RocksDB Read Performance Improvements

Former-commit-id: 80cca4869888e048e10e11f1f20796c482c3e5b3

src/StorageCache.cpp

@@ -2,6 +2,7 @@
 
 void StorageCache::clear()
 {
+    std::unique_lock<fastlock> ul(m_lock);
     if (m_setkeys != nullptr)
         m_setkeys->clear();
     m_spstorage->clear();
@@ -24,6 +25,7 @@ void StorageCache::cacheKey(const char *rgch, size_t cch)
 bool StorageCache::erase(sds key)
 {
     bool result = m_spstorage->erase(key, sdslen(key));
+    std::unique_lock<fastlock> ul(m_lock);
     if (result && m_setkeys != nullptr)
     {
         auto itr = m_setkeys->find(sdsview(key));
@@ -35,15 +37,18 @@ bool StorageCache::erase(sds key)
 
 void StorageCache::insert(sds key, const void *data, size_t cbdata, bool fOverwrite)
 {
+    std::unique_lock<fastlock> ul(m_lock);
     if (!fOverwrite && m_setkeys != nullptr)
     {
         cacheKey(key);
     }
+    ul.unlock();
     m_spstorage->insert(key, sdslen(key), (void*)data, cbdata, fOverwrite);
 }
 
 const StorageCache *StorageCache::clone()
 {
+    std::unique_lock<fastlock> ul(m_lock);
     // Clones never clone the cache
     StorageCache *cacheNew = new StorageCache(const_cast<IStorage*>(m_spstorage->clone()));
     return cacheNew;
@@ -51,6 +56,7 @@ const StorageCache *StorageCache::clone()
 
 void StorageCache::retrieve(sds key, IStorage::callbackSingle fn, sds *cachedKey) const
 {
+    std::unique_lock<fastlock> ul(m_lock);
     if (m_setkeys != nullptr)
     {
         auto itr = m_setkeys->find(sdsview(key));
@@ -59,11 +65,13 @@ void StorageCache::retrieve(sds key, IStorage::callbackSingle fn, sds *cachedKey
         if (cachedKey != nullptr)
             *cachedKey = sdsdupshared(itr->get());
     }
+    ul.unlock();
     m_spstorage->retrieve(key, sdslen(key), fn);
 }
 
 size_t StorageCache::count() const
 {
+    std::unique_lock<fastlock> ul(m_lock);
     size_t count = m_spstorage->count();
     if (m_setkeys != nullptr)
         serverAssert(count == m_setkeys->size());

src/StorageCache.h

@@ -5,6 +5,7 @@ class StorageCache
 {
     std::shared_ptr<IStorage> m_spstorage;
     std::unique_ptr<semiorderedset<sdsimmutablestring, sdsview, true>> m_setkeys;
+    mutable fastlock m_lock {"StorageCache"};
 
     StorageCache(IStorage *storage)
         : m_spstorage(storage)

src/aof.cpp

@@ -972,8 +972,8 @@ int loadAppendOnlyFile(char *filename) {
 loaded_ok: /* DB loaded, cleanup and return C_OK to the caller. */
     for (int idb = 0; idb < cserver.dbnum; ++idb)
     {
-        g_pserver->db[idb]->processChanges(false);
+        if (g_pserver->db[idb]->processChanges(false))
-        g_pserver->db[idb]->commitChanges();
+            g_pserver->db[idb]->commitChanges();
     }
     fclose(fp);
     freeFakeClient(fakeClient);

src/config.cpp

@@ -2181,7 +2181,7 @@ static int updateMaxmemory(long long val, long long prev, const char **err) {
         if ((unsigned long long)val < used) {
             serverLog(LL_WARNING,"WARNING: the new maxmemory value set via CONFIG SET (%llu) is smaller than the current memory usage (%zu). This will result in key eviction and/or the inability to accept new write commands depending on the maxmemory-policy.", g_pserver->maxmemory, used);
         }
-        freeMemoryIfNeededAndSafe(false /*fPreSnapshot*/);
+        freeMemoryIfNeededAndSafe(false /*fQuickCycle*/, false /*fPreSnapshot*/);
     }
     return 1;
 }

src/db.cpp

@@ -2556,7 +2556,7 @@ void redisDbPersistentData::storeDatabase()
     sdsfree(temp);
 }
 
-void redisDbPersistentData::processChanges(bool fSnapshot)
+bool redisDbPersistentData::processChanges(bool fSnapshot)
 {
     serverAssert(GlobalLocksAcquired());
 
@@ -2565,6 +2565,8 @@ void redisDbPersistentData::processChanges(bool fSnapshot)
 
     if (m_spstorage != nullptr)
     {
+        if (!m_fAllChanged && m_setchanged.empty() && m_cnewKeysPending == 0)
+            return false;
         m_spstorage->beginWriteBatch();
         serverAssert(m_pdbSnapshotStorageFlush == nullptr);
         if (fSnapshot && !m_fAllChanged && m_setchanged.size() > 100)
@@ -2596,6 +2598,7 @@ void redisDbPersistentData::processChanges(bool fSnapshot)
         m_setchanged.clear();
         m_cnewKeysPending = 0;
     }
+    return (m_spstorage != nullptr);
 }
 
 void redisDbPersistentData::commitChanges(const redisDbPersistentDataSnapshot **psnapshotFree)
@@ -2698,8 +2701,8 @@ void redisDbPersistentData::removeAllCachedValues()
     // First we have to flush the tracked changes
     if (m_fTrackingChanges)
     {
-        processChanges(false);
+        if (processChanges(false))
-        commitChanges();
+            commitChanges();
         trackChanges(false);
     }
 
@@ -2812,3 +2815,80 @@ int dbnumFromDb(redisDb *db)
     }
     serverPanic("invalid database pointer");
 }
+
+void redisDbPersistentData::prefetchKeysAsync(AeLocker &lock, client *c)
+{
+    if (m_spstorage == nullptr)
+        return;
+
+    std::vector<robj*> veckeys;
+    lock.arm(c);
+    int numkeys = 0;
+    int *keys = getKeysFromCommand(c->cmd, c->argv, c->argc, &numkeys);
+    for (int ikey = 0; ikey < numkeys; ++ikey)
+    {
+        robj *objKey = c->argv[keys[ikey]];
+        if (this->find_cached_threadsafe(szFromObj(objKey)) == nullptr)
+            veckeys.push_back(objKey);
+    }
+    lock.disarm();
+
+    getKeysFreeResult(keys);
+
+    std::vector<std::tuple<sds, robj*, std::unique_ptr<expireEntry>>> vecInserts;
+    for (robj *objKey : veckeys)
+    {
+        sds sharedKey = nullptr;
+        std::unique_ptr<expireEntry> spexpire;
+        robj *o = nullptr;
+        m_spstorage->retrieve((sds)szFromObj(objKey), [&](const char *, size_t, const void *data, size_t cb){
+                size_t offset = 0;
+                spexpire = deserializeExpire((sds)szFromObj(objKey), (const char*)data, cb, &offset);    
+                o = deserializeStoredObject(this, szFromObj(objKey), reinterpret_cast<const char*>(data) + offset, cb - offset);
+                serverAssert(o != nullptr);
+        }, &sharedKey);
+
+        if (sharedKey == nullptr)
+            sharedKey = sdsdupshared(szFromObj(objKey));
+
+        vecInserts.emplace_back(sharedKey, o, std::move(spexpire));
+    }
+
+    lock.arm(c);
+    for (auto &tuple : vecInserts)
+    {
+        sds sharedKey = std::get<0>(tuple);
+        robj *o = std::get<1>(tuple);
+        std::unique_ptr<expireEntry> spexpire = std::move(std::get<2>(tuple));
+
+        if (o != nullptr)
+        {
+            if (this->find_cached_threadsafe(sharedKey) != nullptr)
+            {
+                // While unlocked this was already ensured
+                decrRefCount(o);
+                sdsfree(sharedKey);
+            }
+            else
+            {
+                dictAdd(m_pdict, sharedKey, o);
+                o->SetFExpires(spexpire != nullptr);
+
+                if (spexpire != nullptr)
+                {
+                    auto itr = m_setexpire->find(sharedKey);
+                    if (itr != m_setexpire->end())
+                        m_setexpire->erase(itr);
+                    m_setexpire->insert(std::move(*spexpire));
+                    serverAssert(m_setexpire->find(sharedKey) != m_setexpire->end());
+                }
+                serverAssert(o->FExpires() == (m_setexpire->find(sharedKey) != m_setexpire->end()));
+            }
+        }
+        else
+        {
+            if (sharedKey != nullptr)
+                sdsfree(sharedKey); // BUG but don't bother crashing
+        }
+    }
+}

src/evict.cpp

@@ -425,7 +425,7 @@ size_t freeMemoryGetNotCountedMemory(void) {
  *              limit.
  *              (Populated both for C_ERR and C_OK)
  */
-int getMaxmemoryState(size_t *total, size_t *logical, size_t *tofree, float *level, bool fPreSnapshot) {
+int getMaxmemoryState(size_t *total, size_t *logical, size_t *tofree, float *level, bool fQuickCycle, bool fPreSnapshot) {
     size_t mem_reported, mem_used, mem_tofree;
 
     /* Check if we are over the memory usage limit. If we are not, no need
@@ -464,7 +464,7 @@ int getMaxmemoryState(size_t *total, size_t *logical, size_t *tofree, float *lev
 
     /* Compute how much memory we need to free. */
     mem_tofree = mem_used - maxmemory;
-    if (g_pserver->m_pstorageFactory)
+    if (g_pserver->m_pstorageFactory && !fQuickCycle)
     {
         mem_tofree += static_cast<size_t>(maxmemory * 0.05); // if we have a storage provider be much more aggressive
     }
@@ -484,7 +484,7 @@ int getMaxmemoryState(size_t *total, size_t *logical, size_t *tofree, float *lev
  * were over the limit, but the attempt to free memory was successful.
  * Otehrwise if we are over the memory limit, but not enough memory
  * was freed to return back under the limit, the function returns C_ERR. */
-int freeMemoryIfNeeded(bool fPreSnapshot) {
+int freeMemoryIfNeeded(bool fQuickCycle, bool fPreSnapshot) {
     serverAssert(GlobalLocksAcquired());
 
     /* By default replicas should ignore maxmemory
@@ -498,12 +498,13 @@ int freeMemoryIfNeeded(bool fPreSnapshot) {
     const bool fEvictToStorage = !cserver.delete_on_evict && g_pserver->db[0]->FStorageProvider();
     int result = C_ERR;
     int ckeysFailed = 0;
+    int keys_freed = 0;
 
     /* When clients are paused the dataset should be static not just from the
      * POV of clients not being able to write, but also from the POV of
      * expires and evictions of keys not being performed. */
     if (clientsArePaused()) return C_OK;
-    if (getMaxmemoryState(&mem_reported,NULL,&mem_tofree,NULL,fPreSnapshot) == C_OK)
+    if (getMaxmemoryState(&mem_reported,NULL,&mem_tofree,NULL,fQuickCycle,fPreSnapshot) == C_OK)
         return C_OK;
 
     mem_freed = 0;
@@ -513,7 +514,7 @@ int freeMemoryIfNeeded(bool fPreSnapshot) {
         goto cant_free; /* We need to free memory, but policy forbids. */
 
     while (mem_freed < mem_tofree) {
-        int j, k, i, keys_freed = 0;
+        int j, k, i;
         static unsigned int next_db = 0;
         sds bestkey = NULL;
         int bestdbid;
@@ -737,7 +738,7 @@ cant_free:
  * - Nor we are loading data right now.
  *
  */
-int freeMemoryIfNeededAndSafe(bool fPreSnapshot) {
+int freeMemoryIfNeededAndSafe(bool fQuickCycle, bool fPreSnapshot) {
     if (g_pserver->shutdown_asap || g_pserver->lua_timedout || g_pserver->loading) return C_OK;
-    return freeMemoryIfNeeded(fPreSnapshot);
+    return freeMemoryIfNeeded(fQuickCycle, fPreSnapshot);
 }

src/rdb.cpp

@@ -2520,11 +2520,11 @@ int rdbLoadRio(rio *rdb, int rdbflags, rdbSaveInfo *rsi) {
                 {
                     for (int idb = 0; idb < cserver.dbnum; ++idb)
                     {
-                        g_pserver->db[idb]->processChanges(false);
+                        if (g_pserver->db[idb]->processChanges(false))
-                        g_pserver->db[idb]->commitChanges();
+                            g_pserver->db[idb]->commitChanges();
                         g_pserver->db[idb]->trackChanges(false);
                     }
-                    freeMemoryIfNeeded(false /* fPreSnapshot*/);
+                    freeMemoryIfNeeded(false /*fQuickCycle*/, false /* fPreSnapshot*/);
                 }
             }
             
@@ -2596,8 +2596,8 @@ int rdbLoadRio(rio *rdb, int rdbflags, rdbSaveInfo *rsi) {
 
     for (int idb = 0; idb < cserver.dbnum; ++idb)
     {
-        g_pserver->db[idb]->processChanges(false);
+        if (g_pserver->db[idb]->processChanges(false))
-        g_pserver->db[idb]->commitChanges();
+            g_pserver->db[idb]->commitChanges();
     }
     return C_OK;
 

src/scripting.cpp

@@ -1473,7 +1473,7 @@ void evalGenericCommand(client *c, int evalsha) {
     int delhook = 0, err;
 
     if (g_pserver->m_pstorageFactory != nullptr)
-        freeMemoryIfNeededAndSafe(true);
+        freeMemoryIfNeededAndSafe(false /*fQuickCycle*/, true);
 
     /* When we replicate whole scripts, we want the same PRNG sequence at
      * every call so that our PRNG is not affected by external state. */

src/server.cpp

@@ -2014,8 +2014,8 @@ void flushStorageWeak()
             latencyStartMonitor(storage_process_latency);
             std::vector<redisDb*> vecdb;
             for (int idb = 0; idb < cserver.dbnum; ++idb) {
-                vecdb.push_back(g_pserver->db[idb]);
+                if (g_pserver->db[idb]->processChanges(true))
-                g_pserver->db[idb]->processChanges(true);
+                    vecdb.push_back(g_pserver->db[idb]);
             }
             latencyEndMonitor(storage_process_latency);
             latencyAddSampleIfNeeded("storage-process-changes", storage_process_latency);
@@ -2453,8 +2453,8 @@ void beforeSleep(struct aeEventLoop *eventLoop) {
             mstime_t storage_process_latency;
             latencyStartMonitor(storage_process_latency);
             for (int idb = 0; idb < cserver.dbnum; ++idb) {
-                vecdb.push_back(g_pserver->db[idb]);
+                if (g_pserver->db[idb]->processChanges(false))
-                g_pserver->db[idb]->processChanges(false);
+                    vecdb.push_back(g_pserver->db[idb]);
             }
             latencyEndMonitor(storage_process_latency);
             latencyAddSampleIfNeeded("storage-process-changes", storage_process_latency);
@@ -3936,6 +3936,9 @@ int processCommand(client *c, int callFlags, AeLocker &locker) {
         return C_OK;
     }
 
+    if (!locker.isArmed())
+        c->db->prefetchKeysAsync(locker, c);
+
     /* Check if the user is authenticated. This check is skipped in case
      * the default user is flagged as "nopass" and is active. */
     int auth_required = (!(DefaultUser->flags & USER_FLAG_NOPASS) ||
@@ -4007,7 +4010,7 @@ int processCommand(client *c, int callFlags, AeLocker &locker) {
      * propagation of DELs due to eviction. */
     if (g_pserver->maxmemory && !g_pserver->lua_timedout) {
         locker.arm(c);
-        int out_of_memory = freeMemoryIfNeededAndSafe(false /*fPreSnapshot*/) == C_ERR;
+        int out_of_memory = freeMemoryIfNeededAndSafe(true /*fQuickCycle*/, false /*fPreSnapshot*/) == C_ERR;
         /* freeMemoryIfNeeded may flush replica output buffers. This may result
          * into a replica, that may be the active client, to be freed. */
         if (serverTL->current_client == NULL) return C_ERR;
@@ -4268,8 +4271,8 @@ int prepareForShutdown(int flags) {
 
         // Also Dump To FLASH if Applicable
         for (int idb = 0; idb < cserver.dbnum; ++idb) {
-            g_pserver->db[idb]->processChanges(false);
+            if (g_pserver->db[idb]->processChanges(false))
-            g_pserver->db[idb]->commitChanges();
+                g_pserver->db[idb]->commitChanges();
         }
     }
 

src/server.h

@@ -1329,7 +1329,7 @@ public:
     //  to allow you to release the global lock before commiting.  To prevent deadlocks you *must*
     //  either release the global lock or keep the same global lock between the two functions as
     //  a second look is kept to ensure writes to secondary storage are ordered
-    void processChanges(bool fSnapshot);
+    bool processChanges(bool fSnapshot);
     void commitChanges(const redisDbPersistentDataSnapshot **psnapshotFree = nullptr);
 
     // This should only be used if you look at the key, we do not fixup
@@ -1350,6 +1350,8 @@ public:
     bool removeCachedValue(const char *key);
     void removeAllCachedValues();
 
+    void prefetchKeysAsync(class AeLocker &locker, client *c);
+
     bool FSnapshot() const { return m_spdbSnapshotHOLDER != nullptr; }
 
     dict_iter find_cached_threadsafe(const char *key) const;
@@ -1462,7 +1464,7 @@ struct redisDb : public redisDbPersistentDataSnapshot
     friend void setExpire(client *c, redisDb *db, robj *key, expireEntry &&e);
     friend int evictionPoolPopulate(int dbid, redisDb *db, expireset *setexpire, struct evictionPoolEntry *pool);
     friend void activeDefragCycle(void);
-    friend int freeMemoryIfNeeded(bool);
+    friend int freeMemoryIfNeeded(bool, bool);
     friend void activeExpireCycle(int);
     friend void expireSlaveKeys(void);
 
@@ -1518,6 +1520,7 @@ struct redisDb : public redisDbPersistentDataSnapshot
     using redisDbPersistentData::removeAllCachedValues;
     using redisDbPersistentData::dictUnsafeKeyOnly;
     using redisDbPersistentData::resortExpire;
+    using redisDbPersistentData::prefetchKeysAsync;
 
 public:
     expireset::setiter expireitr;
@@ -2997,10 +3000,10 @@ int zslLexValueGteMin(sds value, zlexrangespec *spec);
 int zslLexValueLteMax(sds value, zlexrangespec *spec);
 
 /* Core functions */
-int getMaxmemoryState(size_t *total, size_t *logical, size_t *tofree, float *level, bool fPreSnapshot=false);
+int getMaxmemoryState(size_t *total, size_t *logical, size_t *tofree, float *level, bool fQuickCycle = false, bool fPreSnapshot=false);
 size_t freeMemoryGetNotCountedMemory();
-int freeMemoryIfNeeded(bool fPreSnapshot);
+int freeMemoryIfNeeded(bool fQuickCycle, bool fPreSnapshot);
-int freeMemoryIfNeededAndSafe(bool fPreSnapshot);
+int freeMemoryIfNeededAndSafe(bool fQuickCycle, bool fPreSnapshot);
 int processCommand(client *c, int callFlags, class AeLocker &locker);
 void setupSignalHandlers(void);
 struct redisCommand *lookupCommand(sds name);

src/snapshot.cpp

@@ -8,7 +8,7 @@ const redisDbPersistentDataSnapshot *redisDbPersistentData::createSnapshot(uint6
     serverAssert(GlobalLocksAcquired());
     serverAssert(m_refCount == 0);  // do not call this on a snapshot
 
-    if (freeMemoryIfNeededAndSafe(true /*fPreSnapshot*/) != C_OK && fOptional)
+    if (freeMemoryIfNeededAndSafe(false /*fQuickCycle*/, true /*fPreSnapshot*/) != C_OK && fOptional)
         return nullptr; // can't create snapshot due to OOM
 
     int levels = 1;
@@ -369,7 +369,7 @@ void redisDbPersistentData::endSnapshot(const redisDbPersistentDataSnapshot *psn
     latencyEndMonitor(latency_endsnapshot);
     latencyAddSampleIfNeeded("end-mvcc-snapshot", latency_endsnapshot);
 
-    freeMemoryIfNeededAndSafe(false);
+    freeMemoryIfNeededAndSafe(false /*fQuickCycle*/, false);
 }
 
 dict_iter redisDbPersistentDataSnapshot::random_cache_threadsafe(bool fPrimaryOnly) const