diskstore bug fixing and negative cache proper implementation

[redis.git] / src / db.c

diff --git a/src/db.c b/src/db.c
index 48663e9c97df48d96467f7fb30a5b4c8a843a905..ae40d2044fcd444ab077707c0c38fc7a39ebd9d9 100644 (file)
--- a/src/db.c
+++ b/src/db.c
@@ -17,12 +17,14 @@ robj *lookupKey(redisDb *db, robj *key) {
         if (server.bgsavechildpid == -1 && server.bgrewritechildpid == -1)
             val->lru = server.lruclock;
 
         if (server.bgsavechildpid == -1 && server.bgrewritechildpid == -1)
             val->lru = server.lruclock;
 

-        if (server.ds_enabled && val->storage == REDIS_DS_SAVING) {
-            /* FIXME: change this code to just wait for our object to
-             * get out of the IO Job. */
+        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();
             waitEmptyIOJobsQueue();
             processAllPendingIOJobs();

-            redisAssert(val->storage != REDIS_DS_SAVING);
+            redisAssert(!(cacheScheduleIOGetFlags(db,key) & REDIS_IO_SAVEINPROG));

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


@@ -32,12 +34,18 @@ robj *lookupKey(redisDb *db, robj *key) {
 
         /* 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
 
         /* 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.
-         * 
-         * FIXME: race condition here. If there was an already scheduled
-         * async loading of this key, what may happen is that the old
-         * key is loaded in memory if this gets deleted in the meantime. */
+         * in a blocking way. */

         if (server.ds_enabled && cacheKeyMayExist(db,key)) {
         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);
             val = dsGet(db,key,&expire);
             if (val) {
                 int retval = dbAdd(db,key,val);


@@ -85,9 +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) {
-            /* FIXME: remove entry from negative cache */
-        }
+        if (server.ds_enabled) cacheSetKeyMayExist(db,key);

         return REDIS_OK;
     }
 }
         return REDIS_OK;
     }
 }


@@ -98,16 +104,18 @@ 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) {
     robj *oldval;
  * On update (key already existed) 0 is returned. Otherwise 1. */
 int dbReplace(redisDb *db, robj *key, robj *val) {
     robj *oldval;

+    int retval;

 
     if ((oldval = dictFetchValue(db->dict,key->ptr)) == NULL) {
         sds copy = sdsdup(key->ptr);
         dictAdd(db->dict, copy, val);
 
     if ((oldval = dictFetchValue(db->dict,key->ptr)) == NULL) {
         sds copy = sdsdup(key->ptr);
         dictAdd(db->dict, copy, val);

-        return 1;
+        retval = 1;

     } else {
     } else {

-        val->storage = oldval->storage;

         dictReplace(db->dict, key->ptr, val);
         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) {


@@ -142,14 +150,13 @@ 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.ds_enabled) handleClientsBlockedOnSwappedKey(db,key);
-
-    /* FIXME: we need to delete the IO Job loading the key, or simply we can
-     * wait for it to finish. */
+    /* 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. */
 
     /* Deleting an entry from the expires dict will not free the sds of
      * the key, because it is shared with the main dictionary. */


@@ -189,13 +196,11 @@ int selectDb(redisClient *c, int id) {
 void signalModifiedKey(redisDb *db, robj *key) {
     touchWatchedKey(db,key);
     if (server.ds_enabled)
 void signalModifiedKey(redisDb *db, robj *key) {
     touchWatchedKey(db,key);
     if (server.ds_enabled)

-        cacheScheduleForFlush(db,key);
+        cacheScheduleIO(db,key,REDIS_IO_SAVE);

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

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

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


@@ -207,6 +212,7 @@ void flushdbCommand(redisClient *c) {
     signalFlushedDb(c->db->id);
     dictEmpty(c->db->dict);
     dictEmpty(c->db->expires);
     signalFlushedDb(c->db->id);
     dictEmpty(c->db->dict);
     dictEmpty(c->db->expires);

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

     addReply(c,shared.ok);
 }
 
     addReply(c,shared.ok);
 }
 


@@ -218,7 +224,10 @@ void flushallCommand(redisClient *c) {
         kill(server.bgsavechildpid,SIGKILL);
         rdbRemoveTempFile(server.bgsavechildpid);
     }
         kill(server.bgsavechildpid,SIGKILL);
         rdbRemoveTempFile(server.bgsavechildpid);
     }

-    rdbSave(server.dbfilename);
+    if (server.ds_enabled)
+        dsFlushDb(-1);
+    else
+        rdbSave(server.dbfilename);

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


@@ -239,7 +248,7 @@ void delCommand(redisClient *c) {
             if (cacheKeyMayExist(c->db,c->argv[j]) &&
                 dsExists(c->db,c->argv[j]))
             {
             if (cacheKeyMayExist(c->db,c->argv[j]) &&
                 dsExists(c->db,c->argv[j]))
             {

-                cacheScheduleForFlush(c->db,c->argv[j]);
+                cacheScheduleIO(c->db,c->argv[j],REDIS_IO_SAVE);

                 deleted = 1;
             }
         }
                 deleted = 1;
             }
         }