Merge pull request #28688 from taosdata/fix/main/TD-32824

Fix: alter wal level from 0 to 1or2 and restart failed

source/libs/wal/src/walMeta.c

@@ -424,9 +424,6 @@ static void printFileSet(int32_t vgId, SArray* fileSet, const char* str) {
 
 int32_t walCheckAndRepairMeta(SWal* pWal) {
   // load log files, get first/snapshot/last version info
-  if (pWal->cfg.level == TAOS_WAL_SKIP) {
-    return TSDB_CODE_SUCCESS;
-  }
   int32_t     code = 0;
   const char* logPattern = "^[0-9]+.log$";
   const char* idxPattern = "^[0-9]+.idx$";

source/libs/wal/src/walMgmt.c

@@ -90,6 +90,45 @@ static int32_t walInitLock(SWal *pWal) {
   return 0;
 }
 
+int32_t walInitWriteFileForSkip(SWal *pWal) {
+  TdFilePtr pIdxTFile = NULL, pLogTFile = NULL;
+  int64_t   fileFirstVer = 0;
+  int32_t   code = 0;
+
+  char fnameStr[WAL_FILE_LEN];
+  walBuildIdxName(pWal, fileFirstVer, fnameStr);
+  pIdxTFile = taosOpenFile(fnameStr, TD_FILE_CREATE | TD_FILE_WRITE | TD_FILE_APPEND);
+  if (pIdxTFile == NULL) {
+    wError("vgId:%d, failed to open file since %s", pWal->cfg.vgId, tstrerror(terrno));
+    code = terrno;
+    goto _exit;
+  }
+  walBuildLogName(pWal, fileFirstVer, fnameStr);
+  pLogTFile = taosOpenFile(fnameStr, TD_FILE_CREATE | TD_FILE_WRITE | TD_FILE_APPEND);
+  if (pLogTFile == NULL) {
+    wError("vgId:%d, failed to open file since %s", pWal->cfg.vgId, tstrerror(terrno));
+    code = terrno;
+    goto _exit;
+  }
+  // switch file
+  pWal->pIdxFile = pIdxTFile;
+  pWal->pLogFile = pLogTFile;
+  SWalFileInfo fileInfo;
+  (void)memset(&fileInfo, -1, sizeof(SWalFileInfo));
+  if (!taosArrayPush(pWal->fileInfoSet, &fileInfo)) {
+    wError("vgId:%d, failed to push fileInfo into array since %s", pWal->cfg.vgId, tstrerror(terrno));
+    code = terrno;
+    goto _exit;
+  }
+  pWal->writeCur = 0;
+_exit:
+  if (code != TSDB_CODE_SUCCESS) {
+    (void)taosCloseFile(&pIdxTFile);
+    (void)taosCloseFile(&pLogTFile);
+  }
+  TAOS_RETURN(code);
+}
+
 SWal *walOpen(const char *path, SWalCfg *pCfg) {
   int32_t code = 0;
   SWal   *pWal = taosMemoryCalloc(1, sizeof(SWal));
@@ -165,17 +204,25 @@ SWal *walOpen(const char *path, SWalCfg *pCfg) {
   if (code < 0) {
     wWarn("vgId:%d, failed to load meta since %s", pWal->cfg.vgId, tstrerror(code));
   }
+  if (pWal->cfg.level != TAOS_WAL_SKIP) {
+    code = walCheckAndRepairMeta(pWal);
+    if (code < 0) {
+      wError("vgId:%d, cannot open wal since repair meta file failed since %s", pWal->cfg.vgId, tstrerror(code));
+      goto _err;
+    }
 
-  code = walCheckAndRepairMeta(pWal);
+    code = walCheckAndRepairIdx(pWal);
-  if (code < 0) {
+    if (code < 0) {
-    wError("vgId:%d, cannot open wal since repair meta file failed since %s", pWal->cfg.vgId, tstrerror(code));
+      wError("vgId:%d, cannot open wal since repair idx file failed since %s", pWal->cfg.vgId, tstrerror(code));
-    goto _err;
+      goto _err;
-  }
+    }
-
+  } else {
-  code = walCheckAndRepairIdx(pWal);
+    code = walInitWriteFileForSkip(pWal);
-  if (code < 0) {
+    if (code < 0) {
-    wError("vgId:%d, cannot open wal since repair idx file failed since %s", pWal->cfg.vgId, tstrerror(code));
+      wError("vgId:%d, cannot open wal since init write file for wal_level = 0 failed since %s", pWal->cfg.vgId,
-    goto _err;
+             tstrerror(code));
+      goto _err;
+    }
   }
 
   // add ref
@@ -217,6 +264,14 @@ int32_t walAlter(SWal *pWal, SWalCfg *pCfg) {
         pWal->cfg.vgId, pWal->cfg.level, pWal->cfg.fsyncPeriod, pWal->cfg.retentionPeriod, pWal->cfg.retentionSize,
         pCfg->level, pCfg->fsyncPeriod, pCfg->retentionPeriod, pCfg->retentionSize);
 
+  if (pWal->cfg.level == TAOS_WAL_SKIP && pCfg->level != TAOS_WAL_SKIP) {
+    wInfo("vgId:%d, remove all wals, path:%s", pWal->cfg.vgId, pWal->path);
+    taosRemoveDir(pWal->path);
+    if (taosMkDir(pWal->path) != 0) {
+      wError("vgId:%d, path:%s, failed to create directory since %s", pWal->cfg.vgId, pWal->path, tstrerror(terrno));
+    }
+  }
+
   pWal->cfg.level = pCfg->level;
   pWal->cfg.fsyncPeriod = pCfg->fsyncPeriod;
   pWal->cfg.retentionPeriod = pCfg->retentionPeriod;

tests/parallel_test/cases.task

@@ -361,6 +361,7 @@
 ,,n,system-test,python3 ./test.py -f 0-others/tag_index_basic.py
 ,,n,system-test,python3 ./test.py -f 0-others/udfpy_main.py
 ,,n,system-test,python3 ./test.py -N 3 -f 0-others/walRetention.py
+,,n,system-test,python3 ./test.py -f 0-others/wal_level_skip.py
 ,,y,system-test,./pytest.sh python3 ./test.py -f 0-others/splitVGroup.py -N 3 -n 1
 ,,y,system-test,./pytest.sh python3 ./test.py -f 0-others/splitVGroupWal.py -N 3 -n 1
 ,,y,system-test,./pytest.sh python3 ./test.py -f 0-others/splitVGroup.py -N 3 -n 3

tests/pytest/util/dnodes.py

@@ -594,8 +594,7 @@ class TDDnode:
 
     def forcestop(self):
         if self.asan:
-            stopCmd = "%s -s stop -n dnode%d -x SIGKILL" + \
+            stopCmd = "%s -s stop -n dnode%d -x SIGKILL" % (self.execPath, self.index)
-                (self.execPath, self.index)
             tdLog.info("execute script: " + stopCmd)
             os.system(stopCmd)
             return

tests/system-test/0-others/wal_level_skip.py

@@ -0,0 +1,170 @@
+import sys
+import taos
+import os
+from util.log import *
+from util.cases import *
+from util.sql import *
+from util.dnodes import *
+
+class TDTestCase:
+
+    def init(self, conn, logSql,replicaVar=1):
+        tdLog.debug(f"start to excute {__file__}")
+        tdSql.init(conn.cursor())
+
+    def getBuildPath(self):
+        selfPath = os.path.dirname(os.path.realpath(__file__))
+
+        if ("community" in selfPath):
+            projPath = selfPath[:selfPath.find("community")]
+        else:
+            projPath = selfPath[:selfPath.find("tests")]
+
+        for root, dirs, files in os.walk(projPath):
+            if ("taosd" in files or "taosd.exe" in files):
+                rootRealPath = os.path.dirname(os.path.realpath(root))
+                if ("packaging" not in rootRealPath):
+                    buildPath = root[:len(root) - len("/build/bin")]
+                    break
+        return buildPath
+
+    def preData(self):
+        tdSql.execute("drop database if exists db0;")
+        tdSql.execute("create database db0 KEEP 30 vgroups 1 buffer 3 wal_level 0;")
+        tdSql.execute("create table if not exists db0.stb (ts timestamp, c1 int, c2 float, c3 double) tags (t1 int unsigned);")
+        tdSql.execute("create table db0.ct1 using db0.stb tags(1000);")
+        tdSql.execute("create table db0.ct2 using db0.stb tags(2000);")
+        tdSql.execute("create table if not exists db0.ntb (ts timestamp, c1 int, c2 float, c3 double) ;")
+        tdSql.query("show db0.stables;")
+        tdSql.execute("insert into db0.ct1 values(now+0s, 10, 2.0, 3.0);")
+        tdSql.execute("insert into db0.ct1 values(now+1s, 11, 2.1, 3.1)(now+2s, 12, 2.2, 3.2)(now+3s, 13, 2.3, 3.3);")
+        tdSql.execute("insert into db0.ntb values(now+2s, 10, 2.0, 3.0);")
+
+    def insertData(self):
+        tdSql.execute("insert into db0.ct1 values(now+0s, 10, 2.0, 3.0);")
+        tdSql.execute("insert into db0.ct1 values(now+1s, 11, 2.1, 3.1)(now+2s, 12, 2.2, 3.2)(now+3s, 13, 2.3, 3.3);")
+        tdSql.execute("insert into db0.ntb values(now+2s, 10, 2.0, 3.0);")
+
+    def createSubTableAndInsertData(self):
+        tdSql.execute("create table db0.ct1 using db0.stb tags(1000);")
+        tdSql.execute("create table db0.ct2 using db0.stb tags(2000);")
+        tdSql.execute("create table if not exists db0.ntb (ts timestamp, c1 int, c2 float, c3 double) ;")
+        tdSql.execute("insert into db0.ct1 values(now+0s, 10, 2.0, 3.0);")
+        tdSql.execute("insert into db0.ct1 values(now+1s, 11, 2.1, 3.1)(now+2s, 12, 2.2, 3.2)(now+3s, 13, 2.3, 3.3);")
+        tdSql.execute("insert into db0.ntb values(now+2s, 10, 2.0, 3.0);")
+
+
+    def alterWalLevel(self,level):
+        tdSql.execute("alter database db0 wal_level %d;"%level)
+
+    def run(self):  
+        tdSql.prepare()
+
+        tdLog.info("-----------test for stop taosd before alter wal level-----------")
+        tdLog.info("create database wal_level = 0 and insert data")
+        self.preData()
+        tdDnodes.stop(1)
+        time.sleep(2)
+        tdLog.info("restart taosd")
+        tdDnodes.start(1)
+
+        tdLog.info(" alter wal level from 0 to 1")
+        self.alterWalLevel(1)
+        self.insertData()
+        tdDnodes.stop(1)
+        tdDnodes.start(1)
+
+        tdLog.info("create database wal_level = 0 and insert data")
+        self.preData()
+        tdDnodes.stop(1)
+        time.sleep(2)
+        tdLog.info("restart taosd")
+        tdDnodes.start(1)
+
+        tdLog.info(" alter wal level from 0 to 2")
+        self.alterWalLevel(2)
+        self.insertData()
+        tdDnodes.forcestop(1)
+        tdDnodes.start(1)
+
+
+        tdLog.info("-----------test for kill taosd before alter wal level-----------")
+        tdLog.info("create database wal_level = 0 and insert data")
+        self.preData()
+        tdDnodes.forcestop(1)
+        time.sleep(2)
+        tdLog.info("restart taosd")
+        tdDnodes.start(1)
+
+        tdLog.info(" alter wal level from 0 to 1")
+        self.alterWalLevel(1)
+        tdDnodes.forcestop(1)
+        tdDnodes.start(1)
+
+        tdLog.info("create database wal_level = 0 and insert data")
+        self.preData()
+        tdDnodes.forcestop(1)
+        time.sleep(2)
+        tdLog.info("restart taosd")
+        tdDnodes.start(1)
+
+        tdLog.info(" alter wal level from 0 to 2")
+        self.alterWalLevel(2)
+        tdDnodes.forcestop(1)
+        tdDnodes.start(1)
+
+        tdLog.info("-----------test for stop taosd after alter wal level-----------")
+        tdLog.info("create database wal_level = 0 and insert data")
+        self.preData()
+        tdLog.info(" alter wal level from 0 to 1")
+        self.alterWalLevel(1)
+        time.sleep(1)
+        self.insertData()
+        tdDnodes.stop(1)
+        time.sleep(2)
+        tdLog.info("restart taosd")
+        tdDnodes.start(1)
+
+
+        tdLog.info("create database wal_level = 0 and insert data")
+        self.preData()
+        tdLog.info(" alter wal level from 0 to 2")
+        self.alterWalLevel(2)
+        time.sleep(1)
+        self.insertData()
+        tdDnodes.stop(1)
+        time.sleep(2)
+        tdLog.info("restart taosd")
+        tdDnodes.start(1)
+
+        tdLog.info("-----------test for kill taosd after alter wal level-----------")
+        tdLog.info("create database wal_level = 0 and insert data")
+        self.preData()
+        tdLog.info(" alter wal level from 0 to 1")
+        self.alterWalLevel(1)
+        time.sleep(1)
+        self.insertData()
+        tdDnodes.forcestop(1)
+        time.sleep(2)
+        tdLog.info("restart taosd")
+        tdDnodes.start(1)
+
+
+        tdLog.info("create database wal_level = 0 and insert data")
+        self.preData()
+        tdLog.info(" alter wal level from 0 to 2")
+        self.alterWalLevel(2)
+        time.sleep(1)
+        self.insertData()
+        tdDnodes.forcestop(1)
+        time.sleep(2)
+        tdLog.info("restart taosd")
+        tdDnodes.start(1)
+
+        
+    def stop(self):
+        tdSql.close()
+        tdLog.success(f"{__file__} successfully executed")
+
+tdCases.addLinux(__file__, TDTestCase())
+tdCases.addWindows(__file__, TDTestCase())