X-Git-Url: https://git.saurik.com/redis.git/blobdiff_plain/96ffb2fe97c3e77879e7a4f6f7457397a18bf233..3a73be752476f44dfb099863ac9424c2f27f2be0:/src/db.c

diff --git a/src/db.c b/src/db.c
index e1e82cb2..669fba13 100644
--- a/src/db.c
+++ b/src/db.c
@@ -6,35 +6,93 @@
  * 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.
+ */
+
+void lookupWaitBusyKey(redisDb *db, robj *key) {
+    /* FIXME: wait just for this key, not everything */
+    waitEmptyIOJobsQueue();
+    processAllPendingIOJobs();
+    redisAssert((cacheScheduleIOGetFlags(db,key) & REDIS_IO_SAVEINPROG) == 0);
+}
+
 robj *lookupKey(redisDb *db, robj *key) {
     dictEntry *de = dictFind(db->dict,key->ptr);
     if (de) {
         robj *val = dictGetEntryVal(de);
 
-        if (server.vm_enabled) {
-            if (val->storage == REDIS_VM_MEMORY ||
-                val->storage == REDIS_VM_SWAPPING)
-            {
-                /* If we were swapping the object out, cancel the operation */
-                if (val->storage == REDIS_VM_SWAPPING)
-                    vmCancelThreadedIOJob(val);
-                /* Update the access time for the aging algorithm. */
-                val->lru = server.lruclock;
-            } else {
-                int notify = (val->storage == REDIS_VM_LOADING);
+        /* Update the access time for the aging algorithm.
+         * Don't do it if we have a saving child, as this will trigger
+         * 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)
+        {
+            /* Need to wait for the key to get unbusy */
+            redisLog(REDIS_DEBUG,"Lookup found a key in SAVEINPROG state. Waiting. (Key was in the cache)");
+            lookupWaitBusyKey(db,key);
+        }
+        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)) {
+            long flags = cacheScheduleIOGetFlags(db,key);
+
+            /* They key is not in cache, but it has a SAVE op in queue?
+             * The only possibility is that the key was deleted, since
+             * dirty keys are not evicted. */
+            if (flags & REDIS_IO_SAVE) {
+                server.stat_keyspace_misses++;
+                return NULL;
+            }
 
-                /* Our value was swapped on disk. Bring it at home. */
-                redisAssert(val->type == REDIS_VMPOINTER);
-                val = vmLoadObject(val);
-                dictGetEntryVal(de) = val;
+            /* At this point we need to blocking load the key in memory.
+             * The first thing we do is waiting here if the key is busy. */
+            if (flags & REDIS_IO_SAVEINPROG) {
+                redisLog(REDIS_DEBUG,"Lookup found a key in SAVEINPROG state. Waiting (while force loading).");
+                lookupWaitBusyKey(db,key);
+            }
 
-                /* Clients blocked by the VM subsystem may be waiting for
-                 * this key... */
-                if (notify) handleClientsBlockedOnSwappedKey(db,key);
+            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;
+            } else {
+                cacheSetKeyDoesNotExist(db,key);
             }
         }
-        return val;
-    } else {
+        server.stat_keyspace_misses++;
         return NULL;
     }
 }
@@ -45,8 +103,7 @@ robj *lookupKeyRead(redisDb *db, robj *key) {
 }
 
 robj *lookupKeyWrite(redisDb *db, robj *key) {
-    deleteIfVolatile(db,key);
-    touchWatchedKey(db,key);
+    expireIfNeeded(db,key);
     return lookupKey(db,key);
 }
 
@@ -73,6 +130,7 @@ int dbAdd(redisDb *db, robj *key, robj *val) {
     } else {
         sds copy = sdsdup(key->ptr);
         dictAdd(db->dict, copy, val);
+        if (server.ds_enabled) cacheSetKeyMayExist(db,key);
         return REDIS_OK;
     }
 }
@@ -82,14 +140,19 @@ int dbAdd(redisDb *db, robj *key, robj *val) {
  *
  * On update (key already existed) 0 is returned. Otherwise 1. */
 int dbReplace(redisDb *db, robj *key, robj *val) {
-    if (dictFind(db->dict,key->ptr) == NULL) {
+    robj *oldval;
+    int retval;
+
+    if ((oldval = dictFetchValue(db->dict,key->ptr)) == NULL) {
         sds copy = sdsdup(key->ptr);
         dictAdd(db->dict, copy, val);
-        return 1;
+        retval = 1;
     } else {
         dictReplace(db->dict, key->ptr, val);
-        return 0;
+        retval = 0;
     }
+    if (server.ds_enabled) cacheSetKeyMayExist(db,key);
+    return retval;
 }
 
 int dbExists(redisDb *db, robj *key) {
@@ -124,13 +187,22 @@ robj *dbRandomKey(redisDb *db) {
 
 /* 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);
+        cacheSetKeyDoesNotExist(db,key);
+    }
+
     /* 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;
 }
 
-/* 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;
@@ -139,6 +211,7 @@ long long emptyDb() {
         removed += dictSize(server.db[j].dict);
         dictEmpty(server.db[j].dict);
         dictEmpty(server.db[j].expires);
+        if (server.ds_enabled) dictEmpty(server.db[j].io_negcache);
     }
     return removed;
 }
@@ -150,27 +223,50 @@ int selectDb(redisClient *c, int id) {
     return REDIS_OK;
 }
 
+/*-----------------------------------------------------------------------------
+ * Hooks for key space changes.
+ *
+ * Every time a key in the database is modified the function
+ * signalModifiedKey() is called.
+ *
+ * Every time a DB is flushed the function signalFlushDb() is called.
+ *----------------------------------------------------------------------------*/
+
+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);
+}
+
 /*-----------------------------------------------------------------------------
  * Type agnostic commands operating on the key space
  *----------------------------------------------------------------------------*/
 
 void flushdbCommand(redisClient *c) {
     server.dirty += dictSize(c->db->dict);
-    touchWatchedKeysOnFlush(c->db->id);
+    signalFlushedDb(c->db->id);
     dictEmpty(c->db->dict);
     dictEmpty(c->db->expires);
+    if (server.ds_enabled) dsFlushDb(c->db->id);
     addReply(c,shared.ok);
 }
 
 void flushallCommand(redisClient *c) {
-    touchWatchedKeysOnFlush(-1);
+    signalFlushedDb(-1);
     server.dirty += emptyDb();
     addReply(c,shared.ok);
     if (server.bgsavechildpid != -1) {
         kill(server.bgsavechildpid,SIGKILL);
         rdbRemoveTempFile(server.bgsavechildpid);
     }
-    rdbSave(server.dbfilename);
+    if (server.ds_enabled)
+        dsFlushDb(-1);
+    else
+        rdbSave(server.dbfilename);
     server.dirty++;
 }
 
@@ -178,10 +274,22 @@ void delCommand(redisClient *c) {
     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])) {
-            touchWatchedKey(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);
@@ -200,7 +308,7 @@ void selectCommand(redisClient *c) {
     int id = atoi(c->argv[1]->ptr);
 
     if (selectDb(c,id) == REDIS_ERR) {
-        addReplySds(c,sdsnew("-ERR invalid DB index\r\n"));
+        addReplyError(c,"invalid DB index");
     } else {
         addReply(c,shared.ok);
     }
@@ -222,19 +330,17 @@ void keysCommand(redisClient *c) {
     dictIterator *di;
     dictEntry *de;
     sds pattern = c->argv[1]->ptr;
-    int plen = sdslen(pattern);
+    int plen = sdslen(pattern), allkeys;
     unsigned long numkeys = 0;
-    robj *lenobj = createObject(REDIS_STRING,NULL);
+    void *replylen = addDeferredMultiBulkLength(c);
 
     di = dictGetIterator(c->db->dict);
-    addReply(c,lenobj);
-    decrRefCount(lenobj);
+    allkeys = (pattern[0] == '*' && pattern[1] == '\0');
     while((de = dictNext(di)) != NULL) {
         sds key = dictGetEntryKey(de);
         robj *keyobj;
 
-        if ((pattern[0] == '*' && pattern[1] == '\0') ||
-            stringmatchlen(pattern,plen,key,sdslen(key),0)) {
+        if (allkeys || stringmatchlen(pattern,plen,key,sdslen(key),0)) {
             keyobj = createStringObject(key,sdslen(key));
             if (expireIfNeeded(c->db,keyobj) == 0) {
                 addReplyBulk(c,keyobj);
@@ -244,17 +350,15 @@ void keysCommand(redisClient *c) {
         }
     }
     dictReleaseIterator(di);
-    lenobj->ptr = sdscatprintf(sdsempty(),"*%lu\r\n",numkeys);
+    setDeferredMultiBulkLength(c,replylen,numkeys);
 }
 
 void dbsizeCommand(redisClient *c) {
-    addReplySds(c,
-        sdscatprintf(sdsempty(),":%lu\r\n",dictSize(c->db->dict)));
+    addReplyLongLong(c,dictSize(c->db->dict));
 }
 
 void lastsaveCommand(redisClient *c) {
-    addReplySds(c,
-        sdscatprintf(sdsempty(),":%lu\r\n",server.lastsave));
+    addReplyLongLong(c,server.lastsave);
 }
 
 void typeCommand(redisClient *c) {
@@ -263,50 +367,24 @@ void typeCommand(redisClient *c) {
 
     o = lookupKeyRead(c->db,c->argv[1]);
     if (o == NULL) {
-        type = "+none";
+        type = "none";
     } else {
         switch(o->type) {
-        case REDIS_STRING: type = "+string"; break;
-        case REDIS_LIST: type = "+list"; break;
-        case REDIS_SET: type = "+set"; break;
-        case REDIS_ZSET: type = "+zset"; break;
-        case REDIS_HASH: type = "+hash"; break;
-        default: type = "+unknown"; break;
+        case REDIS_STRING: type = "string"; break;
+        case REDIS_LIST: type = "list"; break;
+        case REDIS_SET: type = "set"; break;
+        case REDIS_ZSET: type = "zset"; break;
+        case REDIS_HASH: type = "hash"; break;
+        default: type = "unknown"; break;
         }
     }
-    addReplySds(c,sdsnew(type));
-    addReply(c,shared.crlf);
-}
-
-void saveCommand(redisClient *c) {
-    if (server.bgsavechildpid != -1) {
-        addReplySds(c,sdsnew("-ERR background save in progress\r\n"));
-        return;
-    }
-    if (rdbSave(server.dbfilename) == REDIS_OK) {
-        addReply(c,shared.ok);
-    } else {
-        addReply(c,shared.err);
-    }
-}
-
-void bgsaveCommand(redisClient *c) {
-    if (server.bgsavechildpid != -1) {
-        addReplySds(c,sdsnew("-ERR background save already in progress\r\n"));
-        return;
-    }
-    if (rdbSaveBackground(server.dbfilename) == REDIS_OK) {
-        char *status = "+Background saving started\r\n";
-        addReplySds(c,sdsnew(status));
-    } else {
-        addReply(c,shared.err);
-    }
+    addReplyStatus(c,type);
 }
 
 void shutdownCommand(redisClient *c) {
     if (prepareForShutdown() == REDIS_OK)
         exit(0);
-    addReplySds(c, sdsnew("-ERR Errors trying to SHUTDOWN. Check logs.\r\n"));
+    addReplyError(c,"Errors trying to SHUTDOWN. Check logs.");
 }
 
 void renameGenericCommand(redisClient *c, int nx) {
@@ -322,7 +400,6 @@ void renameGenericCommand(redisClient *c, int nx) {
         return;
 
     incrRefCount(o);
-    deleteIfVolatile(c->db,c->argv[2]);
     if (dbAdd(c->db,c->argv[2],o) == REDIS_ERR) {
         if (nx) {
             decrRefCount(o);
@@ -332,7 +409,8 @@ void renameGenericCommand(redisClient *c, int nx) {
         dbReplace(c->db,c->argv[2],o);
     }
     dbDelete(c->db,c->argv[1]);
-    touchWatchedKey(c->db,c->argv[2]);
+    signalModifiedKey(c->db,c->argv[1]);
+    signalModifiedKey(c->db,c->argv[2]);
     server.dirty++;
     addReply(c,nx ? shared.cone : shared.ok);
 }
@@ -375,7 +453,6 @@ void moveCommand(redisClient *c) {
     }
 
     /* Try to add the element to the target DB */
-    deleteIfVolatile(dst,c->argv[1]);
     if (dbAdd(dst,c->argv[1],o) == REDIS_ERR) {
         addReply(c,shared.czero);
         return;
@@ -396,23 +473,16 @@ 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);
-    if (dictDelete(db->expires,key->ptr) == DICT_OK) {
-        return 1;
-    } else {
-        return 0;
-    }
+    return dictDelete(db->expires,key->ptr) == DICT_OK;
 }
 
-int setExpire(redisDb *db, robj *key, time_t when) {
+void setExpire(redisDb *db, robj *key, time_t when) {
     dictEntry *de;
 
     /* Reuse the sds from the main dict in the expire dict */
-    redisAssert((de = dictFind(db->dict,key->ptr)) != NULL);
-    if (dictAdd(db->expires,dictGetEntryKey(de),(void*)when) == DICT_ERR) {
-        return 0;
-    } else {
-        return 1;
-    }
+    de = dictFind(db->dict,key->ptr);
+    redisAssert(de != NULL);
+    dictReplace(db->expires,dictGetEntryKey(de),(void*)when);
 }
 
 /* Return the expire time of the specified key, or -1 if no expire
@@ -430,25 +500,52 @@ time_t getExpire(redisDb *db, robj *key) {
     return (time_t) dictGetEntryVal(de);
 }
 
+/* Propagate expires into slaves and the AOF file.
+ * When a key expires in the master, a DEL operation for this key is sent
+ * to all the slaves and the AOF file if enabled.
+ *
+ * This way the key expiry is centralized in one place, and since both
+ * AOF and the master->slave link guarantee operation ordering, everything
+ * will be consistent even if we allow write operations against expiring
+ * keys. */
+void propagateExpire(redisDb *db, robj *key) {
+    robj *argv[2];
+
+    argv[0] = createStringObject("DEL",3);
+    argv[1] = key;
+    incrRefCount(key);
+
+    if (server.appendonly)
+        feedAppendOnlyFile(server.delCommand,db->id,argv,2);
+    if (listLength(server.slaves))
+        replicationFeedSlaves(server.slaves,db->id,argv,2);
+
+    decrRefCount(argv[0]);
+    decrRefCount(argv[1]);
+}
+
 int expireIfNeeded(redisDb *db, robj *key) {
     time_t when = getExpire(db,key);
-    if (when < 0) return 0;
+
+    if (when < 0) return 0; /* No expire for this key */
+
+    /* 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.
+     *
+     * Still we try to return the right information to the caller, 
+     * that is, 0 if we think the key should be still valid, 1 if
+     * we think the key is expired at this time. */
+    if (server.masterhost != NULL) {
+        return time(NULL) > when;
+    }
 
     /* Return when this key has not expired */
     if (time(NULL) <= when) return 0;
 
     /* Delete the key */
     server.stat_expiredkeys++;
-    server.dirty++;
-    return dbDelete(db,key);
-}
-
-int deleteIfVolatile(redisDb *db, robj *key) {
-    if (getExpire(db,key) < 0) return 0;
-
-    /* Delete the key */
-    server.stat_expiredkeys++;
-    server.dirty++;
+    propagateExpire(db,key);
     return dbDelete(db,key);
 }
 
@@ -458,7 +555,7 @@ int deleteIfVolatile(redisDb *db, robj *key) {
 
 void expireGenericCommand(redisClient *c, robj *key, robj *param, long offset) {
     dictEntry *de;
-    time_t seconds;
+    long seconds;
 
     if (getLongFromObjectOrReply(c, param, &seconds, NULL) != REDIS_OK) return;
 
@@ -472,15 +569,14 @@ void expireGenericCommand(redisClient *c, robj *key, robj *param, long offset) {
     if (seconds <= 0) {
         if (dbDelete(c->db,key)) server.dirty++;
         addReply(c, shared.cone);
+        signalModifiedKey(c->db,key);
         return;
     } else {
         time_t when = time(NULL)+seconds;
-        if (setExpire(c->db,key,when)) {
-            addReply(c,shared.cone);
-            server.dirty++;
-        } else {
-            addReply(c,shared.czero);
-        }
+        setExpire(c->db,key,when);
+        addReply(c,shared.cone);
+        signalModifiedKey(c->db,key);
+        server.dirty++;
         return;
     }
 }
@@ -494,15 +590,28 @@ void expireatCommand(redisClient *c) {
 }
 
 void ttlCommand(redisClient *c) {
-    time_t expire;
-    int ttl = -1;
+    time_t expire, ttl = -1;
 
     expire = getExpire(c->db,c->argv[1]);
     if (expire != -1) {
-        ttl = (int) (expire-time(NULL));
+        ttl = (expire-time(NULL));
         if (ttl < 0) ttl = -1;
     }
-    addReplySds(c,sdscatprintf(sdsempty(),":%d\r\n",ttl));
+    addReplyLongLong(c,(long long)ttl);
 }
 
+void persistCommand(redisClient *c) {
+    dictEntry *de;
 
+    de = dictFind(c->db->dict,c->argv[1]->ptr);
+    if (de == NULL) {
+        addReply(c,shared.czero);
+    } else {
+        if (removeExpire(c->db,c->argv[1])) {
+            addReply(c,shared.cone);
+            server.dirty++;
+        } else {
+            addReply(c,shared.czero);
+        }
+    }
+}