Merge pull request #3634 from taosdata/feature/crash

TD-1388

src/client/src/tscSQLParser.c

@@ -5755,7 +5755,7 @@ int32_t tscCheckCreateDbParams(SSqlCmd* pCmd, SCMCreateDbMsg* pCreate) {
   char msg[512] = {0};
 
   if (pCreate->walLevel != -1 && (pCreate->walLevel < TSDB_MIN_WAL_LEVEL || pCreate->walLevel > TSDB_MAX_WAL_LEVEL)) {
-    snprintf(msg, tListLen(msg), "invalid db option walLevel: %d, only 0-2 allowed", pCreate->walLevel);
+    snprintf(msg, tListLen(msg), "invalid db option walLevel: %d, only 1-2 allowed", pCreate->walLevel);
     return invalidSqlErrMsg(tscGetErrorMsgPayload(pCmd), msg);
   }
 

src/inc/twal.h

@@ -44,6 +44,7 @@ typedef void* twalh;  // WAL HANDLE
 typedef int (*FWalWrite)(void *ahandle, void *pHead, int type);
 
 twalh   walOpen(const char *path, const SWalCfg *pCfg);
+int     walAlter(twalh pWal, const SWalCfg *pCfg);
 void    walClose(twalh);
 int     walRenew(twalh);
 int     walWrite(twalh, SWalHead *);

src/mnode/src/mnodeDb.c

@@ -910,13 +910,13 @@ static SDbCfg mnodeGetAlterDbOption(SDbObj *pDb, SCMAlterDbMsg *pAlter) {
   }
 
   if (walLevel > 0 && walLevel != pDb->cfg.walLevel) {
-    mError("db:%s, can't alter walLevel option", pDb->name);
+    mDebug("db:%s, walLevel:%d change to %d", pDb->name, pDb->cfg.walLevel, walLevel);
-    terrno = TSDB_CODE_MND_INVALID_DB_OPTION;
+    newCfg.walLevel = walLevel;
   }
 
   if (fsyncPeriod >= 0 && fsyncPeriod != pDb->cfg.fsyncPeriod) {
-    mError("db:%s, can't alter fsyncPeriod option", pDb->name);
+    mDebug("db:%s, fsyncPeriod:%d change to %d", pDb->name, pDb->cfg.fsyncPeriod, fsyncPeriod);
-    terrno = TSDB_CODE_MND_INVALID_DB_OPTION;
+    newCfg.fsyncPeriod = fsyncPeriod;
   }
 
   if (replications > 0 && replications != pDb->cfg.replications) {

src/vnode/src/vnodeMain.c

@@ -186,6 +186,12 @@ int32_t vnodeAlter(void *param, SMDCreateVnodeMsg *pVnodeCfg) {
     return code; 
   }
 
+  code = walAlter(pVnode->wal, &pVnode->walCfg);
+  if (code != TSDB_CODE_SUCCESS) {
+    pVnode->status = TAOS_VN_STATUS_READY;
+    return code;
+  }
+
   code = syncReconfig(pVnode->sync, &pVnode->syncCfg);
   if (code != TSDB_CODE_SUCCESS) {
     pVnode->status = TAOS_VN_STATUS_READY;
@@ -390,6 +396,7 @@ void vnodeRelease(void *pVnodeRaw) {
     if (0 == tsEnableVnodeBak) {
       vInfo("vgId:%d, vnode backup not enabled", pVnode->vgId);
     } else {
+      taosRemoveDir(newDir);
       taosRename(rootDir, newDir);
     }
 

src/wal/src/walMain.c

@@ -69,6 +69,13 @@ static void walModuleInitFunc() {
     wDebug("WAL module is initialized");
 }
 
+static inline bool walNeedFsyncTimer(SWal *pWal) {
+  if (pWal->fsyncPeriod > 0 && pWal->level == TAOS_WAL_FSYNC) {
+    return true;
+  }
+  return false;
+}
+
 void *walOpen(const char *path, const SWalCfg *pCfg) {
   SWal *pWal = calloc(sizeof(SWal), 1);
   if (pWal == NULL) {
@@ -95,7 +102,7 @@ void *walOpen(const char *path, const SWalCfg *pCfg) {
   tstrncpy(pWal->path, path, sizeof(pWal->path));
   pthread_mutex_init(&pWal->mutex, NULL);
 
-  if (pWal->fsyncPeriod > 0  && pWal->level == TAOS_WAL_FSYNC) {
+  if (walNeedFsyncTimer(pWal)) {
     pWal->timer = taosTmrStart(walProcessFsyncTimer, pWal->fsyncPeriod, pWal, walTmrCtrl);
     if (pWal->timer == NULL) {
       terrno = TAOS_SYSTEM_ERROR(errno);
@@ -127,6 +134,37 @@ void *walOpen(const char *path, const SWalCfg *pCfg) {
   return pWal;
 }
 
+int walAlter(twalh wal, const SWalCfg *pCfg) {
+  SWal *pWal = wal;
+  if (pWal == NULL) {
+    return TSDB_CODE_WAL_APP_ERROR;
+  }
+
+  if (pWal->level == pCfg->walLevel && pWal->fsyncPeriod == pCfg->fsyncPeriod) {
+    wDebug("wal:%s, old walLevel:%d fsync:%d, new walLevel:%d fsync:%d not change", pWal->name, pWal->level,
+           pWal->fsyncPeriod, pCfg->walLevel, pCfg->fsyncPeriod);
+    return TSDB_CODE_SUCCESS;
+  }
+
+  wInfo("wal:%s, change old walLevel:%d fsync:%d, new walLevel:%d fsync:%d", pWal->name, pWal->level, pWal->fsyncPeriod,
+        pCfg->walLevel, pCfg->fsyncPeriod);
+
+  pthread_mutex_lock(&pWal->mutex);
+  pWal->level = pCfg->walLevel;
+  pWal->fsyncPeriod = pCfg->fsyncPeriod;
+  if (walNeedFsyncTimer(pWal)) {
+    wInfo("wal:%s, reset fsync timer, walLevel:%d fsyncPeriod:%d", pWal->name, pWal->level, pWal->fsyncPeriod);
+    taosTmrReset(walProcessFsyncTimer, pWal->fsyncPeriod, pWal, &pWal->timer,walTmrCtrl);
+  } else {
+    wInfo("wal:%s, stop fsync timer, walLevel:%d fsyncPeriod:%d", pWal->name, pWal->level, pWal->fsyncPeriod);
+    taosTmrStop(pWal->timer);
+    pWal->timer = NULL;
+  }
+  pthread_mutex_unlock(&pWal->mutex);
+
+  return TSDB_CODE_SUCCESS;
+}
+
 void walClose(void *handle) {
   if (handle == NULL) return;
   
@@ -484,6 +522,12 @@ static void walProcessFsyncTimer(void *param, void *tmrId) {
   if (fsync(pWal->fd) < 0) {
     wError("wal:%s, fsync failed(%s)", pWal->name, strerror(errno));
   }
-  
+
-  pWal->timer = taosTmrStart(walProcessFsyncTimer, pWal->fsyncPeriod, pWal, walTmrCtrl);
+  if (walNeedFsyncTimer(pWal)) {
+    pWal->timer = taosTmrStart(walProcessFsyncTimer, pWal->fsyncPeriod, pWal, walTmrCtrl);
+  } else {
+    wInfo("wal:%s, stop fsync timer for walLevel:%d fsyncPeriod:%d", pWal->name, pWal->level, pWal->fsyncPeriod);
+    taosTmrStop(pWal->timer);
+    pWal->timer = NULL;
+  }
 }

tests/script/general/db/alter_option.sim

@@ -218,7 +218,10 @@ if $data12_db != 1 then
   return -1
 endi
 
-sql_error alter database db wal 2
+sql alter database db wal 1
+sql alter database db wal 2
+sql alter database db wal 1
+sql alter database db wal 2
 sql_error alter database db wal 0
 sql_error alter database db wal 3
 sql_error alter database db wal 4
@@ -226,11 +229,13 @@ sql_error alter database db wal -1
 sql_error alter database db wal 1000
 
 print ============== step fsync
-sql_error alter database db fsync 2
+sql alter database db fsync 0
-sql_error alter database db fsync 3
+sql alter database db fsync 1
-sql_error alter database db fsync 4
+sql alter database db fsync 3600
+sql alter database db fsync 18000
+sql alter database db fsync 180000
+sql_error alter database db fsync 180001
 sql_error alter database db fsync -1
-sql_error alter database db fsync 1000
 
 print ============== step comp
 sql show databases