hotfix/TD-4675

src/mnode/src/mnodeSdb.c

@@ -663,10 +663,17 @@ static int32_t sdbProcessWrite(void *wparam, void *hparam, int32_t qtype, void *
                pTable->name, actStr[action], sdbGetKeyStr(pTable, pHead->cont), qtype, pHead->version, tsSdbMgmt.version);
       return TSDB_CODE_SUCCESS;
     } else if (pHead->version != tsSdbMgmt.version + 1) {
+      if (qtype != TAOS_QTYPE_WAL) {
         pthread_mutex_unlock(&tsSdbMgmt.mutex);
-      sdbError("vgId:1, sdb:%s, failed to restore %s key:%s from source(%d), hver:%" PRIu64 " too large, mver:%" PRIu64,
+        sdbError(
+            "vgId:1, sdb:%s, failed to restore %s key:%s from source(%d), hver:%" PRIu64 " too large, mver:%" PRIu64,
             pTable->name, actStr[action], sdbGetKeyStr(pTable, pHead->cont), qtype, pHead->version, tsSdbMgmt.version);
         return TSDB_CODE_SYN_INVALID_VERSION;
+      } else {
+        // If cksum is wrong when recovering wal, use this code
+        tsSdbMgmt.version = pHead->version;
+      }
+
     } else {
       tsSdbMgmt.version = pHead->version;
     }

src/wal/src/walWrite.c

@@ -346,7 +346,7 @@ static int32_t walRestoreWalFile(SWal *pWal, void *pVnode, FWalWrite writeFp, ch
     }
 
 #if defined(WAL_CHECKSUM_WHOLE)
-    if (pHead->sver == 0 && !walValidateChecksum(pHead)) {
+    if ((pHead->sver == 0 && !walValidateChecksum(pHead)) || pHead->sver < 0 || pHead->sver > 1) {
       wError("vgId:%d, file:%s, wal head cksum is messed up, hver:%" PRIu64 " len:%d offset:%" PRId64, pWal->vgId, name,
              pHead->version, pHead->len, offset);
       code = walSkipCorruptedRecord(pWal, pHead, tfd, &offset);