diskstore bug fixing and negative cache proper implementation

[redis.git] / src / db.c

diff --git a/src/db.c b/src/db.c
index c1ce79b56310949f08368dab7b836aa83f3299e7..ae40d2044fcd444ab077707c0c38fc7a39ebd9d9 100644 (file)
--- a/src/db.c
+++ b/src/db.c
@@ -11,31 +11,51 @@ robj *lookupKey(redisDb *db, robj *key) {
     if (de) {
         robj *val = dictGetEntryVal(de);
 
     if (de) {
         robj *val = dictGetEntryVal(de);
 

-        /* Update the access time for the aging algorithm. */
-        val->lru = server.lruclock;
-
-        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);
-            } else {
-                int notify = (val->storage == REDIS_VM_LOADING);
-
-                /* Our value was swapped on disk. Bring it at home. */
-                redisAssert(val->type == REDIS_VMPOINTER);
-                val = vmLoadObject(val);
-                dictGetEntryVal(de) = val;
-
-                /* Clients blocked by the VM subsystem may be waiting for
-                 * this key... */
-                if (notify) handleClientsBlockedOnSwappedKey(db,key);
-            }
+        /* 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)
+        {
+            /* 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 {
         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_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) == 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;
     }
 }
         return NULL;
     }
 }


@@ -73,6 +93,7 @@ int dbAdd(redisDb *db, robj *key, robj *val) {
     } else {
         sds copy = sdsdup(key->ptr);
         dictAdd(db->dict, copy, val);
     } else {
         sds copy = sdsdup(key->ptr);
         dictAdd(db->dict, copy, val);

+        if (server.ds_enabled) cacheSetKeyMayExist(db,key);

         return REDIS_OK;
     }
 }
         return REDIS_OK;
     }
 }


@@ -82,14 +103,19 @@ 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) {
  *
  * 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);
         sds copy = sdsdup(key->ptr);
         dictAdd(db->dict, copy, val);

-        return 1;
+        retval = 1;

     } else {
         dictReplace(db->dict, key->ptr, val);
     } 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) {
 }
 
 int dbExists(redisDb *db, robj *key) {


@@ -124,11 +150,14 @@ 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 VM is enabled make sure to awake waiting clients for this key:
-     * deleting the key will kill the I/O thread bringing the key from swap
-     * to memory, so the client will never be notified and unblocked if we
-     * don't do it now. */
-    if (server.vm_enabled) handleClientsBlockedOnSwappedKey(db,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);
     /* 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);


@@ -155,27 +184,50 @@ int selectDb(redisClient *c, int id) {
     return REDIS_OK;
 }
 
     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);
 /*-----------------------------------------------------------------------------
  * 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);
     dictEmpty(c->db->dict);
     dictEmpty(c->db->expires);

+    if (server.ds_enabled) dsFlushDb(c->db->id);

     addReply(c,shared.ok);
 }
 
 void flushallCommand(redisClient *c) {
     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);
     }
     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++;
 }
 
     server.dirty++;
 }
 


@@ -183,10 +235,22 @@ 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])) {
         if (dbDelete(c->db,c->argv[j])) {

-            touchWatchedKey(c->db,c->argv[j]);
+            signalModifiedKey(c->db,c->argv[j]);

             server.dirty++;
             deleted++;
             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);


@@ -330,8 +394,8 @@ void renameGenericCommand(redisClient *c, int nx) {
         dbReplace(c->db,c->argv[2],o);
     }
     dbDelete(c->db,c->argv[1]);
         dbReplace(c->db,c->argv[2],o);
     }
     dbDelete(c->db,c->argv[1]);

-    touchWatchedKey(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);
 }
     server.dirty++;
     addReply(c,nx ? shared.cone : shared.ok);
 }


@@ -430,16 +494,14 @@ time_t getExpire(redisDb *db, robj *key) {
  * will be consistent even if we allow write operations against expiring
  * keys. */
 void propagateExpire(redisDb *db, robj *key) {
  * will be consistent even if we allow write operations against expiring
  * keys. */
 void propagateExpire(redisDb *db, robj *key) {

-    struct redisCommand *cmd;

     robj *argv[2];
 
     robj *argv[2];
 

-    cmd = lookupCommand("del");

     argv[0] = createStringObject("DEL",3);
     argv[1] = key;
     incrRefCount(key);
 
     if (server.appendonly)
     argv[0] = createStringObject("DEL",3);
     argv[1] = key;
     incrRefCount(key);
 
     if (server.appendonly)

-        feedAppendOnlyFile(cmd,db->id,argv,2);
+        feedAppendOnlyFile(server.delCommand,db->id,argv,2);

     if (listLength(server.slaves))
         replicationFeedSlaves(server.slaves,db->id,argv,2);
 
     if (listLength(server.slaves))
         replicationFeedSlaves(server.slaves,db->id,argv,2);
 


@@ -468,7 +530,6 @@ int expireIfNeeded(redisDb *db, robj *key) {
 
     /* Delete the key */
     server.stat_expiredkeys++;
 
     /* Delete the key */
     server.stat_expiredkeys++;

-    server.dirty++;

     propagateExpire(db,key);
     return dbDelete(db,key);
 }
     propagateExpire(db,key);
     return dbDelete(db,key);
 }


@@ -493,13 +554,13 @@ void expireGenericCommand(redisClient *c, robj *key, robj *param, long offset) {
     if (seconds <= 0) {
         if (dbDelete(c->db,key)) server.dirty++;
         addReply(c, shared.cone);
     if (seconds <= 0) {
         if (dbDelete(c->db,key)) server.dirty++;
         addReply(c, shared.cone);

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

         return;
     } else {
         time_t when = time(NULL)+seconds;
         setExpire(c->db,key,when);
         addReply(c,shared.cone);
         return;
     } else {
         time_t when = time(NULL)+seconds;
         setExpire(c->db,key,when);
         addReply(c,shared.cone);

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

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