Merge branch 'unstable' of github.com:/antirez/redis into unstable

[redis.git] / src / db.c

diff --git a/src/db.c b/src/db.c
index 8684bb1ab0b7e9bca7418dfbc157558ca065d014..cc9810b6bdebe5b2254e5ef46c517608491cde95 100644 (file)
--- a/src/db.c
+++ b/src/db.c
@@ -2,10 +2,35 @@
 
 #include <signal.h>
 
 
 #include <signal.h>
 

+void SlotToKeyAdd(robj *key);
+void SlotToKeyDel(robj *key);
+

 /*-----------------------------------------------------------------------------
  * C-level DB API
  *----------------------------------------------------------------------------*/
 
 /*-----------------------------------------------------------------------------
  * C-level DB API
  *----------------------------------------------------------------------------*/
 

+/* Important notes on lookup and disk store.
+ *
+ * When disk store is enabled on lookup we can have different cases.
+ *
+ * a) The key is in memory:
+ *    - If the key is not in IO_SAVEINPROG state we can access it.
+ *      As if it's just IO_SAVE this means we have the key in the IO queue
+ *      but can't be accessed by the IO thread (it requires to be
+ *      translated into an IO Job by the cache cron function.)
+ *    - If the key is in IO_SAVEINPROG we can't touch the key and have
+ *      to blocking wait completion of operations.
+ * b) The key is not in memory:
+ *    - If it's marked as non existing on disk as well (negative cache)
+ *      we don't need to perform the disk access.
+ *    - if the key MAY EXIST, but is not in memory, and it is marked as IO_SAVE
+ *      then the key can only be a deleted one. As IO_SAVE keys are never
+ *      evicted (dirty state), so the only possibility is that key was deleted.
+ *    - if the key MAY EXIST we need to blocking load it.
+ *      We check that the key is not in IO_SAVEINPROG state before accessing
+ *      the disk object. If it is in this state, we wait.
+ */
+

 robj *lookupKey(redisDb *db, robj *key) {
     dictEntry *de = dictFind(db->dict,key->ptr);
     if (de) {
 robj *lookupKey(redisDb *db, robj *key) {
     dictEntry *de = dictFind(db->dict,key->ptr);
     if (de) {


@@ -16,47 +41,9 @@ robj *lookupKey(redisDb *db, robj *key) {
          * a copy on write madness. */
         if (server.bgsavechildpid == -1 && server.bgrewritechildpid == -1)
             val->lru = server.lruclock;
          * a copy on write madness. */
         if (server.bgsavechildpid == -1 && server.bgrewritechildpid == -1)
             val->lru = server.lruclock;

-
-        if (server.ds_enabled &&
-            cacheScheduleIOGetFlags(db,key) & REDIS_IO_SAVEINPROG)
-        {
-            /* There is a save in progress for this object!
-             * Wait for it to get out. */
-            waitEmptyIOJobsQueue();
-            processAllPendingIOJobs();
-            redisAssert(!(cacheScheduleIOGetFlags(db,key) & REDIS_IO_SAVEINPROG));
-        }

         server.stat_keyspace_hits++;
         return val;
     } else {
         server.stat_keyspace_hits++;
         return val;
     } else {

-        time_t expire;
-        robj *val;
-
-        /* Key not found in the in memory hash table, but if disk store is
-         * enabled we may have this key on disk. If so load it in memory
-         * in a blocking way. */
-        if (server.ds_enabled && cacheKeyMayExist(db,key)) {
-            if (cacheScheduleIOGetFlags(db,key) &
-                 (REDIS_IO_SAVE|REDIS_IO_SAVEINPROG))
-            {
-                /* There is a save in progress for this object!
-                 * Wait for it to get out. */
-                waitEmptyIOJobsQueue();
-                processAllPendingIOJobs();
-                redisAssert((cacheScheduleIOGetFlags(db,key) & (REDIS_IO_SAVE|REDIS_IO_SAVEINPROG)) == 0);
-            }
-
-            redisLog(REDIS_DEBUG,"Force loading key %s via lookup",
-                key->ptr);
-            val = dsGet(db,key,&expire);
-            if (val) {
-                int retval = dbAdd(db,key,val);
-                redisAssert(retval == REDIS_OK);
-                if (expire != -1) setExpire(db,key,expire);
-                server.stat_keyspace_hits++;
-                return val;
-            }
-        }

         server.stat_keyspace_misses++;
         return NULL;
     }
         server.stat_keyspace_misses++;
         return NULL;
     }


@@ -84,39 +71,45 @@ robj *lookupKeyWriteOrReply(redisClient *c, robj *key, robj *reply) {
     return o;
 }
 
     return o;
 }
 

-/* Add the key to the DB. If the key already exists REDIS_ERR is returned,
- * otherwise REDIS_OK is returned, and the caller should increment the
- * refcount of 'val'. */
-int dbAdd(redisDb *db, robj *key, robj *val) {
-    /* Perform a lookup before adding the key, as we need to copy the
-     * key value. */
-    if (dictFind(db->dict, key->ptr) != NULL) {
-        return REDIS_ERR;
-    } else {
-        sds copy = sdsdup(key->ptr);
-        dictAdd(db->dict, copy, val);
-        if (server.ds_enabled) {
-            /* FIXME: remove entry from negative cache */
-        }
-        return REDIS_OK;
-    }
+/* Add the key to the DB. It's up to the caller to increment the reference
+ * counte of the value if needed.
+ *
+ * The program is aborted if the key already exists. */
+void dbAdd(redisDb *db, robj *key, robj *val) {
+    sds copy = sdsdup(key->ptr);
+    int retval = dictAdd(db->dict, copy, val);
+
+    redisAssertWithInfo(NULL,key,retval == REDIS_OK);
+    if (server.cluster_enabled) SlotToKeyAdd(key);
+ }
+
+/* Overwrite an existing key with a new value. Incrementing the reference
+ * count of the new value is up to the caller.
+ * This function does not modify the expire time of the existing key.
+ *
+ * The program is aborted if the key was not already present. */
+void dbOverwrite(redisDb *db, robj *key, robj *val) {
+    struct dictEntry *de = dictFind(db->dict,key->ptr);
+    
+    redisAssertWithInfo(NULL,key,de != NULL);
+    dictReplace(db->dict, key->ptr, val);

 }
 
 }
 

-/* If the key does not exist, this is just like dbAdd(). Otherwise
- * the value associated to the key is replaced with the new one.
+/* High level Set operation. This function can be used in order to set
+ * a key, whatever it was existing or not, to a new object.

  *
  *

- * On update (key already existed) 0 is returned. Otherwise 1. */
-int dbReplace(redisDb *db, robj *key, robj *val) {
-    robj *oldval;
-
-    if ((oldval = dictFetchValue(db->dict,key->ptr)) == NULL) {
-        sds copy = sdsdup(key->ptr);
-        dictAdd(db->dict, copy, val);
-        return 1;
+ * 1) The ref count of the value object is incremented.
+ * 2) clients WATCHing for the destination key notified.
+ * 3) The expire time of the key is reset (the key is made persistent). */
+void setKey(redisDb *db, robj *key, robj *val) {
+    if (lookupKeyWrite(db,key) == NULL) {
+        dbAdd(db,key,val);

     } else {
     } else {

-        dictReplace(db->dict, key->ptr, val);
-        return 0;
+        dbOverwrite(db,key,val);

     }
     }

+    incrRefCount(val);
+    removeExpire(db,key);
+    touchWatchedKey(db,key);

 }
 
 int dbExists(redisDb *db, robj *key) {
 }
 
 int dbExists(redisDb *db, robj *key) {


@@ -151,21 +144,19 @@ robj *dbRandomKey(redisDb *db) {
 
 /* Delete a key, value, and associated expiration entry if any, from the DB */
 int dbDelete(redisDb *db, robj *key) {
 
 /* Delete a key, value, and associated expiration entry if any, from the DB */
 int dbDelete(redisDb *db, robj *key) {

-    /* If diskstore is enabled make sure to awake waiting clients for this key
-     * as it is not really useful to wait for a key already deleted to be
-     * loaded from disk. */
-    if (server.ds_enabled) handleClientsBlockedOnSwappedKey(db,key);
-
-    /* FIXME: we should mark this key as non existing on disk in the negative
-     * cache. */
-

     /* Deleting an entry from the expires dict will not free the sds of
      * the key, because it is shared with the main dictionary. */
     if (dictSize(db->expires) > 0) dictDelete(db->expires,key->ptr);
     /* Deleting an entry from the expires dict will not free the sds of
      * the key, because it is shared with the main dictionary. */
     if (dictSize(db->expires) > 0) dictDelete(db->expires,key->ptr);

-    return dictDelete(db->dict,key->ptr) == DICT_OK;
+    if (dictDelete(db->dict,key->ptr) == DICT_OK) {
+        if (server.cluster_enabled) SlotToKeyDel(key);
+        return 1;
+    } else {
+        return 0;
+    }

 }
 
 }
 

-/* Empty the whole database */
+/* Empty the whole database.
+ * If diskstore is enabled this function will just flush the in-memory cache. */

 long long emptyDb() {
     int j;
     long long removed = 0;
 long long emptyDb() {
     int j;
     long long removed = 0;


@@ -196,14 +187,10 @@ int selectDb(redisClient *c, int id) {
 
 void signalModifiedKey(redisDb *db, robj *key) {
     touchWatchedKey(db,key);
 
 void signalModifiedKey(redisDb *db, robj *key) {
     touchWatchedKey(db,key);

-    if (server.ds_enabled)
-        cacheScheduleIO(db,key,REDIS_IO_SAVE);

 }
 
 void signalFlushedDb(int dbid) {
     touchWatchedKeysOnFlush(dbid);
 }
 
 void signalFlushedDb(int dbid) {
     touchWatchedKeysOnFlush(dbid);

-    if (server.ds_enabled)
-        dsFlushDb(dbid);

 }
 
 /*-----------------------------------------------------------------------------
 }
 
 /*-----------------------------------------------------------------------------


@@ -226,7 +213,13 @@ void flushallCommand(redisClient *c) {
         kill(server.bgsavechildpid,SIGKILL);
         rdbRemoveTempFile(server.bgsavechildpid);
     }
         kill(server.bgsavechildpid,SIGKILL);
         rdbRemoveTempFile(server.bgsavechildpid);
     }

-    rdbSave(server.dbfilename);
+    if (server.saveparamslen > 0) {
+        /* Normally rdbSave() will reset dirty, but we don't want this here
+         * as otherwise FLUSHALL will not be replicated nor put into the AOF. */
+        int saved_dirty = server.dirty;
+        rdbSave(server.dbfilename);
+        server.dirty = saved_dirty;
+    }

     server.dirty++;
 }
 
     server.dirty++;
 }
 


@@ -234,22 +227,10 @@ void delCommand(redisClient *c) {
     int deleted = 0, j;
 
     for (j = 1; j < c->argc; j++) {
     int deleted = 0, j;
 
     for (j = 1; j < c->argc; j++) {

-        if (server.ds_enabled) {
-            lookupKeyRead(c->db,c->argv[j]);
-            /* FIXME: this can be optimized a lot, no real need to load
-             * a possibly huge value. */
-        }

         if (dbDelete(c->db,c->argv[j])) {
             signalModifiedKey(c->db,c->argv[j]);
             server.dirty++;
             deleted++;
         if (dbDelete(c->db,c->argv[j])) {
             signalModifiedKey(c->db,c->argv[j]);
             server.dirty++;
             deleted++;

-        } else if (server.ds_enabled) {
-            if (cacheKeyMayExist(c->db,c->argv[j]) &&
-                dsExists(c->db,c->argv[j]))
-            {
-                cacheScheduleIO(c->db,c->argv[j],REDIS_IO_SAVE);
-                deleted = 1;
-            }

         }
     }
     addReplyLongLong(c,deleted);
         }
     }
     addReplyLongLong(c,deleted);


@@ -267,6 +248,10 @@ void existsCommand(redisClient *c) {
 void selectCommand(redisClient *c) {
     int id = atoi(c->argv[1]->ptr);
 
 void selectCommand(redisClient *c) {
     int id = atoi(c->argv[1]->ptr);
 

+    if (server.cluster_enabled && id != 0) {
+        addReplyError(c,"SELECT is not allowed in cluster mode");
+        return;
+    }

     if (selectDb(c,id) == REDIS_ERR) {
         addReplyError(c,"invalid DB index");
     } else {
     if (selectDb(c,id) == REDIS_ERR) {
         addReplyError(c,"invalid DB index");
     } else {


@@ -341,30 +326,6 @@ void typeCommand(redisClient *c) {
     addReplyStatus(c,type);
 }
 
     addReplyStatus(c,type);
 }
 

-void saveCommand(redisClient *c) {
-    if (server.bgsavechildpid != -1) {
-        addReplyError(c,"Background save already in progress");
-        return;
-    }
-    if (rdbSave(server.dbfilename) == REDIS_OK) {
-        addReply(c,shared.ok);
-    } else {
-        addReply(c,shared.err);
-    }
-}
-
-void bgsaveCommand(redisClient *c) {
-    if (server.bgsavechildpid != -1) {
-        addReplyError(c,"Background save already in progress");
-        return;
-    }
-    if (rdbSaveBackground(server.dbfilename) == REDIS_OK) {
-        addReplyStatus(c,"Background saving started");
-    } else {
-        addReply(c,shared.err);
-    }
-}
-

 void shutdownCommand(redisClient *c) {
     if (prepareForShutdown() == REDIS_OK)
         exit(0);
 void shutdownCommand(redisClient *c) {
     if (prepareForShutdown() == REDIS_OK)
         exit(0);


@@ -373,6 +334,7 @@ void shutdownCommand(redisClient *c) {
 
 void renameGenericCommand(redisClient *c, int nx) {
     robj *o;
 
 void renameGenericCommand(redisClient *c, int nx) {
     robj *o;

+    time_t expire;

 
     /* To use the same key as src and dst is probably an error */
     if (sdscmp(c->argv[1]->ptr,c->argv[2]->ptr) == 0) {
 
     /* To use the same key as src and dst is probably an error */
     if (sdscmp(c->argv[1]->ptr,c->argv[2]->ptr) == 0) {


@@ -384,14 +346,18 @@ void renameGenericCommand(redisClient *c, int nx) {
         return;
 
     incrRefCount(o);
         return;
 
     incrRefCount(o);

-    if (dbAdd(c->db,c->argv[2],o) == REDIS_ERR) {
+    expire = getExpire(c->db,c->argv[1]);
+    if (lookupKeyWrite(c->db,c->argv[2]) != NULL) {

         if (nx) {
             decrRefCount(o);
             addReply(c,shared.czero);
             return;
         }
         if (nx) {
             decrRefCount(o);
             addReply(c,shared.czero);
             return;
         }

-        dbReplace(c->db,c->argv[2],o);
+        /* Overwrite: delete the old key before creating the new one with the same name. */
+        dbDelete(c->db,c->argv[2]);

     }
     }

+    dbAdd(c->db,c->argv[2],o);
+    if (expire != -1) setExpire(c->db,c->argv[2],expire);

     dbDelete(c->db,c->argv[1]);
     signalModifiedKey(c->db,c->argv[1]);
     signalModifiedKey(c->db,c->argv[2]);
     dbDelete(c->db,c->argv[1]);
     signalModifiedKey(c->db,c->argv[1]);
     signalModifiedKey(c->db,c->argv[2]);


@@ -412,6 +378,11 @@ void moveCommand(redisClient *c) {
     redisDb *src, *dst;
     int srcid;
 
     redisDb *src, *dst;
     int srcid;
 

+    if (server.cluster_enabled) {
+        addReplyError(c,"MOVE is not allowed in cluster mode");
+        return;
+    }
+

     /* Obtain source and target DB pointers */
     src = c->db;
     srcid = c->db->id;
     /* Obtain source and target DB pointers */
     src = c->db;
     srcid = c->db->id;


@@ -436,11 +407,12 @@ void moveCommand(redisClient *c) {
         return;
     }
 
         return;
     }
 

-    /* Try to add the element to the target DB */
-    if (dbAdd(dst,c->argv[1],o) == REDIS_ERR) {
+    /* Return zero if the key already exists in the target DB */
+    if (lookupKeyWrite(dst,c->argv[1]) != NULL) {

         addReply(c,shared.czero);
         return;
     }
         addReply(c,shared.czero);
         return;
     }

+    dbAdd(dst,c->argv[1],o);

     incrRefCount(o);
 
     /* OK! key moved, free the entry in the source DB */
     incrRefCount(o);
 
     /* OK! key moved, free the entry in the source DB */


@@ -456,7 +428,7 @@ void moveCommand(redisClient *c) {
 int removeExpire(redisDb *db, robj *key) {
     /* An expire may only be removed if there is a corresponding entry in the
      * main dict. Otherwise, the key will never be freed. */
 int removeExpire(redisDb *db, robj *key) {
     /* An expire may only be removed if there is a corresponding entry in the
      * main dict. Otherwise, the key will never be freed. */

-    redisAssert(dictFind(db->dict,key->ptr) != NULL);
+    redisAssertWithInfo(NULL,key,dictFind(db->dict,key->ptr) != NULL);

     return dictDelete(db->expires,key->ptr) == DICT_OK;
 }
 
     return dictDelete(db->expires,key->ptr) == DICT_OK;
 }
 


@@ -465,7 +437,7 @@ void setExpire(redisDb *db, robj *key, time_t when) {
 
     /* Reuse the sds from the main dict in the expire dict */
     de = dictFind(db->dict,key->ptr);
 
     /* Reuse the sds from the main dict in the expire dict */
     de = dictFind(db->dict,key->ptr);

-    redisAssert(de != NULL);
+    redisAssertWithInfo(NULL,key,de != NULL);

     dictReplace(db->expires,dictGetEntryKey(de),(void*)when);
 }
 
     dictReplace(db->expires,dictGetEntryKey(de),(void*)when);
 }
 


@@ -480,7 +452,7 @@ time_t getExpire(redisDb *db, robj *key) {
 
     /* The entry was found in the expire dict, this means it should also
      * be present in the main dict (safety check). */
 
     /* The entry was found in the expire dict, this means it should also
      * be present in the main dict (safety check). */

-    redisAssert(dictFind(db->dict,key->ptr) != NULL);
+    redisAssertWithInfo(NULL,key,dictFind(db->dict,key->ptr) != NULL);

     return (time_t) dictGetEntryVal(de);
 }
 
     return (time_t) dictGetEntryVal(de);
 }
 


@@ -511,6 +483,11 @@ void propagateExpire(redisDb *db, robj *key) {
 int expireIfNeeded(redisDb *db, robj *key) {
     time_t when = getExpire(db,key);
 
 int expireIfNeeded(redisDb *db, robj *key) {
     time_t when = getExpire(db,key);
 

+    if (when < 0) return 0; /* No expire for this key */
+
+    /* Don't expire anything while loading. It will be done later. */
+    if (server.loading) return 0;
+

     /* If we are running in the context of a slave, return ASAP:
      * the slave key expiration is controlled by the master that will
      * send us synthesized DEL operations for expired keys.
     /* If we are running in the context of a slave, return ASAP:
      * the slave key expiration is controlled by the master that will
      * send us synthesized DEL operations for expired keys.


@@ -522,8 +499,6 @@ int expireIfNeeded(redisDb *db, robj *key) {
         return time(NULL) > when;
     }
 
         return time(NULL) > when;
     }
 

-    if (when < 0) return 0;
-

     /* Return when this key has not expired */
     if (time(NULL) <= when) return 0;
 
     /* Return when this key has not expired */
     if (time(NULL) <= when) return 0;
 


@@ -550,10 +525,24 @@ void expireGenericCommand(redisClient *c, robj *key, robj *param, long offset) {
         addReply(c,shared.czero);
         return;
     }
         addReply(c,shared.czero);
         return;
     }

-    if (seconds <= 0) {
-        if (dbDelete(c->db,key)) server.dirty++;
-        addReply(c, shared.cone);
+    /* EXPIRE with negative TTL, or EXPIREAT with a timestamp into the past
+     * should never be executed as a DEL when load the AOF or in the context
+     * of a slave instance.
+     *
+     * Instead we take the other branch of the IF statement setting an expire
+     * (possibly in the past) and wait for an explicit DEL from the master. */
+    if (seconds <= 0 && !server.loading && !server.masterhost) {
+        robj *aux;
+
+        redisAssertWithInfo(c,key,dbDelete(c->db,key));
+        server.dirty++;
+
+        /* Replicate/AOF this as an explicit DEL. */
+        aux = createStringObject("DEL",3);
+        rewriteClientCommandVector(c,2,aux,key);
+        decrRefCount(aux);

         signalModifiedKey(c->db,key);
         signalModifiedKey(c->db,key);

+        addReply(c, shared.cone);

         return;
     } else {
         time_t when = time(NULL)+seconds;
         return;
     } else {
         time_t when = time(NULL)+seconds;


@@ -599,3 +588,108 @@ void persistCommand(redisClient *c) {
         }
     }
 }
         }
     }
 }

+
+/* -----------------------------------------------------------------------------
+ * API to get key arguments from commands
+ * ---------------------------------------------------------------------------*/
+
+int *getKeysUsingCommandTable(struct redisCommand *cmd,robj **argv, int argc, int *numkeys) {
+    int j, i = 0, last, *keys;
+    REDIS_NOTUSED(argv);
+
+    if (cmd->firstkey == 0) {
+        *numkeys = 0;
+        return NULL;
+    }
+    last = cmd->lastkey;
+    if (last < 0) last = argc+last;
+    keys = zmalloc(sizeof(int)*((last - cmd->firstkey)+1));
+    for (j = cmd->firstkey; j <= last; j += cmd->keystep) {
+        redisAssert(j < argc);
+        keys[i++] = j;
+    }
+    *numkeys = i;
+    return keys;
+}
+
+int *getKeysFromCommand(struct redisCommand *cmd,robj **argv, int argc, int *numkeys, int flags) {
+    if (cmd->getkeys_proc) {
+        return cmd->getkeys_proc(cmd,argv,argc,numkeys,flags);
+    } else {
+        return getKeysUsingCommandTable(cmd,argv,argc,numkeys);
+    }
+}
+
+void getKeysFreeResult(int *result) {
+    zfree(result);
+}
+
+int *noPreloadGetKeys(struct redisCommand *cmd,robj **argv, int argc, int *numkeys, int flags) {
+    if (flags & REDIS_GETKEYS_PRELOAD) {
+        *numkeys = 0;
+        return NULL;
+    } else {
+        return getKeysUsingCommandTable(cmd,argv,argc,numkeys);
+    }
+}
+
+int *renameGetKeys(struct redisCommand *cmd,robj **argv, int argc, int *numkeys, int flags) {
+    if (flags & REDIS_GETKEYS_PRELOAD) {
+        int *keys = zmalloc(sizeof(int));
+        *numkeys = 1;
+        keys[0] = 1;
+        return keys;
+    } else {
+        return getKeysUsingCommandTable(cmd,argv,argc,numkeys);
+    }
+}
+
+int *zunionInterGetKeys(struct redisCommand *cmd,robj **argv, int argc, int *numkeys, int flags) {
+    int i, num, *keys;
+    REDIS_NOTUSED(cmd);
+    REDIS_NOTUSED(flags);
+
+    num = atoi(argv[2]->ptr);
+    /* Sanity check. Don't return any key if the command is going to
+     * reply with syntax error. */
+    if (num > (argc-3)) {
+        *numkeys = 0;
+        return NULL;
+    }
+    keys = zmalloc(sizeof(int)*num);
+    for (i = 0; i < num; i++) keys[i] = 3+i;
+    *numkeys = num;
+    return keys;
+}
+
+/* Slot to Key API. This is used by Redis Cluster in order to obtain in
+ * a fast way a key that belongs to a specified hash slot. This is useful
+ * while rehashing the cluster. */
+void SlotToKeyAdd(robj *key) {
+    unsigned int hashslot = keyHashSlot(key->ptr,sdslen(key->ptr));
+
+    zslInsert(server.cluster.slots_to_keys,hashslot,key);
+    incrRefCount(key);
+}
+
+void SlotToKeyDel(robj *key) {
+    unsigned int hashslot = keyHashSlot(key->ptr,sdslen(key->ptr));
+
+    zslDelete(server.cluster.slots_to_keys,hashslot,key);
+}
+
+unsigned int GetKeysInSlot(unsigned int hashslot, robj **keys, unsigned int count) {
+    zskiplistNode *n;
+    zrangespec range;
+    int j = 0;
+
+    range.min = range.max = hashslot;
+    range.minex = range.maxex = 0;
+    
+    n = zslFirstInRange(server.cluster.slots_to_keys, range);
+    while(n && n->score == hashslot && count--) {
+        keys[j++] = n->obj;
+        n = n->level[0].forward;
+    }
+    return j;
+}