Merge pull request #21335 from taosdata/feat/TD-22746

feat: support client meta control

include/common/tglobal.h

@@ -124,6 +124,7 @@ extern int32_t tsRedirectFactor;
 extern int32_t tsRedirectMaxPeriod;
 extern int32_t tsMaxRetryWaitTime;
 extern bool    tsUseAdapter;
+extern int32_t tsMetaCacheMaxSize;
 extern int32_t tsSlowLogThreshold;
 extern int32_t tsSlowLogScope;
 
@@ -193,7 +194,7 @@ struct SConfig *taosGetCfg();
 
 void    taosSetAllDebugFlag(int32_t flag, bool rewrite);
 void    taosSetDebugFlag(int32_t *pFlagPtr, const char *flagName, int32_t flagVal, bool rewrite);
-int32_t taosSetCfg(SConfig *pCfg, char *name);
+int32_t taosApplyLocalCfg(SConfig *pCfg, char *name);
 void    taosLocalCfgForbiddenToChange(char *name, bool *forbidden);
 
 #ifdef __cplusplus

include/common/tmsg.h

@@ -3191,7 +3191,8 @@ typedef struct {
   char     dbFName[TSDB_DB_FNAME_LEN];
   uint64_t suid;
   int32_t  version;
-  SArray*  pIndex;
+  int32_t  indexSize;
+  SArray*  pIndex;  // STableIndexInfo
 } STableIndexRsp;
 
 int32_t tSerializeSTableIndexRsp(void* buf, int32_t bufLen, const STableIndexRsp* pRsp);

include/libs/catalog/catalog.h

@@ -214,7 +214,7 @@ int32_t catalogGetSTableMeta(SCatalog* pCatalog, SRequestConnInfo* pConn, const
 
 int32_t catalogUpdateTableMeta(SCatalog* pCatalog, STableMetaRsp* rspMsg);
 
-int32_t catalogUpdateTableMeta(SCatalog* pCatalog, STableMetaRsp* rspMsg);
+int32_t catalogAsyncUpdateTableMeta(SCatalog* pCtg, STableMetaRsp* pMsg);
 
 int32_t catalogGetCachedTableMeta(SCatalog* pCtg, const SName* pTableName, STableMeta** pTableMeta);
 

include/libs/qcom/query.h

@@ -90,28 +90,23 @@ typedef struct STbVerInfo {
   int32_t tversion;
 } STbVerInfo;
 
-/*
- * ASSERT(sizeof(SCTableMeta) == 24)
- * ASSERT(tableType == TSDB_CHILD_TABLE)
- * The cached child table meta info. For each child table, 24 bytes are required to keep the essential table info.
- */
+#pragma pack(push, 1) 
 typedef struct SCTableMeta {
-  int32_t  vgId : 24;
-  int8_t   tableType;
   uint64_t uid;
   uint64_t suid;
+  int32_t  vgId;
+  int8_t   tableType;
 } SCTableMeta;
+#pragma pack(pop)
 
-/*
- * Note that the first 24 bytes of STableMeta are identical to SCTableMeta, it is safe to cast a STableMeta to be a
- * SCTableMeta.
- */
+
+#pragma pack(push, 1) 
 typedef struct STableMeta {
   // BEGIN: KEEP THIS PART SAME WITH SCTableMeta
-  int32_t  vgId : 24;
-  int8_t   tableType;
   uint64_t uid;
   uint64_t suid;
+  int32_t  vgId;
+  int8_t   tableType;
   // END: KEEP THIS PART SAME WITH SCTableMeta
 
   // if the table is TSDB_CHILD_TABLE, the following information is acquired from the corresponding super table meta
@@ -121,6 +116,7 @@ typedef struct STableMeta {
   STableComInfo tableInfo;
   SSchema       schema[];
 } STableMeta;
+#pragma pack(pop)
 
 typedef struct SDBVgInfo {
   int32_t   vgVersion;
@@ -130,7 +126,7 @@ typedef struct SDBVgInfo {
   int32_t   numOfTable;  // DB's table num, unit is TSDB_TABLE_NUM_UNIT
   int64_t   stateTs;
   SHashObj* vgHash;  // key:vgId, value:SVgroupInfo
-  SArray*   vgArray;
+  SArray*   vgArray; // SVgroupInfo
 } SDBVgInfo;
 
 typedef struct SUseDbOutput {

source/client/src/clientEnv.c

@@ -656,7 +656,7 @@ int taos_options_imp(TSDB_OPTION option, const char *str) {
   } else {
     tscInfo("set cfg:%s to %s", pItem->name, str);
     if (TSDB_OPTION_SHELL_ACTIVITY_TIMER == option || TSDB_OPTION_USE_ADAPTER == option) {
-      code = taosSetCfg(pCfg, pItem->name);
+      code = taosApplyLocalCfg(pCfg, pItem->name);
     }
   }
 

source/client/src/clientHb.c

@@ -224,7 +224,7 @@ static int32_t hbProcessStbInfoRsp(void *value, int32_t valueLen, struct SCatalo
         return TSDB_CODE_TSC_INVALID_VALUE;
       }
 
-      catalogUpdateTableMeta(pCatalog, rsp);
+      catalogAsyncUpdateTableMeta(pCatalog, rsp);
     }
   }
 

source/client/src/clientImpl.c

@@ -815,7 +815,7 @@ int32_t handleAlterTbExecRes(void* res, SCatalog* pCatalog) {
 }
 
 int32_t handleCreateTbExecRes(void* res, SCatalog* pCatalog) {
-  return catalogUpdateTableMeta(pCatalog, (STableMetaRsp*)res);
+  return catalogAsyncUpdateTableMeta(pCatalog, (STableMetaRsp*)res);
 }
 
 int32_t handleQueryExecRsp(SRequestObj* pRequest) {

source/common/src/tglobal.c

@@ -117,12 +117,10 @@ int32_t tsRedirectFactor = 2;
 int32_t tsRedirectMaxPeriod = 1000;
 int32_t tsMaxRetryWaitTime = 10000;
 bool    tsUseAdapter = false;
+int32_t tsMetaCacheMaxSize = -1; // MB
 int32_t tsSlowLogThreshold = 3; // seconds
 int32_t tsSlowLogScope = SLOW_LOG_TYPE_ALL;
 
-
-
-
 /*
  * denote if the server needs to compress response message at the application layer to client, including query rsp,
  * metricmeta rsp, and multi-meter query rsp message body. The client compress the submit message to server.
@@ -351,6 +349,7 @@ static int32_t taosAddClientCfg(SConfig *pCfg) {
   if (cfgAddBool(pCfg, "useAdapter", tsUseAdapter, true) != 0) return -1;
   if (cfgAddBool(pCfg, "crashReporting", tsEnableCrashReport, true) != 0) return -1;
   if (cfgAddInt64(pCfg, "queryMaxConcurrentTables", tsQueryMaxConcurrentTables, INT64_MIN, INT64_MAX, 1) != 0) return -1;
+  if (cfgAddInt32(pCfg, "metaCacheMaxSize", tsMetaCacheMaxSize, -1, INT32_MAX, 1) != 0) return -1;
   if (cfgAddInt32(pCfg, "slowLogThreshold", tsSlowLogThreshold, 0, INT32_MAX, true) != 0) return -1;
   if (cfgAddString(pCfg, "slowLogScope", "", true) != 0) return -1;
 
@@ -788,6 +787,7 @@ static int32_t taosSetClientCfg(SConfig *pCfg) {
   tsUseAdapter = cfgGetItem(pCfg, "useAdapter")->bval;
   tsEnableCrashReport = cfgGetItem(pCfg, "crashReporting")->bval;
   tsQueryMaxConcurrentTables = cfgGetItem(pCfg, "queryMaxConcurrentTables")->i64;
+  tsMetaCacheMaxSize = cfgGetItem(pCfg, "metaCacheMaxSize")->i32;
   tsSlowLogThreshold = cfgGetItem(pCfg, "slowLogThreshold")->i32;
   if (taosSetSlowLogScope(cfgGetItem(pCfg, "slowLogScope")->str)) {
     return -1;
@@ -916,7 +916,7 @@ void taosLocalCfgForbiddenToChange(char *name, bool *forbidden) {
   *forbidden = false;
 }
 
-int32_t taosSetCfg(SConfig *pCfg, char *name) {
+int32_t taosApplyLocalCfg(SConfig *pCfg, char *name) {
   int32_t len = strlen(name);
   char    lowcaseName[CFG_NAME_MAX_LEN + 1] = {0};
   strntolower(lowcaseName, name, TMIN(CFG_NAME_MAX_LEN, len));
@@ -1051,6 +1051,12 @@ int32_t taosSetCfg(SConfig *pCfg, char *name) {
           }
           break;
         }
+        case 'e': {
+          if (strcasecmp("metaCacheMaxSize", name) == 0) {
+            atomic_store_32(&tsMetaCacheMaxSize, cfgGetItem(pCfg, "metaCacheMaxSize")->i32);
+          }
+          break;
+        }
         case 'i': {
           if (strcasecmp("minimalTmpDirGB", name) == 0) {
             tsTempSpace.reserved = (int64_t)(((double)cfgGetItem(pCfg, "minimalTmpDirGB")->fval) * 1024 * 1024 * 1024);

source/common/src/tmsg.c

@@ -1566,21 +1566,21 @@ int32_t tDeserializeSGetUserAuthRspImpl(SDecoder *pDecoder, SGetUserAuthRsp *pRs
     char db[TSDB_DB_FNAME_LEN] = {0};
     if (tDecodeCStrTo(pDecoder, db) < 0) return -1;
     int32_t len = strlen(db);
-    taosHashPut(pRsp->createdDbs, db, len, db, len);
+    taosHashPut(pRsp->createdDbs, db, len, db, len + 1);
   }
 
   for (int32_t i = 0; i < numOfReadDbs; ++i) {
     char db[TSDB_DB_FNAME_LEN] = {0};
     if (tDecodeCStrTo(pDecoder, db) < 0) return -1;
     int32_t len = strlen(db);
-    taosHashPut(pRsp->readDbs, db, len, db, len);
+    taosHashPut(pRsp->readDbs, db, len, db, len + 1);
   }
 
   for (int32_t i = 0; i < numOfWriteDbs; ++i) {
     char db[TSDB_DB_FNAME_LEN] = {0};
     if (tDecodeCStrTo(pDecoder, db) < 0) return -1;
     int32_t len = strlen(db);
-    taosHashPut(pRsp->writeDbs, db, len, db, len);
+    taosHashPut(pRsp->writeDbs, db, len, db, len + 1);
   }
 
   if (!tDecodeIsEnd(pDecoder)) {
@@ -3416,6 +3416,7 @@ int32_t tSerializeSTableIndexRsp(void *buf, int32_t bufLen, const STableIndexRsp
   if (tEncodeCStr(&encoder, pRsp->dbFName) < 0) return -1;
   if (tEncodeU64(&encoder, pRsp->suid) < 0) return -1;
   if (tEncodeI32(&encoder, pRsp->version) < 0) return -1;
+  if (tEncodeI32(&encoder, pRsp->indexSize) < 0) return -1;
   int32_t num = taosArrayGetSize(pRsp->pIndex);
   if (tEncodeI32(&encoder, num) < 0) return -1;
   if (num > 0) {
@@ -3461,6 +3462,7 @@ int32_t tDeserializeSTableIndexRsp(void *buf, int32_t bufLen, STableIndexRsp *pR
   if (tDecodeCStrTo(&decoder, pRsp->dbFName) < 0) return -1;
   if (tDecodeU64(&decoder, &pRsp->suid) < 0) return -1;
   if (tDecodeI32(&decoder, &pRsp->version) < 0) return -1;
+  if (tDecodeI32(&decoder, &pRsp->indexSize) < 0) return -1;
   int32_t num = 0;
   if (tDecodeI32(&decoder, &num) < 0) return -1;
   if (num > 0) {
@@ -3735,6 +3737,7 @@ int32_t tSerializeSSTbHbRsp(void *buf, int32_t bufLen, SSTbHbRsp *pRsp) {
     if (tEncodeCStr(&encoder, pIndexRsp->dbFName) < 0) return -1;
     if (tEncodeU64(&encoder, pIndexRsp->suid) < 0) return -1;
     if (tEncodeI32(&encoder, pIndexRsp->version) < 0) return -1;
+    if (tEncodeI32(&encoder, pIndexRsp->indexSize) < 0) return -1;
     int32_t num = taosArrayGetSize(pIndexRsp->pIndex);
     if (tEncodeI32(&encoder, num) < 0) return -1;
     for (int32_t i = 0; i < num; ++i) {
@@ -3797,6 +3800,7 @@ int32_t tDeserializeSSTbHbRsp(void *buf, int32_t bufLen, SSTbHbRsp *pRsp) {
     if (tDecodeCStrTo(&decoder, tableIndexRsp.dbFName) < 0) return -1;
     if (tDecodeU64(&decoder, &tableIndexRsp.suid) < 0) return -1;
     if (tDecodeI32(&decoder, &tableIndexRsp.version) < 0) return -1;
+    if (tDecodeI32(&decoder, &tableIndexRsp.indexSize) < 0) return -1;
     int32_t num = 0;
     if (tDecodeI32(&decoder, &num) < 0) return -1;
     if (num > 0) {

source/dnode/mnode/impl/src/mndSma.c

@@ -1114,6 +1114,7 @@ int32_t mndGetTableSma(SMnode *pMnode, char *tbFName, STableIndexRsp *rsp, bool
       return code;
     }
 
+    rsp->indexSize += sizeof(info) + pSma->exprLen + 1;
     *exist = true;
 
     sdbRelease(pSdb, pSma);

source/libs/catalog/inc/catalogInt.h

@@ -23,6 +23,8 @@ extern "C" {
 #include "catalog.h"
 #include "query.h"
 #include "tcommon.h"
+#include "ttimer.h"
+#include "tglobal.h"
 
 #define CTG_DEFAULT_CACHE_CLUSTER_NUMBER 6
 #define CTG_DEFAULT_CACHE_VGROUP_NUMBER  100
@@ -34,6 +36,8 @@ extern "C" {
 #define CTG_DEFAULT_BATCH_NUM            64
 #define CTG_DEFAULT_FETCH_NUM            8
 #define CTG_MAX_COMMAND_LEN              512
+#define CTG_DEFAULT_CACHE_MON_MSEC       5000
+#define CTG_CLEAR_CACHE_ROUND_TB_NUM     3000  
 
 #define CTG_RENT_SLOT_SECOND 1.5
 
@@ -131,6 +135,7 @@ typedef struct SCtgDebug {
 
 typedef struct SCtgCacheStat {
   uint64_t cacheNum[CTG_CI_MAX_VALUE];
+  uint64_t cacheSize[CTG_CI_MAX_VALUE];
   uint64_t cacheHit[CTG_CI_MAX_VALUE];
   uint64_t cacheNHit[CTG_CI_MAX_VALUE];
 } SCtgCacheStat;
@@ -239,8 +244,8 @@ typedef STableIndexRsp STableIndex;
 
 typedef struct SCtgTbCache {
   SRWLatch     metaLock;
-  STableMeta*  pMeta;
   SRWLatch     indexLock;
+  STableMeta*  pMeta;
   STableIndex* pIndex;
 } SCtgTbCache;
 
@@ -263,6 +268,7 @@ typedef struct SCtgDBCache {
   SHashObj*    tbCache;   // key:tbname, value:SCtgTbCache
   SHashObj*    stbCache;  // key:suid, value:char*
   uint64_t     dbCacheNum[CTG_CI_MAX_VALUE];
+  uint64_t     dbCacheSize;
 } SCtgDBCache;
 
 typedef struct SCtgRentSlot {
@@ -276,12 +282,15 @@ typedef struct SCtgRentMgmt {
   uint16_t      slotNum;
   uint16_t      slotRIdx;
   int64_t       lastReadMsec;
+  uint64_t      rentCacheSize;
+  int32_t       metaSize;
   SCtgRentSlot* slots;
 } SCtgRentMgmt;
 
 typedef struct SCtgUserAuth {
   SRWLatch        lock;
   SGetUserAuthRsp userAuth;
+  uint64_t        userCacheSize;
 } SCtgUserAuth;
 
 typedef struct SCatalog {
@@ -412,6 +421,7 @@ typedef struct SCtgRuntimeStat {
   uint64_t numOfOpAbort;
   uint64_t numOfOpEnqueue;
   uint64_t numOfOpDequeue;
+  uint64_t numOfOpClearMeta;
   uint64_t numOfOpClearCache;
 } SCtgRuntimeStat;
 
@@ -488,6 +498,7 @@ typedef struct SCtgDropTbIndexMsg {
 
 typedef struct SCtgClearCacheMsg {
   SCatalog* pCtg;
+  bool      clearMeta;
   bool      freeCtg;
 } SCtgClearCacheMsg;
 
@@ -526,6 +537,8 @@ typedef struct SCatalogMgmt {
   int32_t      jobPool;
   SRWLatch     lock;
   SCtgQueue    queue;
+  void        *timer;
+  tmr_h        cacheTimer;
   TdThread     updateThread;
   SHashObj*    pCluster;  // key: clusterId, value: SCatalog*
   SCatalogStat statInfo;
@@ -542,8 +555,8 @@ typedef struct SCtgOperation {
 } SCtgOperation;
 
 typedef struct SCtgCacheItemInfo {
-  char*   name;
-  int32_t flag;
+  char*            name;
+  int32_t          flag;
 } SCtgCacheItemInfo;
 
 #define CTG_AUTH_READ(_t)  ((_t) == AUTH_TYPE_READ || (_t) == AUTH_TYPE_READ_OR_WRITE)
@@ -556,11 +569,6 @@ typedef struct SCtgCacheItemInfo {
 #define CTG_STAT_DEC(_item, _n) atomic_sub_fetch_64(&(_item), _n)
 #define CTG_STAT_GET(_item)     atomic_load_64(&(_item))
 
-#define CTG_DB_NUM_INC(_item)   dbCache->dbCacheNum[_item] += 1
-#define CTG_DB_NUM_DEC(_item)   dbCache->dbCacheNum[_item] -= 1
-#define CTG_DB_NUM_SET(_item)   dbCache->dbCacheNum[_item] = 1
-#define CTG_DB_NUM_RESET(_item) dbCache->dbCacheNum[_item] = 0
-
 #define CTG_STAT_API_INC(item, n)  (CTG_STAT_INC(gCtgMgmt.statInfo.api.item, n))
 #define CTG_STAT_RT_INC(item, n)   (CTG_STAT_INC(gCtgMgmt.statInfo.runtime.item, n))
 #define CTG_STAT_NUM_INC(item, n)  (CTG_STAT_INC(gCtgMgmt.statInfo.cache.cacheNum[item], n))
@@ -575,6 +583,11 @@ typedef struct SCtgCacheItemInfo {
 #define CTG_CACHE_HIT_INC(item, n)  (CTG_STAT_INC(pCtg->cacheStat.cacheHit[item], n))
 #define CTG_CACHE_NHIT_INC(item, n) (CTG_STAT_INC(pCtg->cacheStat.cacheNHit[item], n))
 
+#define CTG_DB_NUM_INC(_item)   dbCache->dbCacheNum[_item] += 1
+#define CTG_DB_NUM_DEC(_item)   dbCache->dbCacheNum[_item] -= 1
+#define CTG_DB_NUM_SET(_item)   dbCache->dbCacheNum[_item] = 1
+#define CTG_DB_NUM_RESET(_item) dbCache->dbCacheNum[_item] = 0
+
 #define CTG_META_NUM_INC(type)                  \
   do {                                          \
     switch (type) {                             \
@@ -685,6 +698,10 @@ typedef struct SCtgCacheItemInfo {
 #define CTG_DB_NOT_EXIST(code) \
   (code == TSDB_CODE_MND_DB_NOT_EXIST || code == TSDB_CODE_MND_DB_IN_CREATING || code == TSDB_CODE_MND_DB_IN_DROPPING)
 
+#define CTG_CACHE_OVERFLOW(_csize, _maxsize) ((_maxsize >= 0) ? ((_csize) >= (_maxsize) * 1048576L * 0.9) : false)
+#define CTG_CACHE_LOW(_csize, _maxsize) ((_maxsize >= 0) ? ((_csize) <= (_maxsize) * 1048576L * 0.75) : true)
+
+
 #define ctgFatal(param, ...) qFatal("CTG:%p " param, pCtg, __VA_ARGS__)
 #define ctgError(param, ...) qError("CTG:%p " param, pCtg, __VA_ARGS__)
 #define ctgWarn(param, ...)  qWarn("CTG:%p " param, pCtg, __VA_ARGS__)
@@ -787,6 +804,12 @@ typedef struct SCtgCacheItemInfo {
     CTG_RET(__code);                                 \
   } while (0)
 
+#define CTG_API_NLEAVE()                             \
+  do {                                               \
+    CTG_UNLOCK(CTG_READ, &gCtgMgmt.lock);            \
+    CTG_API_DEBUG("CTG API leave %s", __FUNCTION__); \
+  } while (0)  
+
 #define CTG_API_ENTER()                              \
   do {                                               \
     CTG_API_DEBUG("CTG API enter %s", __FUNCTION__); \
@@ -796,6 +819,15 @@ typedef struct SCtgCacheItemInfo {
     }                                                \
   } while (0)
 
+#define CTG_API_NENTER()                             \
+  do {                                               \
+    CTG_API_DEBUG("CTG API enter %s", __FUNCTION__); \
+    CTG_LOCK(CTG_READ, &gCtgMgmt.lock);              \
+    if (atomic_load_8((int8_t*)&gCtgMgmt.exit)) {    \
+      CTG_API_NLEAVE();                              \
+    }                                                \
+  } while (0)  
+
 #define CTG_API_JENTER()                             \
   do {                                               \
     CTG_API_DEBUG("CTG API enter %s", __FUNCTION__); \
@@ -859,8 +891,8 @@ int32_t ctgUpdateTbMetaEnqueue(SCatalog* pCtg, STableMetaOutput* output, bool sy
 int32_t ctgUpdateUserEnqueue(SCatalog* pCtg, SGetUserAuthRsp* pAuth, bool syncReq);
 int32_t ctgUpdateVgEpsetEnqueue(SCatalog* pCtg, char* dbFName, int32_t vgId, SEpSet* pEpSet);
 int32_t ctgUpdateTbIndexEnqueue(SCatalog* pCtg, STableIndex** pIndex, bool syncOp);
-int32_t ctgClearCacheEnqueue(SCatalog* pCtg, bool freeCtg, bool stopQueue, bool syncOp);
-int32_t ctgMetaRentInit(SCtgRentMgmt* mgmt, uint32_t rentSec, int8_t type);
+int32_t ctgClearCacheEnqueue(SCatalog* pCtg, bool clearMeta, bool freeCtg, bool stopQueue, bool syncOp);
+int32_t ctgMetaRentInit(SCtgRentMgmt* mgmt, uint32_t rentSec, int8_t type, int32_t size);
 int32_t ctgMetaRentAdd(SCtgRentMgmt* mgmt, void* meta, int64_t id, int32_t size);
 int32_t ctgMetaRentGet(SCtgRentMgmt* mgmt, void** res, uint32_t* num, int32_t size);
 int32_t ctgUpdateTbMetaToCache(SCatalog* pCtg, STableMetaOutput* pOut, bool syncReq);
@@ -941,7 +973,7 @@ void    ctgFreeSTableIndex(void* info);
 void    ctgClearSubTaskRes(SCtgSubRes* pRes);
 void    ctgFreeQNode(SCtgQNode* node);
 void    ctgClearHandle(SCatalog* pCtg);
-void    ctgFreeTbCacheImpl(SCtgTbCache* pCache);
+void    ctgFreeTbCacheImpl(SCtgTbCache* pCache, bool lock);
 int32_t ctgRemoveTbMeta(SCatalog* pCtg, SName* pTableName);
 int32_t ctgGetTbHashVgroup(SCatalog* pCtg, SRequestConnInfo* pConn, const SName* pTableName, SVgroupInfo* pVgroup,
                            bool* exists);
@@ -960,6 +992,16 @@ void    ctgReleaseVgMetaToCache(SCatalog* pCtg, SCtgDBCache* dbCache, SCtgTbCach
 void    ctgReleaseTbMetaToCache(SCatalog* pCtg, SCtgDBCache* dbCache, SCtgTbCache* pCache);
 void    ctgGetGlobalCacheStat(SCtgCacheStat* pStat);
 int32_t ctgChkSetAuthRes(SCatalog* pCtg, SCtgAuthReq* req, SCtgAuthRsp* res);
+void    ctgGetGlobalCacheSize(uint64_t *pSize);
+uint64_t ctgGetTbIndexCacheSize(STableIndex *pIndex);
+uint64_t ctgGetTbMetaCacheSize(STableMeta *pMeta);
+uint64_t ctgGetDbVgroupCacheSize(SDBVgInfo *pVg);
+uint64_t ctgGetUserCacheSize(SGetUserAuthRsp *pAuth);
+uint64_t ctgGetClusterCacheSize(SCatalog *pCtg);
+void     ctgClearHandleMeta(SCatalog* pCtg, int64_t *pClearedSize, int64_t *pCleardNum, bool *roundDone);
+void     ctgClearAllHandleMeta(int64_t *clearedSize, int64_t *clearedNum, bool *roundDone);
+void     ctgProcessTimerEvent(void *param, void *tmrId);
+
 int32_t ctgGetTbMeta(SCatalog* pCtg, SRequestConnInfo* pConn, SCtgTbMetaCtx* ctx, STableMeta** pTableMeta);
 int32_t ctgGetCachedStbNameFromSuid(SCatalog* pCtg, char* dbFName, uint64_t suid, char **stbName);
 int32_t ctgGetTbTagCb(SCtgTask* pTask);

source/libs/catalog/src/catalog.c

@@ -668,6 +668,36 @@ _return:
   CTG_RET(code);
 }
 
+void ctgProcessTimerEvent(void *param, void *tmrId) {
+  CTG_API_NENTER();
+
+  int32_t cacheMaxSize = atomic_load_32(&tsMetaCacheMaxSize);
+  if (cacheMaxSize >= 0) {
+    uint64_t cacheSize = 0;
+    ctgGetGlobalCacheSize(&cacheSize);
+    bool overflow = CTG_CACHE_OVERFLOW(cacheSize, cacheMaxSize);
+    
+    qDebug("catalog cache size: %" PRIu64"B, maxCaseSize:%dMB, %s", cacheSize, cacheMaxSize, overflow ? "overflow" : "NO overflow");
+
+    if (overflow) {
+      int32_t code = ctgClearCacheEnqueue(NULL, true, false, false, false);
+      if (code) {
+        qError("clear cache enqueue failed, error:%s", tstrerror(code));
+        taosTmrReset(ctgProcessTimerEvent, CTG_DEFAULT_CACHE_MON_MSEC, NULL, gCtgMgmt.timer, &gCtgMgmt.cacheTimer);
+      }
+
+      goto _return;
+    }
+  }
+
+  qTrace("reset catalog timer");
+  taosTmrReset(ctgProcessTimerEvent, CTG_DEFAULT_CACHE_MON_MSEC, NULL, gCtgMgmt.timer, &gCtgMgmt.cacheTimer);
+
+_return:
+
+  CTG_API_NLEAVE();
+}
+
 int32_t ctgGetDBCfg(SCatalog* pCtg, SRequestConnInfo* pConn, const char* dbFName, SDbCfgInfo* pDbCfg) {
   CTG_ERR_RET(ctgReadDBCfgFromCache(pCtg, dbFName, pDbCfg));
 
@@ -686,6 +716,7 @@ int32_t ctgGetDBCfg(SCatalog* pCtg, SRequestConnInfo* pConn, const char* dbFName
 
 
 int32_t catalogInit(SCatalogCfg* cfg) {
+  qDebug("catalogInit start");
   if (gCtgMgmt.pCluster) {
     qError("catalog already initialized");
     CTG_ERR_RET(TSDB_CODE_CTG_INVALID_INPUT);
@@ -743,6 +774,18 @@ int32_t catalogInit(SCatalogCfg* cfg) {
     CTG_ERR_RET(terrno);
   }
 
+  gCtgMgmt.timer = taosTmrInit(0, 0, 0, "catalog");
+  if (NULL == gCtgMgmt.timer) {
+    qError("init timer failed, error:%s", tstrerror(terrno));
+    CTG_ERR_RET(TSDB_CODE_OUT_OF_MEMORY);
+  }
+
+  gCtgMgmt.cacheTimer = taosTmrStart(ctgProcessTimerEvent, CTG_DEFAULT_CACHE_MON_MSEC, NULL, gCtgMgmt.timer);
+  if (NULL == gCtgMgmt.cacheTimer) {
+    qError("start cache timer failed");
+    CTG_ERR_RET(TSDB_CODE_OUT_OF_MEMORY);
+  }
+
   CTG_ERR_RET(ctgStartUpdateThread());
 
   qDebug("catalog initialized, maxDb:%u, maxTbl:%u, dbRentSec:%u, stbRentSec:%u", gCtgMgmt.cfg.maxDBCacheNum,
@@ -786,8 +829,8 @@ int32_t catalogGetHandle(uint64_t clusterId, SCatalog** catalogHandle) {
 
     clusterCtg->clusterId = clusterId;
 
-    CTG_ERR_JRET(ctgMetaRentInit(&clusterCtg->dbRent, gCtgMgmt.cfg.dbRentSec, CTG_RENT_DB));
-    CTG_ERR_JRET(ctgMetaRentInit(&clusterCtg->stbRent, gCtgMgmt.cfg.stbRentSec, CTG_RENT_STABLE));
+    CTG_ERR_JRET(ctgMetaRentInit(&clusterCtg->dbRent, gCtgMgmt.cfg.dbRentSec, CTG_RENT_DB, sizeof(SDbCacheInfo)));
+    CTG_ERR_JRET(ctgMetaRentInit(&clusterCtg->stbRent, gCtgMgmt.cfg.stbRentSec, CTG_RENT_STABLE, sizeof(SSTableVersion)));
 
     clusterCtg->dbCache = taosHashInit(gCtgMgmt.cfg.maxDBCacheNum, taosGetDefaultHashFunction(TSDB_DATA_TYPE_BINARY),
                                        false, HASH_ENTRY_LOCK);
@@ -1139,6 +1182,22 @@ _return:
   CTG_API_LEAVE(code);
 }
 
+int32_t catalogAsyncUpdateTableMeta(SCatalog* pCtg, STableMetaRsp* pMsg) {
+  CTG_API_ENTER();
+
+  if (NULL == pCtg || NULL == pMsg) {
+    CTG_API_LEAVE(TSDB_CODE_CTG_INVALID_INPUT);
+  }
+
+  int32_t code = 0;
+  CTG_ERR_JRET(ctgUpdateTbMeta(pCtg, pMsg, false));
+
+_return:
+
+  CTG_API_LEAVE(code);
+}
+
+
 int32_t catalogChkTbMetaVersion(SCatalog* pCtg, SRequestConnInfo* pConn, SArray* pTables) {
   CTG_API_ENTER();
 
@@ -1600,11 +1659,11 @@ int32_t catalogClearCache(void) {
 
   qInfo("start to clear catalog cache");
 
-  if (NULL == gCtgMgmt.pCluster || atomic_load_8((int8_t*)&gCtgMgmt.exit)) {
+  if (NULL == gCtgMgmt.pCluster) {
     CTG_API_LEAVE_NOLOCK(TSDB_CODE_SUCCESS);
   }
 
-  int32_t code = ctgClearCacheEnqueue(NULL, false, false, true);
+  int32_t code = ctgClearCacheEnqueue(NULL, false, false, false, true);
 
   qInfo("clear catalog cache end, code: %s", tstrerror(code));
 
@@ -1618,10 +1677,17 @@ void catalogDestroy(void) {
     return;
   }
 
+  if (gCtgMgmt.cacheTimer) {
+    taosTmrStop(gCtgMgmt.cacheTimer);
+    gCtgMgmt.cacheTimer = NULL;
+    taosTmrCleanUp(gCtgMgmt.timer);
+    gCtgMgmt.timer = NULL;
+  }
+
   atomic_store_8((int8_t*)&gCtgMgmt.exit, true);
 
   if (!taosCheckCurrentInDll()) {
-    ctgClearCacheEnqueue(NULL, true, true, true);
+    ctgClearCacheEnqueue(NULL, false, true, true, true);
     taosThreadJoin(gCtgMgmt.updateThread, NULL);
   }
 

source/libs/catalog/src/ctgCache.c

@@ -33,24 +33,25 @@ SCtgOperation gCtgCacheOperation[CTG_OP_MAX] = {{CTG_OP_UPDATE_VGROUP, "update v
                                                 {CTG_OP_CLEAR_CACHE, "clear cache", ctgOpClearCache}};
 
 SCtgCacheItemInfo gCtgStatItem[CTG_CI_MAX_VALUE] = {
-    {"Cluster   ", CTG_CI_FLAG_LEVEL_GLOBAL},   // CTG_CI_CLUSTER
-    {"Dnode     ", CTG_CI_FLAG_LEVEL_CLUSTER},  // CTG_CI_DNODE,
-    {"Qnode     ", CTG_CI_FLAG_LEVEL_CLUSTER},  // CTG_CI_QNODE,
-    {"DB        ", CTG_CI_FLAG_LEVEL_CLUSTER},  // CTG_CI_DB,
-    {"DbVgroup  ", CTG_CI_FLAG_LEVEL_DB},       // CTG_CI_DB_VGROUP,
-    {"DbCfg     ", CTG_CI_FLAG_LEVEL_DB},       // CTG_CI_DB_CFG,
-    {"DbInfo    ", CTG_CI_FLAG_LEVEL_DB},       // CTG_CI_DB_INFO,
-    {"StbMeta   ", CTG_CI_FLAG_LEVEL_DB},       // CTG_CI_STABLE_META,
-    {"NtbMeta   ", CTG_CI_FLAG_LEVEL_DB},       // CTG_CI_NTABLE_META,
-    {"CtbMeta   ", CTG_CI_FLAG_LEVEL_DB},       // CTG_CI_CTABLE_META,
-    {"SysTblMeta", CTG_CI_FLAG_LEVEL_DB},       // CTG_CI_SYSTABLE_META,
-    {"OthTblMeta", CTG_CI_FLAG_LEVEL_DB},       // CTG_CI_OTHERTABLE_META,
-    {"TblSMA    ", CTG_CI_FLAG_LEVEL_DB},       // CTG_CI_TBL_SMA,
-    {"TblCfg    ", CTG_CI_FLAG_LEVEL_DB},       // CTG_CI_TBL_CFG,
-    {"IndexInfo ", CTG_CI_FLAG_LEVEL_DB},       // CTG_CI_INDEX_INFO,
-    {"User      ", CTG_CI_FLAG_LEVEL_CLUSTER},  // CTG_CI_USER,
-    {"UDF       ", CTG_CI_FLAG_LEVEL_CLUSTER},  // CTG_CI_UDF,
-    {"SvrVer    ", CTG_CI_FLAG_LEVEL_CLUSTER}   // CTG_CI_SVR_VER,
+    {"Cluster   ", CTG_CI_FLAG_LEVEL_GLOBAL},  //CTG_CI_CLUSTER
+    {"Dnode     ", CTG_CI_FLAG_LEVEL_CLUSTER}, //CTG_CI_DNODE,
+    {"Qnode     ", CTG_CI_FLAG_LEVEL_CLUSTER}, //CTG_CI_QNODE,
+    {"DB        ", CTG_CI_FLAG_LEVEL_CLUSTER}, //CTG_CI_DB,
+    {"DbVgroup  ", CTG_CI_FLAG_LEVEL_DB},      //CTG_CI_DB_VGROUP,
+    {"DbCfg     ", CTG_CI_FLAG_LEVEL_DB},      //CTG_CI_DB_CFG,
+    {"DbInfo    ", CTG_CI_FLAG_LEVEL_DB},      //CTG_CI_DB_INFO,
+    {"StbMeta   ", CTG_CI_FLAG_LEVEL_DB},      //CTG_CI_STABLE_META,
+    {"NtbMeta   ", CTG_CI_FLAG_LEVEL_DB},      //CTG_CI_NTABLE_META,
+    {"CtbMeta   ", CTG_CI_FLAG_LEVEL_DB},      //CTG_CI_CTABLE_META,
+    {"SysTblMeta", CTG_CI_FLAG_LEVEL_DB},      //CTG_CI_SYSTABLE_META,
+    {"OthTblMeta", CTG_CI_FLAG_LEVEL_DB},      //CTG_CI_OTHERTABLE_META,
+    {"TblSMA    ", CTG_CI_FLAG_LEVEL_DB},      //CTG_CI_TBL_SMA,
+    {"TblCfg    ", CTG_CI_FLAG_LEVEL_DB},      //CTG_CI_TBL_CFG,
+    {"TblTag    ", CTG_CI_FLAG_LEVEL_DB},      //CTG_CI_TBL_TAG,
+    {"IndexInfo ", CTG_CI_FLAG_LEVEL_DB},      //CTG_CI_INDEX_INFO,
+    {"User      ", CTG_CI_FLAG_LEVEL_CLUSTER}, //CTG_CI_USER,
+    {"UDF       ", CTG_CI_FLAG_LEVEL_CLUSTER}, //CTG_CI_UDF,
+    {"SvrVer    ", CTG_CI_FLAG_LEVEL_CLUSTER}  //CTG_CI_SVR_VER,
 };
 
 int32_t ctgRLockVgInfo(SCatalog *pCtg, SCtgDBCache *dbCache, bool *inCache) {
@@ -860,7 +861,7 @@ int32_t ctgEnqueue(SCatalog *pCtg, SCtgCacheOperation *operation) {
 
   CTG_UNLOCK(CTG_WRITE, &gCtgMgmt.queue.qlock);
 
-  ctgDebug("action [%s] added into queue", opName);
+  ctgDebug("%sync action [%s] added into queue", syncOp ? "S": "As", opName);
 
   CTG_QUEUE_INC();
   CTG_STAT_RT_INC(numOfOpEnqueue, 1);
@@ -1242,7 +1243,7 @@ _return:
   CTG_RET(code);
 }
 
-int32_t ctgClearCacheEnqueue(SCatalog *pCtg, bool freeCtg, bool stopQueue, bool syncOp) {
+int32_t ctgClearCacheEnqueue(SCatalog *pCtg, bool clearMeta, bool freeCtg, bool stopQueue, bool syncOp) {
   int32_t             code = 0;
   SCtgCacheOperation *op = taosMemoryCalloc(1, sizeof(SCtgCacheOperation));
   op->opId = CTG_OP_CLEAR_CACHE;
@@ -1258,6 +1259,7 @@ int32_t ctgClearCacheEnqueue(SCatalog *pCtg, bool freeCtg, bool stopQueue, bool
   }
 
   msg->pCtg = pCtg;
+  msg->clearMeta = clearMeta;
   msg->freeCtg = freeCtg;
   op->data = msg;
 
@@ -1270,10 +1272,11 @@ _return:
   CTG_RET(code);
 }
 
-int32_t ctgMetaRentInit(SCtgRentMgmt *mgmt, uint32_t rentSec, int8_t type) {
+int32_t ctgMetaRentInit(SCtgRentMgmt *mgmt, uint32_t rentSec, int8_t type, int32_t size) {
   mgmt->slotRIdx = 0;
   mgmt->slotNum = rentSec / CTG_RENT_SLOT_SECOND;
   mgmt->type = type;
+  mgmt->metaSize = size;
 
   size_t msgSize = sizeof(SCtgRentSlot) * mgmt->slotNum;
 
@@ -1283,6 +1286,8 @@ int32_t ctgMetaRentInit(SCtgRentMgmt *mgmt, uint32_t rentSec, int8_t type) {
     CTG_ERR_RET(TSDB_CODE_OUT_OF_MEMORY);
   }
 
+  mgmt->rentCacheSize = msgSize;
+
   qDebug("meta rent initialized, type:%d, slotNum:%d", type, mgmt->slotNum);
 
   return TSDB_CODE_SUCCESS;
@@ -1309,6 +1314,7 @@ int32_t ctgMetaRentAdd(SCtgRentMgmt *mgmt, void *meta, int64_t id, int32_t size)
     CTG_ERR_JRET(TSDB_CODE_OUT_OF_MEMORY);
   }
 
+  mgmt->rentCacheSize += size;
   slot->needSort = true;
 
   qDebug("add meta to rent, id:0x%" PRIx64 ", slot idx:%d, type:%d", id, widx, mgmt->type);
@@ -1389,6 +1395,7 @@ int32_t ctgMetaRentRemove(SCtgRentMgmt *mgmt, int64_t id, __compar_fn_t sortComp
   }
 
   taosArrayRemove(slot->meta, idx);
+  mgmt->rentCacheSize -= mgmt->metaSize;
 
   qDebug("meta in rent removed, id:0x%" PRIx64 ", slot idx:%d, type:%d", id, widx, mgmt->type);
 
@@ -1656,10 +1663,15 @@ int32_t ctgWriteTbMetaToCache(SCatalog *pCtg, SCtgDBCache *dbCache, char *dbFNam
     }
 
     if (origType == TSDB_SUPER_TABLE) {
-      if (taosHashRemove(dbCache->stbCache, &orig->suid, sizeof(orig->suid))) {
-        ctgError("stb not exist in stbCache, dbFName:%s, stb:%s, suid:0x%" PRIx64, dbFName, tbName, orig->suid);
-      } else {
-        ctgDebug("stb removed from stbCache, dbFName:%s, stb:%s, suid:0x%" PRIx64, dbFName, tbName, orig->suid);
+      char *stbName = taosHashGet(dbCache->stbCache, &orig->suid, sizeof(orig->suid));
+      if (stbName) {
+        uint64_t metaSize = strlen(stbName) + 1 + sizeof(orig->suid);
+        if (taosHashRemove(dbCache->stbCache, &orig->suid, sizeof(orig->suid))) {
+          ctgError("stb not exist in stbCache, dbFName:%s, stb:%s, suid:0x%" PRIx64, dbFName, tbName, orig->suid);
+        } else {
+          ctgDebug("stb removed from stbCache, dbFName:%s, stb:%s, suid:0x%" PRIx64, dbFName, tbName, orig->suid);
+          atomic_sub_fetch_64(&dbCache->dbCacheSize, metaSize);
+        }
       }
     }
   }
@@ -1673,14 +1685,20 @@ int32_t ctgWriteTbMetaToCache(SCatalog *pCtg, SCtgDBCache *dbCache, char *dbFNam
       CTG_ERR_RET(TSDB_CODE_OUT_OF_MEMORY);
     }
 
+    atomic_add_fetch_64(&dbCache->dbCacheSize, strlen(tbName) + sizeof(SCtgTbCache) + ctgGetTbMetaCacheSize(meta));
+
     pCache = taosHashGet(dbCache->tbCache, tbName, strlen(tbName));
   } else {
     CTG_LOCK(CTG_WRITE, &pCache->metaLock);
     if (orig) {
       CTG_META_NUM_DEC(origType);
     }
+
+    atomic_add_fetch_64(&dbCache->dbCacheSize, ctgGetTbMetaCacheSize(meta) - ctgGetTbMetaCacheSize(pCache->pMeta));
+
     taosMemoryFree(pCache->pMeta);
     pCache->pMeta = meta;
+    
     CTG_UNLOCK(CTG_WRITE, &pCache->metaLock);
   }
 
@@ -1698,6 +1716,8 @@ int32_t ctgWriteTbMetaToCache(SCatalog *pCtg, SCtgDBCache *dbCache, char *dbFNam
     CTG_ERR_RET(TSDB_CODE_OUT_OF_MEMORY);
   }
 
+  atomic_add_fetch_64(&dbCache->dbCacheSize, sizeof(meta->suid) + strlen(tbName) + 1);
+
   ctgDebug("stb 0x%" PRIx64 " updated to cache, dbFName:%s, tbName:%s, tbType:%d", meta->suid, dbFName, tbName,
            meta->tableType);
 
@@ -1730,6 +1750,8 @@ int32_t ctgWriteTbIndexToCache(SCatalog *pCtg, SCtgDBCache *dbCache, char *dbFNa
       CTG_ERR_RET(TSDB_CODE_OUT_OF_MEMORY);
     }
 
+    atomic_add_fetch_64(&dbCache->dbCacheSize, strlen(tbName) + sizeof(SCtgTbCache) + ctgGetTbIndexCacheSize(pIndex));
+
     CTG_DB_NUM_INC(CTG_CI_TBL_SMA);
 
     *index = NULL;
@@ -1746,6 +1768,7 @@ int32_t ctgWriteTbIndexToCache(SCatalog *pCtg, SCtgDBCache *dbCache, char *dbFNa
   CTG_LOCK(CTG_WRITE, &pCache->indexLock);
 
   if (pCache->pIndex) {
+    atomic_sub_fetch_64(&dbCache->dbCacheSize, ctgGetTbIndexCacheSize(pCache->pIndex));
     if (0 == suid) {
       suid = pCache->pIndex->suid;
     }
@@ -1756,6 +1779,8 @@ int32_t ctgWriteTbIndexToCache(SCatalog *pCtg, SCtgDBCache *dbCache, char *dbFNa
   pCache->pIndex = pIndex;
   CTG_UNLOCK(CTG_WRITE, &pCache->indexLock);
 
+  atomic_add_fetch_64(&dbCache->dbCacheSize, ctgGetTbIndexCacheSize(pIndex));
+
   *index = NULL;
 
   ctgDebug("table %s index updated to cache, ver:%d, num:%d", tbName, pIndex->version,
@@ -1785,7 +1810,7 @@ _return:
   CTG_RET(code);
 }
 
-void ctgClearAllInstance(void) {
+void ctgClearAllHandles(void) {
   SCatalog *pCtg = NULL;
 
   void *pIter = taosHashIterate(gCtgMgmt.pCluster, NULL);
@@ -1800,7 +1825,7 @@ void ctgClearAllInstance(void) {
   }
 }
 
-void ctgFreeAllInstance(void) {
+void ctgFreeAllHandles(void) {
   SCatalog *pCtg = NULL;
 
   void *pIter = taosHashIterate(gCtgMgmt.pCluster, NULL);
@@ -1841,7 +1866,7 @@ int32_t ctgOpUpdateVgroup(SCtgCacheOperation *operation) {
     goto _return;
   }
 
-  if (dbInfo->vgVersion < 0 || taosHashGetSize(dbInfo->vgHash) <= 0) {
+  if (dbInfo->vgVersion < 0 || (taosHashGetSize(dbInfo->vgHash) <= 0 && !IS_SYS_DBNAME(dbFName))) {
     ctgDebug("invalid db vgInfo, dbFName:%s, vgHash:%p, vgVersion:%d, vgHashSize:%d", dbFName, dbInfo->vgHash,
              dbInfo->vgVersion, taosHashGetSize(dbInfo->vgHash));
     CTG_ERR_JRET(TSDB_CODE_APP_ERROR);
@@ -1859,6 +1884,7 @@ int32_t ctgOpUpdateVgroup(SCtgCacheOperation *operation) {
   }
 
   SCtgVgCache *vgCache = &dbCache->vgCache;
+  
   CTG_ERR_JRET(ctgWLockVgInfo(msg->pCtg, dbCache));
 
   if (vgCache->vgInfo) {
@@ -1881,6 +1907,11 @@ int32_t ctgOpUpdateVgroup(SCtgCacheOperation *operation) {
       goto _return;
     }
 
+    uint64_t groupCacheSize = ctgGetDbVgroupCacheSize(vgCache->vgInfo);
+    ctgDebug("sub dbGroupCacheSize %" PRIu64 " from db, dbFName:%s", groupCacheSize, dbFName);
+
+    atomic_sub_fetch_64(&dbCache->dbCacheSize, groupCacheSize);
+    
     freeVgInfo(vgInfo);
     CTG_DB_NUM_RESET(CTG_CI_DB_VGROUP);
   }
@@ -1898,6 +1929,10 @@ int32_t ctgOpUpdateVgroup(SCtgCacheOperation *operation) {
 
   ctgWUnlockVgInfo(dbCache);
 
+  uint64_t groupCacheSize = ctgGetDbVgroupCacheSize(vgCache->vgInfo);
+  atomic_add_fetch_64(&dbCache->dbCacheSize, groupCacheSize);
+  ctgDebug("add dbGroupCacheSize %" PRIu64 " from db, dbFName:%s", groupCacheSize, dbFName);
+
   dbCache = NULL;
 
   // if (!IS_SYS_DBNAME(dbFName)) {
@@ -2022,6 +2057,8 @@ int32_t ctgOpDropDbVgroup(SCtgCacheOperation *operation) {
 
   CTG_ERR_JRET(ctgWLockVgInfo(pCtg, dbCache));
 
+  atomic_sub_fetch_64(&dbCache->dbCacheSize, ctgGetDbVgroupCacheSize(dbCache->vgCache.vgInfo));
+
   freeVgInfo(dbCache->vgCache.vgInfo);
   dbCache->vgCache.vgInfo = NULL;
 
@@ -2113,26 +2150,32 @@ int32_t ctgOpDropStbMeta(SCtgCacheOperation *operation) {
     goto _return;
   }
 
-  if (taosHashRemove(dbCache->stbCache, &msg->suid, sizeof(msg->suid))) {
-    ctgDebug("stb not exist in stbCache, may be removed, dbFName:%s, stb:%s, suid:0x%" PRIx64, msg->dbFName,
-             msg->stbName, msg->suid);
+  char *stbName = taosHashGet(dbCache->stbCache, &msg->suid, sizeof(msg->suid));
+  if (stbName) {
+    uint64_t metaSize = strlen(stbName) + 1 + sizeof(msg->suid);
+    if (taosHashRemove(dbCache->stbCache, &msg->suid, sizeof(msg->suid))) {
+      ctgDebug("stb not exist in stbCache, may be removed, dbFName:%s, stb:%s, suid:0x%" PRIx64, msg->dbFName,
+               msg->stbName, msg->suid);
+    } else {
+      atomic_sub_fetch_64(&dbCache->dbCacheSize, metaSize);
+    }
   }
-
+  
   SCtgTbCache *pTbCache = taosHashGet(dbCache->tbCache, msg->stbName, strlen(msg->stbName));
   if (NULL == pTbCache) {
     ctgDebug("stb %s already not in cache", msg->stbName);
     goto _return;
   }
 
-  CTG_LOCK(CTG_WRITE, &pTbCache->metaLock);
   tblType = pTbCache->pMeta->tableType;
-  ctgFreeTbCacheImpl(pTbCache);
-  CTG_UNLOCK(CTG_WRITE, &pTbCache->metaLock);
+  atomic_sub_fetch_64(&dbCache->dbCacheSize, ctgGetTbMetaCacheSize(pTbCache->pMeta) + ctgGetTbIndexCacheSize(pTbCache->pIndex));
+  ctgFreeTbCacheImpl(pTbCache, true);
 
   if (taosHashRemove(dbCache->tbCache, msg->stbName, strlen(msg->stbName))) {
     ctgError("stb not exist in cache, dbFName:%s, stb:%s, suid:0x%" PRIx64, msg->dbFName, msg->stbName, msg->suid);
   } else {
     CTG_META_NUM_DEC(tblType);
+    atomic_sub_fetch_64(&dbCache->dbCacheSize, sizeof(*pTbCache) + strlen(msg->stbName));
   }
 
   ctgInfo("stb removed from cache, dbFName:%s, stbName:%s, suid:0x%" PRIx64, msg->dbFName, msg->stbName, msg->suid);
@@ -2176,15 +2219,15 @@ int32_t ctgOpDropTbMeta(SCtgCacheOperation *operation) {
     goto _return;
   }
 
-  CTG_LOCK(CTG_WRITE, &pTbCache->metaLock);
   tblType = pTbCache->pMeta->tableType;
-  ctgFreeTbCacheImpl(pTbCache);
-  CTG_UNLOCK(CTG_WRITE, &pTbCache->metaLock);
+  atomic_sub_fetch_64(&dbCache->dbCacheSize, ctgGetTbMetaCacheSize(pTbCache->pMeta) + ctgGetTbIndexCacheSize(pTbCache->pIndex));
+  ctgFreeTbCacheImpl(pTbCache, true);
 
   if (taosHashRemove(dbCache->tbCache, msg->tbName, strlen(msg->tbName))) {
     ctgError("tb %s not exist in cache, dbFName:%s", msg->tbName, msg->dbFName);
     CTG_ERR_JRET(TSDB_CODE_CTG_INTERNAL_ERROR);
   } else {
+    atomic_sub_fetch_64(&dbCache->dbCacheSize, sizeof(*pTbCache) + strlen(msg->tbName));
     CTG_META_NUM_DEC(tblType);
   }
 
@@ -2211,7 +2254,8 @@ int32_t ctgOpUpdateUser(SCtgCacheOperation *operation) {
     SCtgUserAuth userAuth = {0};
 
     memcpy(&userAuth.userAuth, &msg->userAuth, sizeof(msg->userAuth));
-
+    userAuth.userCacheSize = ctgGetUserCacheSize(&userAuth.userAuth);
+    
     if (taosHashPut(pCtg->userCache, msg->userAuth.user, strlen(msg->userAuth.user), &userAuth, sizeof(userAuth))) {
       ctgError("taosHashPut user %s to cache failed", msg->userAuth.user);
       CTG_ERR_JRET(TSDB_CODE_OUT_OF_MEMORY);
@@ -2243,6 +2287,8 @@ int32_t ctgOpUpdateUser(SCtgCacheOperation *operation) {
   msg->userAuth.useDbs = NULL;
 
   CTG_UNLOCK(CTG_WRITE, &pUser->lock);
+  
+  atomic_store_64(&pUser->userCacheSize, ctgGetUserCacheSize(&pUser->userAuth));
 
 _return:
 
@@ -2378,33 +2424,78 @@ _return:
   CTG_RET(code);
 }
 
-int32_t ctgOpClearCache(SCtgCacheOperation *operation) {
-  int32_t            code = 0;
+void ctgClearFreeCache(SCtgCacheOperation *operation) {
   SCtgClearCacheMsg *msg = operation->data;
   SCatalog          *pCtg = msg->pCtg;
 
   CTG_LOCK(CTG_WRITE, &gCtgMgmt.lock);
-
+  
   if (pCtg) {
     if (msg->freeCtg) {
       ctgFreeHandle(pCtg);
     } else {
       ctgClearHandle(pCtg);
     }
+  } else if (msg->freeCtg) {
+    ctgFreeAllHandles();
+  } else {
+    ctgClearAllHandles();
+  }
+  
+  CTG_UNLOCK(CTG_WRITE, &gCtgMgmt.lock);
+}
 
-    goto _return;
+void ctgClearMetaCache(SCtgCacheOperation *operation) {
+  SCtgClearCacheMsg *msg = operation->data;
+  SCatalog          *pCtg = msg->pCtg;
+  int64_t            clearedSize = 0;
+  int64_t            clearedNum = 0;
+  int64_t            remainSize = 0;
+  bool               roundDone = false;
+
+  if (pCtg) {
+    ctgClearHandleMeta(pCtg, &clearedSize, &clearedNum, &roundDone);
+  } else {
+    ctgClearAllHandleMeta(&clearedSize, &clearedNum, &roundDone);
+  }
+
+  qDebug("catalog finish one round meta clear, clearedSize:%" PRId64 ", clearedNum:%" PRId64 ", done:%d", clearedSize, clearedNum, roundDone);
+
+  ctgGetGlobalCacheSize(&remainSize);
+  int32_t cacheMaxSize = atomic_load_32(&tsMetaCacheMaxSize);
+  
+  if (CTG_CACHE_LOW(remainSize, cacheMaxSize)) {
+    qDebug("catalog finish meta clear, remainSize:%" PRId64 ", cacheMaxSize:%dMB", remainSize, cacheMaxSize);
+    taosTmrReset(ctgProcessTimerEvent, CTG_DEFAULT_CACHE_MON_MSEC, NULL, gCtgMgmt.timer, &gCtgMgmt.cacheTimer);
+    return;
+  }
+
+  if (!roundDone) {
+    qDebug("catalog all meta cleared, remainSize:%" PRId64 ", cacheMaxSize:%dMB, to clear handle", remainSize, cacheMaxSize);
+    ctgClearFreeCache(operation);
+    taosTmrReset(ctgProcessTimerEvent, CTG_DEFAULT_CACHE_MON_MSEC, NULL, gCtgMgmt.timer, &gCtgMgmt.cacheTimer);
+    return;
+  }
+  
+  int32_t code = ctgClearCacheEnqueue(NULL, true, false, false, false);
+  if (code) {
+    qError("clear cache enqueue failed, error:%s", tstrerror(code));
+    taosTmrReset(ctgProcessTimerEvent, CTG_DEFAULT_CACHE_MON_MSEC, NULL, gCtgMgmt.timer, &gCtgMgmt.cacheTimer);
   }
+}
 
-  if (msg->freeCtg) {
-    ctgFreeAllInstance();
+int32_t ctgOpClearCache(SCtgCacheOperation *operation) {
+  int32_t            code = 0;
+  SCtgClearCacheMsg *msg = operation->data;
+
+  if (msg->clearMeta) {
+    ctgClearMetaCache(operation);
   } else {
-    ctgClearAllInstance();
+    ctgClearFreeCache(operation);
   }
 
 _return:
 
-  CTG_UNLOCK(CTG_WRITE, &gCtgMgmt.lock);
-
   taosMemoryFreeClear(msg);
 
   CTG_RET(code);

source/libs/catalog/src/ctgDbg.c

@@ -19,7 +19,7 @@
 #include "trpc.h"
 
 extern SCatalogMgmt gCtgMgmt;
-SCtgDebug           gCTGDebug = {0};
+SCtgDebug           gCTGDebug = {.statEnable = true};
 
 #if 0
 
@@ -547,7 +547,10 @@ int32_t ctgdShowStatInfo(void) {
   CTG_API_ENTER();
 
   SCtgCacheStat cache;
+  uint64_t cacheSize = 0;
+  
   ctgGetGlobalCacheStat(&cache);
+  ctgGetGlobalCacheSize(&cacheSize);
 
   qDebug("## Global Stat Info %s ##", "begin");
   qDebug("##            \t%s \t%s \t%s ##", "Num", "Hit", "Nhit");
@@ -555,6 +558,7 @@ int32_t ctgdShowStatInfo(void) {
     qDebug("#  %s \t%" PRIu64 " \t%" PRIu64 " \t%" PRIu64 " #", gCtgStatItem[i].name, cache.cacheNum[i], cache.cacheHit[i], cache.cacheNHit[i]);
   }
   qDebug("## Global Stat Info %s ##", "end");
+  qDebug("## Global Cache Size: %" PRIu64, cacheSize);
 
   CTG_API_LEAVE(TSDB_CODE_SUCCESS);
 }

source/libs/catalog/src/ctgUtil.c

@@ -199,6 +199,7 @@ void ctgFreeMetaRent(SCtgRentMgmt* mgmt) {
   }
 
   taosMemoryFreeClear(mgmt->slots);
+  mgmt->rentCacheSize = 0;
 }
 
 void ctgFreeStbMetaCache(SCtgDBCache* dbCache) {
@@ -211,12 +212,26 @@ void ctgFreeStbMetaCache(SCtgDBCache* dbCache) {
   dbCache->stbCache = NULL;
 }
 
-void ctgFreeTbCacheImpl(SCtgTbCache* pCache) {
-  qDebug("tbMeta freed, p:%p", pCache->pMeta);
-  taosMemoryFreeClear(pCache->pMeta);
+void ctgFreeTbCacheImpl(SCtgTbCache* pCache, bool lock) {
+  if (pCache->pMeta) {
+    if (lock) { 
+      CTG_LOCK(CTG_WRITE, &pCache->metaLock);
+    }
+    taosMemoryFreeClear(pCache->pMeta);
+    if (lock) { 
+      CTG_UNLOCK(CTG_WRITE, &pCache->metaLock);
+    }
+  }
+
   if (pCache->pIndex) {
+    if (lock) { 
+      CTG_LOCK(CTG_WRITE, &pCache->indexLock);
+    }
     taosArrayDestroyEx(pCache->pIndex->pIndex, tFreeSTableIndexInfo);
     taosMemoryFreeClear(pCache->pIndex);
+    if (lock) { 
+      CTG_UNLOCK(CTG_WRITE, &pCache->indexLock);
+    }
   }
 }
 
@@ -228,7 +243,7 @@ void ctgFreeTbCache(SCtgDBCache* dbCache) {
   int32_t      tblNum = taosHashGetSize(dbCache->tbCache);
   SCtgTbCache* pCache = taosHashIterate(dbCache->tbCache, NULL);
   while (NULL != pCache) {
-    ctgFreeTbCacheImpl(pCache);
+    ctgFreeTbCacheImpl(pCache, false);
     pCache = taosHashIterate(dbCache->tbCache, pCache);
   }
   taosHashCleanup(dbCache->tbCache);
@@ -316,21 +331,86 @@ void ctgFreeHandle(SCatalog* pCtg) {
   ctgInfo("handle freed, clusterId:0x%" PRIx64, clusterId);
 }
 
+void ctgClearHandleMeta(SCatalog* pCtg, int64_t *pClearedSize, int64_t *pCleardNum, bool *roundDone) {
+  int64_t cacheSize = 0;
+  void* pIter = taosHashIterate(pCtg->dbCache, NULL);
+  while (pIter) {
+    SCtgDBCache* dbCache = pIter;
+    
+    SCtgTbCache* pCache = taosHashIterate(dbCache->tbCache, NULL);
+    while (NULL != pCache) {
+      size_t len = 0;
+      void*  key = taosHashGetKey(pCache, &len);
+
+      if (pCache->pMeta && TSDB_SUPER_TABLE == pCache->pMeta->tableType) {
+        pCache = taosHashIterate(dbCache->tbCache, pCache);
+        continue;
+      }
+      
+      taosHashRemove(dbCache->tbCache, key, len);
+      cacheSize = len + sizeof(SCtgTbCache) + ctgGetTbMetaCacheSize(pCache->pMeta) + ctgGetTbIndexCacheSize(pCache->pIndex);
+      atomic_sub_fetch_64(&dbCache->dbCacheSize, cacheSize);
+      *pClearedSize += cacheSize;
+      (*pCleardNum)++;
+
+      if (pCache->pMeta) {
+        CTG_META_NUM_DEC(pCache->pMeta->tableType);
+      }
+      
+      ctgFreeTbCacheImpl(pCache, true);
+      
+      if (*pCleardNum >= CTG_CLEAR_CACHE_ROUND_TB_NUM) {
+        taosHashCancelIterate(dbCache->tbCache, pCache);
+        goto _return;
+      }
+            
+      pCache = taosHashIterate(dbCache->tbCache, pCache);
+    }
+
+    pIter = taosHashIterate(pCtg->dbCache, pIter);
+  }
+
+_return:
+
+  if (*pCleardNum >= CTG_CLEAR_CACHE_ROUND_TB_NUM) {
+    *roundDone = true;
+  }
+}
+
+void ctgClearAllHandleMeta(int64_t *clearedSize, int64_t *clearedNum, bool *roundDone) {
+  SCatalog *pCtg = NULL;
+
+  void *pIter = taosHashIterate(gCtgMgmt.pCluster, NULL);
+  while (pIter) {
+    pCtg = *(SCatalog **)pIter;
+
+    if (pCtg) {
+      ctgClearHandleMeta(pCtg, clearedSize, clearedNum, roundDone);
+      if (*roundDone) {
+        taosHashCancelIterate(gCtgMgmt.pCluster, pIter);
+        break;
+      }
+    }
+
+    pIter = taosHashIterate(gCtgMgmt.pCluster, pIter);
+  }
+}
+
 void ctgClearHandle(SCatalog* pCtg) {
   if (NULL == pCtg) {
     return;
   }
 
   uint64_t clusterId = pCtg->clusterId;
-
+  
   ctgFreeMetaRent(&pCtg->dbRent);
   ctgFreeMetaRent(&pCtg->stbRent);
 
   ctgFreeInstDbCache(pCtg->dbCache);
   ctgFreeInstUserCache(pCtg->userCache);
 
-  ctgMetaRentInit(&pCtg->dbRent, gCtgMgmt.cfg.dbRentSec, CTG_RENT_DB);
-  ctgMetaRentInit(&pCtg->stbRent, gCtgMgmt.cfg.stbRentSec, CTG_RENT_STABLE);
+  ctgMetaRentInit(&pCtg->dbRent, gCtgMgmt.cfg.dbRentSec, CTG_RENT_DB, sizeof(SDbCacheInfo));
+  ctgMetaRentInit(&pCtg->stbRent, gCtgMgmt.cfg.stbRentSec, CTG_RENT_STABLE, sizeof(SSTableVersion));
 
   pCtg->dbCache = taosHashInit(gCtgMgmt.cfg.maxDBCacheNum, taosGetDefaultHashFunction(TSDB_DATA_TYPE_BINARY), false,
                                HASH_ENTRY_LOCK);
@@ -1624,6 +1704,130 @@ void catalogFreeMetaData(SMetaData* pData) {
 }
 #endif
 
+uint64_t ctgGetTbIndexCacheSize(STableIndex *pIndex) {
+  if (NULL == pIndex) {
+    return 0;
+  }
+
+  return sizeof(*pIndex) + pIndex->indexSize;
+}
+
+FORCE_INLINE uint64_t ctgGetTbMetaCacheSize(STableMeta *pMeta) {
+  if (NULL == pMeta) {
+    return 0;
+  }
+
+  switch (pMeta->tableType) {
+    case TSDB_SUPER_TABLE:
+      return sizeof(*pMeta) + (pMeta->tableInfo.numOfColumns + pMeta->tableInfo.numOfTags) * sizeof(SSchema);
+    case TSDB_CHILD_TABLE:
+      return sizeof(SCTableMeta);
+    default:
+      return sizeof(*pMeta) + pMeta->tableInfo.numOfColumns * sizeof(SSchema);
+  }
+
+  return 0;
+}
+
+uint64_t ctgGetDbVgroupCacheSize(SDBVgInfo *pVg) {
+  if (NULL == pVg) {
+    return 0;
+  }
+
+  return sizeof(*pVg) + taosHashGetSize(pVg->vgHash) * (sizeof(SVgroupInfo) + sizeof(int32_t)) 
+    + taosArrayGetSize(pVg->vgArray) * sizeof(SVgroupInfo);
+}
+
+uint64_t ctgGetUserCacheSize(SGetUserAuthRsp *pAuth) {
+  if (NULL == pAuth) {
+    return 0;
+  }
+
+  uint64_t cacheSize = 0;
+  char* p = taosHashIterate(pAuth->createdDbs, NULL);
+  while (p != NULL) {
+    size_t len = 0;
+    void*  key = taosHashGetKey(p, &len);
+    cacheSize += len + strlen(p);
+
+    p = taosHashIterate(pAuth->createdDbs, p);
+  }
+
+  p = taosHashIterate(pAuth->readDbs, NULL);
+  while (p != NULL) {
+    size_t len = 0;
+    void*  key = taosHashGetKey(p, &len);
+    cacheSize += len + strlen(p);
+
+    p = taosHashIterate(pAuth->readDbs, p);
+  }  
+
+  p = taosHashIterate(pAuth->writeDbs, NULL);
+  while (p != NULL) {
+    size_t len = 0;
+    void*  key = taosHashGetKey(p, &len);
+    cacheSize += len + strlen(p);
+
+    p = taosHashIterate(pAuth->writeDbs, p);
+  } 
+
+  p = taosHashIterate(pAuth->readTbs, NULL);
+  while (p != NULL) {
+    size_t len = 0;
+    void*  key = taosHashGetKey(p, &len);
+    cacheSize += len + strlen(p);
+
+    p = taosHashIterate(pAuth->readTbs, p);
+  } 
+
+  p = taosHashIterate(pAuth->writeTbs, NULL);
+  while (p != NULL) {
+    size_t len = 0;
+    void*  key = taosHashGetKey(p, &len);
+    cacheSize += len + strlen(p);
+
+    p = taosHashIterate(pAuth->writeTbs, p);
+  } 
+
+  int32_t *ref = taosHashIterate(pAuth->useDbs, NULL);
+  while (ref != NULL) {
+    size_t len = 0;
+    void*  key = taosHashGetKey(ref, &len);
+    cacheSize += len + sizeof(*ref);
+
+    ref = taosHashIterate(pAuth->useDbs, ref);
+  } 
+
+  return cacheSize;
+}
+
+uint64_t ctgGetClusterCacheSize(SCatalog *pCtg) {
+  uint64_t cacheSize = sizeof(SCatalog);
+  
+  SCtgUserAuth* pAuth = taosHashIterate(pCtg->userCache, NULL);
+  while (pAuth != NULL) {
+    size_t len = 0;
+    void*  key = taosHashGetKey(pAuth, &len);
+    cacheSize += len + sizeof(SCtgUserAuth) + atomic_load_64(&pAuth->userCacheSize);
+
+    pAuth = taosHashIterate(pCtg->userCache, pAuth);
+  }
+
+  SCtgDBCache* pDb = taosHashIterate(pCtg->dbCache, NULL);
+  while (pDb != NULL) {
+    size_t len = 0;
+    void*  key = taosHashGetKey(pDb, &len);
+    cacheSize += len + sizeof(SCtgDBCache) + atomic_load_64(&pDb->dbCacheSize);
+
+    pDb = taosHashIterate(pCtg->dbCache, pDb);
+  }
+
+  cacheSize += pCtg->dbRent.rentCacheSize;
+  cacheSize += pCtg->stbRent.rentCacheSize;
+
+  return cacheSize;
+}
+
 void ctgGetClusterCacheStat(SCatalog* pCtg) {
   for (int32_t i = 0; i < CTG_CI_MAX_VALUE; ++i) {
     if (0 == (gCtgStatItem[i].flag & CTG_CI_FLAG_LEVEL_DB)) {
@@ -1688,3 +1892,23 @@ void ctgGetGlobalCacheStat(SCtgCacheStat* pStat) {
 
   memcpy(pStat, &gCtgMgmt.statInfo.cache, sizeof(gCtgMgmt.statInfo.cache));
 }
+
+void ctgGetGlobalCacheSize(uint64_t *pSize) {
+  *pSize = 0;
+
+  SCatalog* pCtg = NULL;
+  void*     pIter = taosHashIterate(gCtgMgmt.pCluster, NULL);
+  while (pIter) {
+    size_t len = 0;
+    void*  key = taosHashGetKey(pIter, &len);
+    *pSize += len + POINTER_BYTES; 
+    
+    pCtg = *(SCatalog**)pIter;
+    if (pCtg) {
+      *pSize += ctgGetClusterCacheSize(pCtg);
+    }
+
+    pIter = taosHashIterate(gCtgMgmt.pCluster, pIter);
+  }
+}
+

source/libs/catalog/test/catalogTests.cpp

@@ -37,6 +37,7 @@
 #include "tglobal.h"
 #include "trpc.h"
 #include "tvariant.h"
+#include "ttimer.h"
 
 namespace {
 
@@ -150,6 +151,7 @@ void ctgTestInitLogFile() {
 
   tsAsyncLog = 0;
   qDebugFlag = 159;
+  tmrDebugFlag = 159;
   strcpy(tsLogDir, TD_LOG_DIR_PATH);
 
   ctgdEnableDebug("api", true);
@@ -1746,6 +1748,8 @@ TEST(tableMeta, updateStbMeta) {
 
   code = catalogUpdateTableMeta(pCtg, &rsp);
   ASSERT_EQ(code, 0);
+  code = catalogAsyncUpdateTableMeta(pCtg, &rsp);
+  ASSERT_EQ(code, 0);
   taosMemoryFreeClear(rsp.pSchemas);
 
   while (true) {

source/libs/command/src/command.c

@@ -748,7 +748,7 @@ static int32_t execAlterLocal(SAlterLocalStmt* pStmt) {
     return terrno;
   }
 
-  if (taosSetCfg(tsCfg, pStmt->config)) {
+  if (taosApplyLocalCfg(tsCfg, pStmt->config)) {
     return terrno;
   }
 

source/util/src/ttimer.c

@@ -113,7 +113,7 @@ typedef struct time_wheel_t {
 
 static int32_t tsMaxTmrCtrl = TSDB_MAX_VNODES_PER_DB + 100;
 
-static TdThreadOnce  tmrModuleInit = PTHREAD_ONCE_INIT;
+static int32_t       tmrModuleInit = 0;
 static TdThreadMutex tmrCtrlMutex;
 static tmr_ctrl_t*   tmrCtrls;
 static tmr_ctrl_t*   unusedTmrCtrl = NULL;
@@ -512,11 +512,11 @@ bool taosTmrReset(TAOS_TMR_CALLBACK fp, int32_t mseconds, void* param, void* han
   return stopped;
 }
 
-static void taosTmrModuleInit(void) {
+static int32_t taosTmrModuleInit(void) {
   tmrCtrls = taosMemoryMalloc(sizeof(tmr_ctrl_t) * tsMaxTmrCtrl);
   if (tmrCtrls == NULL) {
     tmrError("failed to allocate memory for timer controllers.");
-    return;
+    return -1;
   }
 
   memset(&timerMap, 0, sizeof(timerMap));
@@ -535,14 +535,14 @@ static void taosTmrModuleInit(void) {
     time_wheel_t* wheel = wheels + i;
     if (taosThreadMutexInit(&wheel->mutex, NULL) != 0) {
       tmrError("failed to create the mutex for wheel, reason:%s", strerror(errno));
-      return;
+      return -1;
     }
     wheel->nextScanAt = now + wheel->resolution;
     wheel->index = 0;
     wheel->slots = (tmr_obj_t**)taosMemoryCalloc(wheel->size, sizeof(tmr_obj_t*));
     if (wheel->slots == NULL) {
       tmrError("failed to allocate wheel slots");
-      return;
+      return -1;
     }
     timerMap.size += wheel->size;
   }
@@ -551,20 +551,48 @@ static void taosTmrModuleInit(void) {
   timerMap.slots = (timer_list_t*)taosMemoryCalloc(timerMap.size, sizeof(timer_list_t));
   if (timerMap.slots == NULL) {
     tmrError("failed to allocate hash map");
-    return;
+    return -1;
   }
 
   tmrQhandle = taosInitScheduler(10000, taosTmrThreads, "tmr", NULL);
   taosInitTimer(taosTimerLoopFunc, MSECONDS_PER_TICK);
 
   tmrDebug("timer module is initialized, number of threads: %d", taosTmrThreads);
+
+  return 2;
+}
+
+static int32_t taosTmrInitModule(void) {
+  if (atomic_load_32(&tmrModuleInit) == 2) {
+    return 0;
+  }
+
+  if (atomic_load_32(&tmrModuleInit) < 0) {
+    return -1;
+  }
+  
+  while (true) {
+    if (0 == atomic_val_compare_exchange_32(&tmrModuleInit, 0, 1)) {
+      atomic_store_32(&tmrModuleInit, taosTmrModuleInit());
+    } else if (atomic_load_32(&tmrModuleInit) < 0) {
+      return -1;
+    } else if (atomic_load_32(&tmrModuleInit) == 2) {
+      return 0;
+    } else {
+      taosMsleep(1);
+    }
+  }
+
+  return -1;
 }
 
 void* taosTmrInit(int32_t maxNumOfTmrs, int32_t resolution, int32_t longest, const char* label) {
   const char* ret = taosMonotonicInit();
   tmrDebug("ttimer monotonic clock source:%s", ret);
 
-  taosThreadOnce(&tmrModuleInit, taosTmrModuleInit);
+  if (taosTmrInitModule() < 0) {
+    return NULL;
+  }
 
   taosThreadMutexLock(&tmrCtrlMutex);
   tmr_ctrl_t* ctrl = unusedTmrCtrl;
@@ -581,6 +609,7 @@ void* taosTmrInit(int32_t maxNumOfTmrs, int32_t resolution, int32_t longest, con
   }
 
   tstrncpy(ctrl->label, label, sizeof(ctrl->label));
+  
   tmrDebug("%s timer controller is initialized, number of timer controllers: %d.", label, numOfTmrCtrl);
   return ctrl;
 }
@@ -629,8 +658,6 @@ void taosTmrCleanUp(void* handle) {
 
     tmrCtrls = NULL;
     unusedTmrCtrl = NULL;
-#if defined(LINUX)
-    tmrModuleInit = PTHREAD_ONCE_INIT;  // to support restart
-#endif
+    atomic_store_32(&tmrModuleInit, 0);
   }
 }