Merge branch '3.0' into fix/3_liaohj

include/common/tmsg.h

@@ -2361,12 +2361,19 @@ typedef struct {
 } SMArbUpdateGroupMember;
 
 typedef struct {
-  int32_t                vgId;
+  int32_t dnodeId;
-  int64_t                dbUid;
+  char*   token;
-  SMArbUpdateGroupMember members[2];
+  int8_t  acked;
-  int8_t                 isSync;
+} SMArbUpdateGroupAssigned;
-  SMArbUpdateGroupMember assignedLeader;
+
-  int64_t                version;
+typedef struct {
+  int32_t                  vgId;
+  int64_t                  dbUid;
+  SMArbUpdateGroupMember   members[2];
+  int8_t                   isSync;
+  int8_t                   assignedAcked;
+  SMArbUpdateGroupAssigned assignedLeader;
+  int64_t                  version;
 } SMArbUpdateGroup;
 
 typedef struct {

include/dnode/mnode/mnode.h

@@ -101,6 +101,8 @@ int32_t mndGetMonitorInfo(SMnode *pMnode, SMonClusterInfo *pClusterInfo, SMonVgr
  */
 int32_t mndGetLoad(SMnode *pMnode, SMnodeLoad *pLoad);
 
+int64_t mndGetRoleTimeMs(SMnode *pMnode);
+
 /**
  * @brief Process the rpc, sync request.
  *

include/libs/nodes/plannodes.h

@@ -183,6 +183,7 @@ typedef struct SProjectLogicNode {
   char       stmtName[TSDB_TABLE_NAME_LEN];
   bool       ignoreGroupId;
   bool       inputIgnoreGroup;
+  bool       isSetOpProj;
 } SProjectLogicNode;
 
 typedef struct SIndefRowsFuncLogicNode {

source/common/src/systable.c

@@ -76,6 +76,7 @@ static const SSysDbTableSchema arbGroupsSchema[] = {
     {.name = "is_sync", .bytes = 2, .type = TSDB_DATA_TYPE_SMALLINT, .sysInfo = true},
     {.name = "assigned_dnode", .bytes = 2, .type = TSDB_DATA_TYPE_SMALLINT, .sysInfo = true},
     {.name = "assigned_token", .bytes = TSDB_ARB_TOKEN_SIZE + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_VARCHAR, .sysInfo = true},
+    {.name = "assigned_acked", .bytes = 2, .type = TSDB_DATA_TYPE_SMALLINT, .sysInfo = true},
 };
 
 static const SSysDbTableSchema clusterSchema[] = {

source/common/src/tmsg.c

@@ -6460,6 +6460,11 @@ int32_t tSerializeSMArbUpdateGroupBatchReq(void *buf, int32_t bufLen, SMArbUpdat
     if (tEncodeI64(&encoder, pGroup->version) < 0) return -1;
   }
 
+  for (int32_t i = 0; i < sz; i++) {
+    SMArbUpdateGroup *pGroup = taosArrayGet(pReq->updateArray, i);
+    if (tEncodeI8(&encoder, pGroup->assignedLeader.acked) < 0) return -1;
+  }
+
   tEndEncode(&encoder);
 
   int32_t tlen = encoder.pos;
@@ -6492,8 +6497,18 @@ int32_t tDeserializeSMArbUpdateGroupBatchReq(void *buf, int32_t bufLen, SMArbUpd
     group.assignedLeader.token = taosMemoryMalloc(TSDB_ARB_TOKEN_SIZE);
     if (tDecodeCStrTo(&decoder, group.assignedLeader.token) < 0) return -1;
     if (tDecodeI64(&decoder, &group.version) < 0) return -1;
+    group.assignedLeader.acked = false;
+
     taosArrayPush(updateArray, &group);
   }
+
+  if (!tDecodeIsEnd(&decoder)) {
+    for (int32_t i = 0; i < sz; i++) {
+      SMArbUpdateGroup *pGroup = taosArrayGet(updateArray, i);
+      if (tDecodeI8(&decoder, &pGroup->assignedLeader.acked) < 0) return -1;
+    }
+  }
+
   pReq->updateArray = updateArray;
 
   tEndDecode(&decoder);

source/dnode/mnode/impl/inc/mndDef.h

@@ -255,6 +255,7 @@ typedef struct {
 typedef struct {
   int32_t dnodeId;
   char    token[TSDB_ARB_TOKEN_SIZE];
+  int8_t  acked;
 } SArbAssignedLeader;
 
 typedef struct {

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

@@ -25,7 +25,7 @@
 #include "mndVgroup.h"
 
 #define ARBGROUP_VER_NUMBER   1
-#define ARBGROUP_RESERVE_SIZE 64
+#define ARBGROUP_RESERVE_SIZE 63
 
 static SHashObj *arbUpdateHash = NULL;
 
@@ -129,6 +129,7 @@ SSdbRaw *mndArbGroupActionEncode(SArbGroup *pGroup) {
   SDB_SET_INT32(pRaw, dataPos, pLeader->dnodeId, _OVER)
   SDB_SET_BINARY(pRaw, dataPos, pLeader->token, TSDB_ARB_TOKEN_SIZE, _OVER)
   SDB_SET_INT64(pRaw, dataPos, pGroup->version, _OVER)
+  SDB_SET_INT8(pRaw, dataPos, pLeader->acked, _OVER)
 
   SDB_SET_RESERVE(pRaw, dataPos, ARBGROUP_RESERVE_SIZE, _OVER)
 
@@ -182,6 +183,7 @@ SSdbRow *mndArbGroupActionDecode(SSdbRaw *pRaw) {
   SDB_GET_INT32(pRaw, dataPos, &pLeader->dnodeId, _OVER)
   SDB_GET_BINARY(pRaw, dataPos, pLeader->token, TSDB_ARB_TOKEN_SIZE, _OVER)
   SDB_GET_INT64(pRaw, dataPos, &pGroup->version, _OVER)
+  SDB_GET_INT8(pRaw, dataPos, &pLeader->acked, _OVER)
 
   pGroup->mutexInited = false;
 
@@ -235,6 +237,7 @@ static int32_t mndArbGroupActionUpdate(SSdb *pSdb, SArbGroup *pOld, SArbGroup *p
   pOld->isSync = pNew->isSync;
   pOld->assignedLeader.dnodeId = pNew->assignedLeader.dnodeId;
   memcpy(pOld->assignedLeader.token, pNew->assignedLeader.token, TSDB_ARB_TOKEN_SIZE);
+  pOld->assignedLeader.acked = pNew->assignedLeader.acked;
   pOld->version++;
 
 _OVER:
@@ -540,6 +543,14 @@ static int32_t mndProcessArbCheckSyncTimer(SRpcMsg *pReq) {
     return -1;
   }
 
+  int64_t roleTimeMs = mndGetRoleTimeMs(pMnode);
+  int64_t nowMs = taosGetTimestampMs();
+  if (nowMs - roleTimeMs < tsArbHeartBeatIntervalSec * 1000 * 2) {
+    mInfo("arb skip to check sync since mnd had just switch over, roleTime:%" PRId64 " now:%" PRId64, roleTimeMs,
+          nowMs);
+    return 0;
+  }
+
   SArray *pUpdateArray = taosArrayInit(16, sizeof(SArbGroup));
 
   while (1) {
@@ -551,15 +562,14 @@ static int32_t mndProcessArbCheckSyncTimer(SRpcMsg *pReq) {
     taosThreadMutexUnlock(&pArbGroup->mutex);
 
     int32_t vgId = arbGroupDup.vgId;
-    int64_t nowMs = taosGetTimestampMs();
 
     bool                member0IsTimeout = mndCheckArbMemberHbTimeout(&arbGroupDup, 0, nowMs);
     bool                member1IsTimeout = mndCheckArbMemberHbTimeout(&arbGroupDup, 1, nowMs);
     SArbAssignedLeader *pAssignedLeader = &arbGroupDup.assignedLeader;
     int32_t             currentAssignedDnodeId = pAssignedLeader->dnodeId;
 
-    // 1. has assigned && is sync => send req
+    // 1. has assigned && is sync && no response => send req
-    if (currentAssignedDnodeId != 0 && arbGroupDup.isSync == true) {
+    if (currentAssignedDnodeId != 0 && arbGroupDup.isSync == true && pAssignedLeader->acked == false) {
       (void)mndSendArbSetAssignedLeaderReq(pMnode, currentAssignedDnodeId, vgId, arbToken, term,
                                            pAssignedLeader->token);
       mInfo("vgId:%d, arb send set assigned leader to dnodeId:%d", vgId, currentAssignedDnodeId);
@@ -651,6 +661,7 @@ static void mndInitArbUpdateGroup(SArbGroup *pGroup, SMArbUpdateGroup *outGroup)
   outGroup->isSync = pGroup->isSync;
   outGroup->assignedLeader.dnodeId = pGroup->assignedLeader.dnodeId;
   outGroup->assignedLeader.token = pGroup->assignedLeader.token;  // just copy the pointer
+  outGroup->assignedLeader.acked = pGroup->assignedLeader.acked;
   outGroup->version = pGroup->version;
 }
 
@@ -766,6 +777,7 @@ static int32_t mndProcessArbUpdateGroupBatchReq(SRpcMsg *pReq) {
     newGroup.isSync = pUpdateGroup->isSync;
     newGroup.assignedLeader.dnodeId = pUpdateGroup->assignedLeader.dnodeId;
     memcpy(newGroup.assignedLeader.token, pUpdateGroup->assignedLeader.token, TSDB_ARB_TOKEN_SIZE);
+    newGroup.assignedLeader.acked = pUpdateGroup->assignedLeader.acked;
     newGroup.version = pUpdateGroup->version;
 
     SArbGroup *pOldGroup = sdbAcquire(pMnode->pSdb, SDB_ARBGROUP, &newGroup.vgId);
@@ -783,10 +795,10 @@ static int32_t mndProcessArbUpdateGroupBatchReq(SRpcMsg *pReq) {
       goto _OVER;
     }
 
-    mInfo("trans:%d, used to update arbgroup:%d, member0:[%d][%s] member1:[%d][%s] isSync:%d assigned:[%d][%s]",
+    mInfo("trans:%d, used to update arbgroup:%d, member0:[%d][%s] member1:[%d][%s] isSync:%d assigned:[%d][%s][%d]",
           pTrans->id, newGroup.vgId, newGroup.members[0].info.dnodeId, newGroup.members[0].state.token,
           newGroup.members[1].info.dnodeId, newGroup.members[1].state.token, newGroup.isSync,
-          newGroup.assignedLeader.dnodeId, newGroup.assignedLeader.token);
+          newGroup.assignedLeader.dnodeId, newGroup.assignedLeader.token, newGroup.assignedLeader.acked);
 
     sdbRelease(pMnode->pSdb, pOldGroup);
   }
@@ -819,11 +831,13 @@ static void mndArbGroupSetAssignedLeader(SArbGroup *pGroup, int32_t index) {
 
   pGroup->assignedLeader.dnodeId = pMember->info.dnodeId;
   strncpy(pGroup->assignedLeader.token, pMember->state.token, TSDB_ARB_TOKEN_SIZE);
+  pGroup->assignedLeader.acked = false;
 }
 
 static void mndArbGroupResetAssignedLeader(SArbGroup *pGroup) {
   pGroup->assignedLeader.dnodeId = 0;
   memset(pGroup->assignedLeader.token, 0, TSDB_ARB_TOKEN_SIZE);
+  pGroup->assignedLeader.acked = false;
 }
 
 static int32_t mndArbGroupUpdateTrans(SMnode *pMnode, SArbGroup *pNew) {
@@ -834,10 +848,10 @@ static int32_t mndArbGroupUpdateTrans(SMnode *pMnode, SArbGroup *pNew) {
     goto _OVER;
   }
 
-  mInfo("trans:%d, used to update arbgroup:%d, member0:[%d][%s] member1:[%d][%s] isSync:%d assigned:[%d][%s]",
+  mInfo("trans:%d, used to update arbgroup:%d, member0:[%d][%s] member1:[%d][%s] isSync:%d assigned:[%d][%s][%d]",
         pTrans->id, pNew->vgId, pNew->members[0].info.dnodeId, pNew->members[0].state.token,
         pNew->members[1].info.dnodeId, pNew->members[1].state.token, pNew->isSync, pNew->assignedLeader.dnodeId,
-        pNew->assignedLeader.token);
+        pNew->assignedLeader.token, pNew->assignedLeader.acked);
 
   mndTransAddArbGroupId(pTrans, pNew->vgId);
   if (mndTransCheckConflict(pMnode, pTrans) != 0) {
@@ -1103,11 +1117,12 @@ bool mndUpdateArbGroupBySetAssignedLeader(SArbGroup *pGroup, int32_t vgId, char
     goto _OVER;
   }
 
-  if (pGroup->isSync) {
+  if (pGroup->assignedLeader.acked == false) {
     mndArbGroupDupObj(pGroup, pNewGroup);
     pNewGroup->isSync = false;
+    pNewGroup->assignedLeader.acked = true;
 
-    mInfo("vgId:%d, arb isSync is setting to false", vgId);
+    mInfo("vgId:%d, arb received assigned ack", vgId);
     updateAssigned = true;
     goto _OVER;
   }
@@ -1217,12 +1232,18 @@ static int32_t mndRetrieveArbGroups(SRpcMsg *pReq, SShowObj *pShow, SSDataBlock
       STR_WITH_MAXSIZE_TO_VARSTR(token, pGroup->assignedLeader.token, TSDB_ARB_TOKEN_SIZE + VARSTR_HEADER_SIZE);
       pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
       colDataSetVal(pColInfo, numOfRows, (const char *)token, false);
+
+      pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
+      colDataSetVal(pColInfo, numOfRows, (const char *)&pGroup->assignedLeader.acked, false);
     } else {
       pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
       colDataSetNULL(pColInfo, numOfRows);
 
       pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
       colDataSetNULL(pColInfo, numOfRows);
+
+      pColInfo = taosArrayGet(pBlock->pDataBlock, cols++);
+      colDataSetNULL(pColInfo, numOfRows);
     }
 
     taosThreadMutexUnlock(&pGroup->mutex);

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

@@ -310,6 +310,8 @@ static int32_t minCronTime() {
   min = TMIN(min, tsStreamCheckpointInterval);
   min = TMIN(min, 6);  // checkpointRemain
   min = TMIN(min, tsStreamNodeCheckInterval);
+  min = TMIN(min, tsArbHeartBeatIntervalSec);
+  min = TMIN(min, tsArbCheckSyncIntervalSec);
 
   int64_t telemInt = TMIN(60, (tsTelemInterval - 1));
   min = TMIN(min, telemInt);
@@ -362,6 +364,18 @@ void mndDoTimerPullupTask(SMnode *pMnode, int64_t sec) {
   if (sec % tsUptimeInterval == 0) {
     mndIncreaseUpTime(pMnode);
   }
+
+  if (sec % (tsArbHeartBeatIntervalSec) == 0) {
+    if (mndPullupArbHeartbeat(pMnode) != 0) {
+      mError("failed to pullup arb heartbeat, since:%s", terrstr());
+    }
+  }
+
+  if (sec % (tsArbCheckSyncIntervalSec) == 0) {
+    if (mndPullupArbCheckSync(pMnode) != 0) {
+      mError("failed to pullup arb check sync, since:%s", terrstr());
+    }
+  }
 }
 void mndDoTimerCheckTask(SMnode *pMnode, int64_t sec) {
   if (sec % (tsStatusInterval * 5) == 0) {
@@ -394,18 +408,6 @@ static void *mndThreadFp(void *param) {
       continue;
     }
     mndDoTimerPullupTask(pMnode, sec);
-
-    if (sec % (tsArbHeartBeatIntervalSec) == 0) {
-      if (mndPullupArbHeartbeat(pMnode) != 0) {
-        mError("failed to pullup arb heartbeat, since:%s", terrstr());
-      }
-    }
-
-    if (sec % (tsArbCheckSyncIntervalSec) == 0) {
-      if (mndPullupArbCheckSync(pMnode) != 0) {
-        mError("failed to pullup arb check sync, since:%s", terrstr());
-      }
-    }
   }
 
   return NULL;
@@ -1048,6 +1050,11 @@ int32_t mndGetLoad(SMnode *pMnode, SMnodeLoad *pLoad) {
   return 0;
 }
 
+int64_t mndGetRoleTimeMs(SMnode *pMnode) {
+  SSyncState state = syncGetState(pMnode->syncMgmt.sync);
+  return state.roleTimeMs;
+}
+
 void mndSetRestored(SMnode *pMnode, bool restored) {
   if (restored) {
     taosThreadRwlockWrlock(&pMnode->lock);

source/libs/planner/src/planLogicCreater.c

@@ -1587,6 +1587,7 @@ static int32_t createSetOpProjectLogicNode(SLogicPlanContext* pCxt, SSetOperator
     TSWAP(pProject->node.pLimit, pSetOperator->pLimit);
   }
   pProject->ignoreGroupId = true;
+  pProject->isSetOpProj = true;
 
   int32_t code = TSDB_CODE_SUCCESS;
 

source/libs/planner/src/planOptimizer.c

@@ -3284,18 +3284,34 @@ static int32_t eliminateProjOptimizeImpl(SOptimizeContext* pCxt, SLogicSubplan*
   SNodeList*  pNewChildTargets = nodesMakeList();
 
   if (NULL == pProjectNode->node.pParent) {
-    SNode* pProjection = NULL;
+    SNode *pProjection = NULL, *pChildTarget = NULL;
-    FOREACH(pProjection, pProjectNode->pProjections) {
+    bool needOrderMatch = QUERY_NODE_LOGIC_PLAN_PROJECT == nodeType(pChild) && ((SProjectLogicNode*)pChild)->isSetOpProj;
-      SNode* pChildTarget = NULL;
+    bool orderMatch = true;
-      FOREACH(pChildTarget, pChild->pTargets) {
+    if (needOrderMatch) {
-        if (0 == strcmp(((SColumnNode*)pProjection)->colName, ((SColumnNode*)pChildTarget)->colName)) {
+      // For sql: select ... from (select ... union all select ...);
-          nodesListAppend(pNewChildTargets, nodesCloneNode(pChildTarget));
+      // When eliminating the outer proj (the outer select), we have to make sure that the outer proj projections and
+      // union all project targets have same columns in the same order. See detail in TD-30188
+      FORBOTH(pProjection, pProjectNode->pProjections, pChildTarget, pChild->pTargets) {
+        if (!pProjection) break;
+        if (0 != strcmp(((SColumnNode*)pProjection)->colName, ((SColumnNode*)pChildTarget)->colName)) {
+          orderMatch = false;
           break;
         }
+        nodesListAppend(pNewChildTargets, nodesCloneNode(pChildTarget));
+      }
+    } else {
+      FOREACH(pProjection, pProjectNode->pProjections) {
+        FOREACH(pChildTarget, pChild->pTargets) {
+          if (0 == strcmp(((SColumnNode*)pProjection)->colName, ((SColumnNode*)pChildTarget)->colName)) {
+            nodesListAppend(pNewChildTargets, nodesCloneNode(pChildTarget));
+            break;
+          }
+        }
       }
     }
 
-    if (eliminateProjOptCanChildConditionUseChildTargets(pChild, pNewChildTargets)) {
+    if (eliminateProjOptCanChildConditionUseChildTargets(pChild, pNewChildTargets) &&
+        (!needOrderMatch || (needOrderMatch && orderMatch))) {
       nodesDestroyList(pChild->pTargets);
       pChild->pTargets = pNewChildTargets;
     } else {

tests/system-test/2-query/union1.py

@@ -240,9 +240,22 @@ class TDTestCase:
         tdSql.error( " '' union all select c1 from ct1 " )
         # tdSql.error( "select c1 from ct1 union select c1 from ct2 union select c1 from ct4 ")
 
+    def test_select_from_union_all(self):
+        tdSql.query('select c8, ts from ((select ts, c8,c1 from stb1 order by c1) union all select ts, c8, c1 from stb1 limit 15)')
+        tdSql.checkRows(15)
+        tdSql.query('select c8, ts from ((select ts, c8,c1 from stb1 order by c1) union all (select ts, c8, c1 from stb1 order by c8 limit 10) limit 15)')
+        tdSql.checkRows(15)
+        tdSql.query('select ts, c1 from ((select ts, c8,c1 from stb1 order by c1) union all (select ts, c8, c1 from stb1 order by c8 limit 10) limit 15)')
+        tdSql.checkRows(15)
+        tdSql.query('select ts, c1, c8 from ((select ts, c8,c1 from stb1 order by c1) union all (select ts, c8, c1 from stb1 order by c8 limit 10) limit 15)')
+        tdSql.checkRows(15)
+        tdSql.query('select ts, c8, c1, 123 from ((select ts, c8,c1 from stb1 order by c1) union all (select ts, c8, c1 from stb1 order by c8 limit 10) limit 15)')
+        tdSql.checkRows(15)
+
     def all_test(self):
         self.__test_error()
         self.union_check()
+        self.test_select_from_union_all()
 
 
     def __create_tb(self):