + + /* indexes sanity checks */ + if (start > end || start >= llen) { addReply(c,shared.czero); - } else { - long deleted; + return; + } + if (end >= llen) end = llen-1; - if (zsetobj->type != REDIS_ZSET) { - addReply(c,shared.wrongtypeerr); - return; + /* increment start and end because zsl*Rank functions + * use 1-based rank */ + deleted = zslDeleteRangeByRank(zs->zsl,start+1,end+1,zs->dict); + if (htNeedsResize(zs->dict)) dictResize(zs->dict); + server.dirty += deleted; + addReplyLong(c, deleted); +} + +typedef struct { + dict *dict; + double weight; +} zsetopsrc; + +static int qsortCompareZsetopsrcByCardinality(const void *s1, const void *s2) { + zsetopsrc *d1 = (void*) s1, *d2 = (void*) s2; + unsigned long size1, size2; + size1 = d1->dict ? dictSize(d1->dict) : 0; + size2 = d2->dict ? dictSize(d2->dict) : 0; + return size1 - size2; +} + +static void zunionInterGenericCommand(redisClient *c, robj *dstkey, int op) { + int i, j, zsetnum; + zsetopsrc *src; + robj *dstobj; + zset *dstzset; + dictIterator *di; + dictEntry *de; + + /* expect zsetnum input keys to be given */ + zsetnum = atoi(c->argv[2]->ptr); + if (zsetnum < 1) { + addReplySds(c,sdsnew("-ERR at least 1 input key is needed for ZUNION/ZINTER\r\n")); + return; + } + + /* test if the expected number of keys would overflow */ + if (3+zsetnum > c->argc) { + addReply(c,shared.syntaxerr); + return; + } + + /* read keys to be used for input */ + src = zmalloc(sizeof(zsetopsrc) * zsetnum); + for (i = 0, j = 3; i < zsetnum; i++, j++) { + robj *zsetobj = lookupKeyWrite(c->db,c->argv[j]); + if (!zsetobj) { + src[i].dict = NULL; + } else { + if (zsetobj->type != REDIS_ZSET) { + zfree(src); + addReply(c,shared.wrongtypeerr); + return; + } + src[i].dict = ((zset*)zsetobj->ptr)->dict; } - zs = zsetobj->ptr; - deleted = zslDeleteRange(zs->zsl,min,max,zs->dict); - if (htNeedsResize(zs->dict)) dictResize(zs->dict); - server.dirty += deleted; - addReplySds(c,sdscatprintf(sdsempty(),":%lu\r\n",deleted)); + + /* default all weights to 1 */ + src[i].weight = 1.0; } + + /* parse optional extra arguments */ + if (j < c->argc) { + int remaining = c->argc-j; + + while (remaining) { + if (!strcasecmp(c->argv[j]->ptr,"weights")) { + j++; remaining--; + if (remaining < zsetnum) { + zfree(src); + addReplySds(c,sdsnew("-ERR not enough weights for ZUNION/ZINTER\r\n")); + return; + } + for (i = 0; i < zsetnum; i++, j++, remaining--) { + src[i].weight = strtod(c->argv[j]->ptr, NULL); + } + } else { + zfree(src); + addReply(c,shared.syntaxerr); + return; + } + } + } + + dstobj = createZsetObject(); + dstzset = dstobj->ptr; + + if (op == REDIS_OP_INTER) { + /* sort sets from the smallest to largest, this will improve our + * algorithm's performance */ + qsort(src,zsetnum,sizeof(zsetopsrc), qsortCompareZsetopsrcByCardinality); + + /* skip going over all entries if the smallest zset is NULL or empty */ + if (src[0].dict && dictSize(src[0].dict) > 0) { + /* precondition: as src[0].dict is non-empty and the zsets are ordered + * from small to large, all src[i > 0].dict are non-empty too */ + di = dictGetIterator(src[0].dict); + while((de = dictNext(di)) != NULL) { + double *score = zmalloc(sizeof(double)); + *score = 0.0; + + for (j = 0; j < zsetnum; j++) { + dictEntry *other = (j == 0) ? de : dictFind(src[j].dict,dictGetEntryKey(de)); + if (other) { + *score = *score + src[j].weight * (*(double*)dictGetEntryVal(other)); + } else { + break; + } + } + + /* skip entry when not present in every source dict */ + if (j != zsetnum) { + zfree(score); + } else { + robj *o = dictGetEntryKey(de); + dictAdd(dstzset->dict,o,score); + incrRefCount(o); /* added to dictionary */ + zslInsert(dstzset->zsl,*score,o); + incrRefCount(o); /* added to skiplist */ + } + } + dictReleaseIterator(di); + } + } else if (op == REDIS_OP_UNION) { + for (i = 0; i < zsetnum; i++) { + if (!src[i].dict) continue; + + di = dictGetIterator(src[i].dict); + while((de = dictNext(di)) != NULL) { + /* skip key when already processed */ + if (dictFind(dstzset->dict,dictGetEntryKey(de)) != NULL) continue; + + double *score = zmalloc(sizeof(double)); + *score = 0.0; + for (j = 0; j < zsetnum; j++) { + if (!src[j].dict) continue; + + dictEntry *other = (i == j) ? de : dictFind(src[j].dict,dictGetEntryKey(de)); + if (other) { + *score = *score + src[j].weight * (*(double*)dictGetEntryVal(other)); + } + } + + robj *o = dictGetEntryKey(de); + dictAdd(dstzset->dict,o,score); + incrRefCount(o); /* added to dictionary */ + zslInsert(dstzset->zsl,*score,o); + incrRefCount(o); /* added to skiplist */ + } + dictReleaseIterator(di); + } + } else { + /* unknown operator */ + redisAssert(op == REDIS_OP_INTER || op == REDIS_OP_UNION); + } + + deleteKey(c->db,dstkey); + dictAdd(c->db->dict,dstkey,dstobj); + incrRefCount(dstkey); + + addReplyLong(c, dstzset->zsl->length); + server.dirty++; + zfree(src); +} + +static void zunionCommand(redisClient *c) { + zunionInterGenericCommand(c,c->argv[1], REDIS_OP_UNION); +} + +static void zinterCommand(redisClient *c) { + zunionInterGenericCommand(c,c->argv[1], REDIS_OP_INTER); } static void zrangeGenericCommand(redisClient *c, int reverse) { @@ -5136,6 +5544,12 @@ static void zrangeGenericCommand(redisClient *c, int reverse) { int start = atoi(c->argv[2]->ptr); int end = atoi(c->argv[3]->ptr); int withscores = 0; + int llen; + int rangelen, j; + zset *zsetobj; + zskiplist *zsl; + zskiplistNode *ln; + robj *ele; if (c->argc == 5 && !strcasecmp(c->argv[4]->ptr,"withscores")) { withscores = 1; @@ -5144,59 +5558,45 @@ static void zrangeGenericCommand(redisClient *c, int reverse) { return; } - o = lookupKeyRead(c->db,c->argv[1]); - if (o == NULL) { - addReply(c,shared.nullmultibulk); - } else { - if (o->type != REDIS_ZSET) { - addReply(c,shared.wrongtypeerr); - } else { - zset *zsetobj = o->ptr; - zskiplist *zsl = zsetobj->zsl; - zskiplistNode *ln; + if ((o = lookupKeyReadOrReply(c,c->argv[1],shared.nullmultibulk)) == NULL || + checkType(c,o,REDIS_ZSET)) return; + zsetobj = o->ptr; + zsl = zsetobj->zsl; + llen = zsl->length; - int llen = zsl->length; - int rangelen, j; - robj *ele; + /* convert negative indexes */ + if (start < 0) start = llen+start; + if (end < 0) end = llen+end; + if (start < 0) start = 0; + if (end < 0) end = 0; - /* convert negative indexes */ - if (start < 0) start = llen+start; - if (end < 0) end = llen+end; - if (start < 0) start = 0; - if (end < 0) end = 0; + /* indexes sanity checks */ + if (start > end || start >= llen) { + /* Out of range start or start > end result in empty list */ + addReply(c,shared.emptymultibulk); + return; + } + if (end >= llen) end = llen-1; + rangelen = (end-start)+1; - /* indexes sanity checks */ - if (start > end || start >= llen) { - /* Out of range start or start > end result in empty list */ - addReply(c,shared.emptymultibulk); - return; - } - if (end >= llen) end = llen-1; - rangelen = (end-start)+1; - - /* Return the result in form of a multi-bulk reply */ - if (reverse) { - ln = zsl->tail; - while (start--) - ln = ln->backward; - } else { - ln = zsl->header->forward[0]; - while (start--) - ln = ln->forward[0]; - } + /* check if starting point is trivial, before searching + * the element in log(N) time */ + if (reverse) { + ln = start == 0 ? zsl->tail : zslGetElementByRank(zsl, llen-start); + } else { + ln = start == 0 ? + zsl->header->forward[0] : zslGetElementByRank(zsl, start+1); + } - addReplySds(c,sdscatprintf(sdsempty(),"*%d\r\n", - withscores ? (rangelen*2) : rangelen)); - for (j = 0; j < rangelen; j++) { - ele = ln->obj; - addReplyBulkLen(c,ele); - addReply(c,ele); - addReply(c,shared.crlf); - if (withscores) - addReplyDouble(c,ln->score); - ln = reverse ? ln->backward : ln->forward[0]; - } - } + /* Return the result in form of a multi-bulk reply */ + addReplySds(c,sdscatprintf(sdsempty(),"*%d\r\n", + withscores ? (rangelen*2) : rangelen)); + for (j = 0; j < rangelen; j++) { + ele = ln->obj; + addReplyBulk(c,ele); + if (withscores) + addReplyDouble(c,ln->score); + ln = reverse ? ln->backward : ln->forward[0]; } } @@ -5306,9 +5706,7 @@ static void genericZrangebyscoreCommand(redisClient *c, int justcount) { if (limit == 0) break; if (!justcount) { ele = ln->obj; - addReplyBulkLen(c,ele); - addReply(c,ele); - addReply(c,shared.crlf); + addReplyBulk(c,ele); if (withscores) addReplyDouble(c,ln->score); } @@ -5337,46 +5735,281 @@ static void zcountCommand(redisClient *c) { static void zcardCommand(redisClient *c) { robj *o; zset *zs; - - o = lookupKeyRead(c->db,c->argv[1]); - if (o == NULL) { - addReply(c,shared.czero); - return; + + if ((o = lookupKeyReadOrReply(c,c->argv[1],shared.czero)) == NULL || + checkType(c,o,REDIS_ZSET)) return; + + zs = o->ptr; + addReplyUlong(c,zs->zsl->length); +} + +static void zscoreCommand(redisClient *c) { + robj *o; + zset *zs; + dictEntry *de; + + if ((o = lookupKeyReadOrReply(c,c->argv[1],shared.nullbulk)) == NULL || + checkType(c,o,REDIS_ZSET)) return; + + zs = o->ptr; + de = dictFind(zs->dict,c->argv[2]); + if (!de) { + addReply(c,shared.nullbulk); } else { - if (o->type != REDIS_ZSET) { - addReply(c,shared.wrongtypeerr); - } else { - zs = o->ptr; - addReplySds(c,sdscatprintf(sdsempty(),":%lu\r\n",zs->zsl->length)); - } + double *score = dictGetEntryVal(de); + + addReplyDouble(c,*score); } } -static void zscoreCommand(redisClient *c) { +static void zrankGenericCommand(redisClient *c, int reverse) { robj *o; zset *zs; - - o = lookupKeyRead(c->db,c->argv[1]); - if (o == NULL) { + zskiplist *zsl; + dictEntry *de; + unsigned long rank; + double *score; + + if ((o = lookupKeyReadOrReply(c,c->argv[1],shared.nullbulk)) == NULL || + checkType(c,o,REDIS_ZSET)) return; + + zs = o->ptr; + zsl = zs->zsl; + de = dictFind(zs->dict,c->argv[2]); + if (!de) { addReply(c,shared.nullbulk); return; + } + + score = dictGetEntryVal(de); + rank = zslGetRank(zsl, *score, c->argv[2]); + if (rank) { + if (reverse) { + addReplyLong(c, zsl->length - rank); + } else { + addReplyLong(c, rank-1); + } } else { - if (o->type != REDIS_ZSET) { + addReply(c,shared.nullbulk); + } +} + +static void zrankCommand(redisClient *c) { + zrankGenericCommand(c, 0); +} + +static void zrevrankCommand(redisClient *c) { + zrankGenericCommand(c, 1); +} + +/* =================================== Hashes =============================== */ +static void hsetCommand(redisClient *c) { + int update = 0; + robj *o = lookupKeyWrite(c->db,c->argv[1]); + + if (o == NULL) { + o = createHashObject(); + dictAdd(c->db->dict,c->argv[1],o); + incrRefCount(c->argv[1]); + } else { + if (o->type != REDIS_HASH) { addReply(c,shared.wrongtypeerr); + return; + } + } + /* We want to convert the zipmap into an hash table right now if the + * entry to be added is too big. Note that we check if the object + * is integer encoded before to try fetching the length in the test below. + * This is because integers are small, but currently stringObjectLen() + * performs a slow conversion: not worth it. */ + if (o->encoding == REDIS_ENCODING_ZIPMAP && + ((c->argv[2]->encoding == REDIS_ENCODING_RAW && + sdslen(c->argv[2]->ptr) > server.hash_max_zipmap_value) || + (c->argv[3]->encoding == REDIS_ENCODING_RAW && + sdslen(c->argv[3]->ptr) > server.hash_max_zipmap_value))) + { + convertToRealHash(o); + } + + if (o->encoding == REDIS_ENCODING_ZIPMAP) { + unsigned char *zm = o->ptr; + robj *valobj = getDecodedObject(c->argv[3]); + + zm = zipmapSet(zm,c->argv[2]->ptr,sdslen(c->argv[2]->ptr), + valobj->ptr,sdslen(valobj->ptr),&update); + decrRefCount(valobj); + o->ptr = zm; + + /* And here there is the second check for hash conversion... + * we want to do it only if the operation was not just an update as + * zipmapLen() is O(N). */ + if (!update && zipmapLen(zm) > server.hash_max_zipmap_entries) + convertToRealHash(o); + } else { + tryObjectEncoding(c->argv[2]); + /* note that c->argv[3] is already encoded, as the latest arg + * of a bulk command is always integer encoded if possible. */ + if (dictAdd(o->ptr,c->argv[2],c->argv[3]) == DICT_OK) { + incrRefCount(c->argv[2]); } else { - dictEntry *de; + update = 1; + } + incrRefCount(c->argv[3]); + } + server.dirty++; + addReplySds(c,sdscatprintf(sdsempty(),":%d\r\n",update == 0)); +} - zs = o->ptr; - de = dictFind(zs->dict,c->argv[2]); - if (!de) { - addReply(c,shared.nullbulk); - } else { - double *score = dictGetEntryVal(de); +static void hgetCommand(redisClient *c) { + robj *o; + + if ((o = lookupKeyReadOrReply(c,c->argv[1],shared.nullbulk)) == NULL || + checkType(c,o,REDIS_HASH)) return; + + if (o->encoding == REDIS_ENCODING_ZIPMAP) { + unsigned char *zm = o->ptr; + unsigned char *val; + unsigned int vlen; + + if (zipmapGet(zm,c->argv[2]->ptr,sdslen(c->argv[2]->ptr), &val,&vlen)) { + addReplySds(c,sdscatprintf(sdsempty(),"$%u\r\n", vlen)); + addReplySds(c,sdsnewlen(val,vlen)); + addReply(c,shared.crlf); + return; + } else { + addReply(c,shared.nullbulk); + return; + } + } else { + struct dictEntry *de; + + de = dictFind(o->ptr,c->argv[2]); + if (de == NULL) { + addReply(c,shared.nullbulk); + } else { + robj *e = dictGetEntryVal(de); + + addReplyBulk(c,e); + } + } +} + +static void hdelCommand(redisClient *c) { + robj *o; + int deleted = 0; + + if ((o = lookupKeyWriteOrReply(c,c->argv[1],shared.czero)) == NULL || + checkType(c,o,REDIS_HASH)) return; + + if (o->encoding == REDIS_ENCODING_ZIPMAP) { + o->ptr = zipmapDel((unsigned char*) o->ptr, + (unsigned char*) c->argv[2]->ptr, + sdslen(c->argv[2]->ptr), &deleted); + } else { + deleted = dictDelete((dict*)o->ptr,c->argv[2]) == DICT_OK; + } + addReply(c,deleted ? shared.cone : shared.czero); +} + +static void hlenCommand(redisClient *c) { + robj *o; + unsigned long len; + + if ((o = lookupKeyReadOrReply(c,c->argv[1],shared.czero)) == NULL || + checkType(c,o,REDIS_HASH)) return; + + len = (o->encoding == REDIS_ENCODING_ZIPMAP) ? + zipmapLen((unsigned char*)o->ptr) : dictSize((dict*)o->ptr); + addReplyUlong(c,len); +} + +#define REDIS_GETALL_KEYS 1 +#define REDIS_GETALL_VALS 2 +static void genericHgetallCommand(redisClient *c, int flags) { + robj *o, *lenobj; + unsigned long count = 0; + + if ((o = lookupKeyReadOrReply(c,c->argv[1],shared.nullmultibulk)) == NULL + || checkType(c,o,REDIS_HASH)) return; + + lenobj = createObject(REDIS_STRING,NULL); + addReply(c,lenobj); + decrRefCount(lenobj); + + if (o->encoding == REDIS_ENCODING_ZIPMAP) { + unsigned char *p = zipmapRewind(o->ptr); + unsigned char *field, *val; + unsigned int flen, vlen; + + while((p = zipmapNext(p,&field,&flen,&val,&vlen)) != NULL) { + robj *aux; + + if (flags & REDIS_GETALL_KEYS) { + aux = createStringObject((char*)field,flen); + addReplyBulk(c,aux); + decrRefCount(aux); + count++; + } + if (flags & REDIS_GETALL_VALS) { + aux = createStringObject((char*)val,vlen); + addReplyBulk(c,aux); + decrRefCount(aux); + count++; + } + } + } else { + dictIterator *di = dictGetIterator(o->ptr); + dictEntry *de; - addReplyDouble(c,*score); + while((de = dictNext(di)) != NULL) { + robj *fieldobj = dictGetEntryKey(de); + robj *valobj = dictGetEntryVal(de); + + if (flags & REDIS_GETALL_KEYS) { + addReplyBulk(c,fieldobj); + count++; + } + if (flags & REDIS_GETALL_VALS) { + addReplyBulk(c,valobj); + count++; } } + dictReleaseIterator(di); } + lenobj->ptr = sdscatprintf(sdsempty(),"*%lu\r\n",count); +} + +static void hkeysCommand(redisClient *c) { + genericHgetallCommand(c,REDIS_GETALL_KEYS); +} + +static void hvalsCommand(redisClient *c) { + genericHgetallCommand(c,REDIS_GETALL_VALS); +} + +static void hgetallCommand(redisClient *c) { + genericHgetallCommand(c,REDIS_GETALL_KEYS|REDIS_GETALL_VALS); +} + +static void convertToRealHash(robj *o) { + unsigned char *key, *val, *p, *zm = o->ptr; + unsigned int klen, vlen; + dict *dict = dictCreate(&hashDictType,NULL); + + assert(o->type == REDIS_HASH && o->encoding != REDIS_ENCODING_HT); + p = zipmapRewind(zm); + while((p = zipmapNext(p,&key,&klen,&val,&vlen)) != NULL) { + robj *keyobj, *valobj; + + keyobj = createStringObject((char*)key,klen); + valobj = createStringObject((char*)val,vlen); + tryObjectEncoding(keyobj); + tryObjectEncoding(valobj); + dictAdd(dict,keyobj,valobj); + } + o->encoding = REDIS_ENCODING_HT; + o->ptr = dict; + zfree(zm); } /* ========================= Non type-specific commands ==================== */ @@ -5683,11 +6316,7 @@ static void sortCommand(redisClient *c) { listNode *ln; listIter li; - if (!getop) { - addReplyBulkLen(c,vector[j].obj); - addReply(c,vector[j].obj); - addReply(c,shared.crlf); - } + if (!getop) addReplyBulk(c,vector[j].obj); listRewind(operations,&li); while((ln = listNext(&li))) { redisSortOperation *sop = ln->value; @@ -5698,9 +6327,7 @@ static void sortCommand(redisClient *c) { if (!val || val->type != REDIS_STRING) { addReply(c,shared.nullbulk); } else { - addReplyBulkLen(c,val); - addReply(c,val); - addReply(c,shared.crlf); + addReplyBulk(c,val); } } else { redisAssert(sop->type == REDIS_SORT_GET); /* always fails */ @@ -5787,6 +6414,9 @@ static sds genRedisInfoString(void) { time_t uptime = time(NULL)-server.stat_starttime; int j; char hmem[64]; + + server.hash_max_zipmap_entries = REDIS_HASH_MAX_ZIPMAP_ENTRIES; + server.hash_max_zipmap_value = REDIS_HASH_MAX_ZIPMAP_VALUE; bytesToHuman(hmem,zmalloc_used_memory()); info = sdscatprintf(sdsempty(), @@ -5807,6 +6437,8 @@ static sds genRedisInfoString(void) { "bgrewriteaof_in_progress:%d\r\n" "total_connections_received:%lld\r\n" "total_commands_processed:%lld\r\n" + "hash_max_zipmap_entries:%ld\r\n" + "hash_max_zipmap_value:%ld\r\n" "vm_enabled:%d\r\n" "role:%s\r\n" ,REDIS_VERSION, @@ -5826,6 +6458,8 @@ static sds genRedisInfoString(void) { server.bgrewritechildpid != -1, server.stat_numconnections, server.stat_numcommands, + server.hash_max_zipmap_entries, + server.hash_max_zipmap_value, server.vm_enabled != 0, server.masterhost == NULL ? "master" : "slave" ); @@ -6211,12 +6845,8 @@ static int handleClientsWaitingListPush(redisClient *c, robj *key, robj *ele) { 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); + addReplyBulk(receiver,key); + addReplyBulk(receiver,ele); unblockClientWaitingData(receiver); return 1; } @@ -6255,9 +6885,7 @@ static void blockingPopGenericCommand(redisClient *c, int where) { * 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); + addReplyBulk(c,argv[1]); popGenericCommand(c,where); /* Fix the client structure with the original stuff */ @@ -6519,9 +7147,9 @@ static void updateSlavesWaitingBgsave(int bgsaveerr) { static int syncWithMaster(void) { char buf[1024], tmpfile[256], authcmd[1024]; - int dumpsize; + long dumpsize; int fd = anetTcpConnect(NULL,server.masterhost,server.masterport); - int dfd; + int dfd, maxtries = 5; if (fd == -1) { redisLog(REDIS_WARNING,"Unable to connect to MASTER: %s", @@ -6571,11 +7199,16 @@ static int syncWithMaster(void) { redisLog(REDIS_WARNING,"Bad protocol from MASTER, the first byte is not '$', are you sure the host and port are right?"); return REDIS_ERR; } - dumpsize = atoi(buf+1); - redisLog(REDIS_NOTICE,"Receiving %d bytes data dump from MASTER",dumpsize); + dumpsize = strtol(buf+1,NULL,10); + redisLog(REDIS_NOTICE,"Receiving %ld bytes data dump from MASTER",dumpsize); /* Read the bulk write data on a temp file */ - snprintf(tmpfile,256,"temp-%d.%ld.rdb",(int)time(NULL),(long int)random()); - dfd = open(tmpfile,O_CREAT|O_WRONLY,0644); + while(maxtries--) { + snprintf(tmpfile,256, + "temp-%d.%ld.rdb",(int)time(NULL),(long int)getpid()); + dfd = open(tmpfile,O_CREAT|O_WRONLY|O_EXCL,0644); + if (dfd != -1) break; + sleep(1); + } if (dfd == -1) { close(fd); redisLog(REDIS_WARNING,"Opening the temp file needed for MASTER <-> SLAVE synchronization: %s",strerror(errno)); @@ -6915,7 +7548,7 @@ fmterr: } /* Write an object into a file in the bulk format $\r\n\r\n */ -static int fwriteBulk(FILE *fp, robj *obj) { +static int fwriteBulkObject(FILE *fp, robj *obj) { char buf[128]; int decrrc = 0; @@ -6940,6 +7573,18 @@ err: return 0; } +/* Write binary-safe string into a file in the bulkformat + * $\r\n\r\n */ +static int fwriteBulkString(FILE *fp, char *s, unsigned long len) { + char buf[128]; + + snprintf(buf,sizeof(buf),"$%ld\r\n",(unsigned long)len); + if (fwrite(buf,strlen(buf),1,fp) == 0) return 0; + if (len && fwrite(s,len,1,fp) == 0) return 0; + if (fwrite("\r\n",2,1,fp) == 0) return 0; + return 1; +} + /* Write a double value in bulk format $\r\n\r\n */ static int fwriteBulkDouble(FILE *fp, double d) { char buf[128], dbuf[128]; @@ -7022,8 +7667,8 @@ static int rewriteAppendOnlyFile(char *filename) { 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; + if (fwriteBulkObject(fp,key) == 0) goto werr; + if (fwriteBulkObject(fp,o) == 0) goto werr; } else if (o->type == REDIS_LIST) { /* Emit the RPUSHes needed to rebuild the list */ list *list = o->ptr; @@ -7036,8 +7681,8 @@ static int rewriteAppendOnlyFile(char *filename) { 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; + if (fwriteBulkObject(fp,key) == 0) goto werr; + if (fwriteBulkObject(fp,eleobj) == 0) goto werr; } } else if (o->type == REDIS_SET) { /* Emit the SADDs needed to rebuild the set */ @@ -7050,8 +7695,8 @@ static int rewriteAppendOnlyFile(char *filename) { 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; + if (fwriteBulkObject(fp,key) == 0) goto werr; + if (fwriteBulkObject(fp,eleobj) == 0) goto werr; } dictReleaseIterator(di); } else if (o->type == REDIS_ZSET) { @@ -7066,13 +7711,45 @@ static int rewriteAppendOnlyFile(char *filename) { double *score = dictGetEntryVal(de); if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr; - if (fwriteBulk(fp,key) == 0) goto werr; + if (fwriteBulkObject(fp,key) == 0) goto werr; if (fwriteBulkDouble(fp,*score) == 0) goto werr; - if (fwriteBulk(fp,eleobj) == 0) goto werr; + if (fwriteBulkObject(fp,eleobj) == 0) goto werr; } dictReleaseIterator(di); + } else if (o->type == REDIS_HASH) { + char cmd[]="*4\r\n$4\r\nHSET\r\n"; + + /* Emit the HSETs needed to rebuild the hash */ + if (o->encoding == REDIS_ENCODING_ZIPMAP) { + unsigned char *p = zipmapRewind(o->ptr); + unsigned char *field, *val; + unsigned int flen, vlen; + + while((p = zipmapNext(p,&field,&flen,&val,&vlen)) != NULL) { + if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr; + if (fwriteBulkObject(fp,key) == 0) goto werr; + if (fwriteBulkString(fp,(char*)field,flen) == -1) + return -1; + if (fwriteBulkString(fp,(char*)val,vlen) == -1) + return -1; + } + } else { + dictIterator *di = dictGetIterator(o->ptr); + dictEntry *de; + + while((de = dictNext(di)) != NULL) { + robj *field = dictGetEntryKey(de); + robj *val = dictGetEntryVal(de); + + if (fwrite(cmd,sizeof(cmd)-1,1,fp) == 0) goto werr; + if (fwriteBulkObject(fp,key) == 0) goto werr; + if (fwriteBulkObject(fp,field) == -1) return -1; + if (fwriteBulkObject(fp,val) == -1) return -1; + } + dictReleaseIterator(di); + } } else { - redisAssert(0 != 0); + redisAssert(0); } /* Save the expire time */ if (expiretime != -1) { @@ -7080,7 +7757,7 @@ static int rewriteAppendOnlyFile(char *filename) { /* 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 (fwriteBulkObject(fp,key) == 0) goto werr; if (fwriteBulkLong(fp,expiretime) == 0) goto werr; } if (swapped) decrRefCount(o); @@ -8249,11 +8926,20 @@ static void debugCommand(redisClient *c) { val = dictGetEntryVal(de); if (!server.vm_enabled || (key->storage == REDIS_VM_MEMORY || key->storage == REDIS_VM_SWAPPING)) { + char *strenc; + char buf[128]; + + if (val->encoding < (sizeof(strencoding)/sizeof(char*))) { + strenc = strencoding[val->encoding]; + } else { + snprintf(buf,64,"unknown encoding %d\n", val->encoding); + strenc = buf; + } addReplySds(c,sdscatprintf(sdsempty(), "+Key at:%p refcount:%d, value at:%p refcount:%d " - "encoding:%d serializedlength:%lld\r\n", + "encoding:%s serializedlength:%lld\r\n", (void*)key, key->refcount, (void*)val, val->refcount, - val->encoding, (long long) rdbSavedObjectLen(val,NULL))); + strenc, (long long) rdbSavedObjectLen(val,NULL))); } else { addReplySds(c,sdscatprintf(sdsempty(), "+Key at:%p refcount:%d, value swapped at: page %llu "