Merge pull request #2004 from taosdata/feature/dropdnode

[TD-388] drop dnode in community version

src/inc/taosmsg.h

@@ -725,6 +725,8 @@ typedef struct {
 typedef struct {
   uint32_t  queryId;
   uint32_t  streamId;
+  uint32_t  totalDnodes;
+  uint32_t  onlineDnodes;
   int8_t    killConnection;
   SRpcIpSet ipList;
 } SCMHeartBeatRsp;

src/mnode/inc/mgmtDef.h

@@ -66,7 +66,6 @@ typedef struct SMnodeObj {
   SDnodeObj *pDnode;
 } SMnodeObj;
 
-// todo use dynamic length string
 typedef struct {
   char  *tableId;
   int8_t type;

src/mnode/inc/mgmtDnode.h

@@ -34,6 +34,7 @@ char*   mgmtGetDnodeStatusStr(int32_t dnodeStatus);
 void    mgmtMonitorDnodeModule();
 
 int32_t mgmtGetDnodesNum();
+int32_t mgmtGetOnlinDnodesNum();
 void *  mgmtGetNextDnode(void *pIter, SDnodeObj **pDnode);
 void    mgmtIncDnodeRef(SDnodeObj *pDnode);
 void    mgmtDecDnodeRef(SDnodeObj *pDnode);

src/mnode/inc/mgmtTable.h

@@ -31,6 +31,7 @@ void *  mgmtGetNextChildTable(void *pIter, SChildTableObj **pTable);
 void *  mgmtGetNextSuperTable(void *pIter, SSuperTableObj **pTable);
 void    mgmtDropAllChildTables(SDbObj *pDropDb);
 void    mgmtDropAllSuperTables(SDbObj *pDropDb);
+void    mgmtDropAllChildTablesInVgroups(SVgObj *pVgroup);
 
 #ifdef __cplusplus
 }

src/mnode/src/mgmtDnode.c

@@ -74,7 +74,6 @@ static int32_t mgmtDnodeActionDelete(SSdbOper *pOper) {
   SDnodeObj *pDnode = pOper->pObj;
  
 #ifndef _SYNC 
-  //TODO: drop dnode local
   mgmtDropAllDnodeVgroups(pDnode);
 #endif  
   mgmtDropMnodeLocal(pDnode->dnodeId);
@@ -179,6 +178,23 @@ int32_t mgmtGetDnodesNum() {
   return sdbGetNumOfRows(tsDnodeSdb);
 }
 
+int32_t mgmtGetOnlinDnodesNum(char *ep) {
+  SDnodeObj *pDnode = NULL;
+  void *     pIter = NULL;
+  int32_t    onlineDnodes = 0;
+
+  while (1) {
+    pIter = mgmtGetNextDnode(pIter, &pDnode);
+    if (pDnode == NULL) break;
+    if (pDnode->status != TAOS_DN_STATUS_OFFLINE) onlineDnodes++;
+    mgmtDecDnodeRef(pDnode);
+  }
+
+  sdbFreeIter(pIter);
+
+  return onlineDnodes;
+}
+
 void *mgmtGetDnode(int32_t dnodeId) {
   return sdbGetRow(tsDnodeSdb, &dnodeId);
 }
@@ -397,7 +413,6 @@ static int32_t mgmtCreateDnode(char *ep) {
   return code;
 }
 
-//TODO drop others tables
 int32_t mgmtDropDnode(SDnodeObj *pDnode) {
   SSdbOper oper = {
     .type = SDB_OPER_GLOBAL,
@@ -410,7 +425,7 @@ int32_t mgmtDropDnode(SDnodeObj *pDnode) {
     code = TSDB_CODE_SDB_ERROR;
   }
 
-  mLPrint("dnode:%d is dropped from cluster, result:%s", pDnode->dnodeId, tstrerror(code));
+  mLPrint("dnode:%d, is dropped from cluster, result:%s", pDnode->dnodeId, tstrerror(code));
   return code;
 }
 

src/mnode/src/mgmtSdb.c

@@ -353,8 +353,8 @@ void sdbIncRef(void *handle, void *pObj) {
   SSdbTable *pTable = handle;
   int32_t *  pRefCount = (int32_t *)(pObj + pTable->refCountPos);
   atomic_add_fetch_32(pRefCount, 1);
-  if (0 && (pTable->tableId == SDB_TABLE_MNODE || pTable->tableId == SDB_TABLE_DNODE)) {
-    sdbTrace("table:%s, add ref to record:%s:%d", pTable->tableName, sdbGetKeyStrFromObj(pTable, pObj), *pRefCount);
+  if (1 && (pTable->tableId == SDB_TABLE_MNODE || pTable->tableId == SDB_TABLE_DNODE)) {
+    sdbTrace("add ref to table:%s record:%s:%d", pTable->tableName, sdbGetKeyStrFromObj(pTable, pObj), *pRefCount);
   }
 }
 
@@ -364,8 +364,8 @@ void sdbDecRef(void *handle, void *pObj) {
   SSdbTable *pTable = handle;
   int32_t *  pRefCount = (int32_t *)(pObj + pTable->refCountPos);
   int32_t    refCount = atomic_sub_fetch_32(pRefCount, 1);
-  if (0 && (pTable->tableId == SDB_TABLE_MNODE || pTable->tableId == SDB_TABLE_DNODE)) {
-    sdbTrace("table:%s, def ref of record:%s:%d", pTable->tableName, sdbGetKeyStrFromObj(pTable, pObj), *pRefCount);
+  if (1 && (pTable->tableId == SDB_TABLE_MNODE || pTable->tableId == SDB_TABLE_DNODE)) {
+    sdbTrace("def ref of table:%s record:%s:%d", pTable->tableName, sdbGetKeyStrFromObj(pTable, pObj), *pRefCount);
   }
 
   int8_t *updateEnd = pObj + pTable->refCountPos - 1;

src/mnode/src/mgmtShell.c

@@ -325,6 +325,8 @@ static void mgmtProcessHeartBeatMsg(SQueuedMsg *pMsg) {
     return;
   }
 
+  pHBRsp->onlineDnodes = htonl(mgmtGetOnlinDnodesNum());
+  pHBRsp->totalDnodes = htonl(mgmtGetDnodesNum());
   mgmtGetMnodeIpSet(&pHBRsp->ipList);
   
   /*

src/mnode/src/mgmtTable.c

@@ -1305,7 +1305,7 @@ static void mgmtProcessSuperTableVgroupMsg(SQueuedMsg *pMsg) {
         if (pDnode == NULL) break;
 
         strncpy(pVgroupInfo->vgroups[vgSize].ipAddr[vn].fqdn, pDnode->dnodeFqdn, tListLen(pDnode->dnodeFqdn));
-        pVgroupInfo->vgroups[vgSize].ipAddr[vn].port = htons(tsDnodeShellPort);
+        pVgroupInfo->vgroups[vgSize].ipAddr[vn].port = htons(pDnode->dnodePort);
 
         pVgroupInfo->vgroups[vgSize].numOfIps++;
       }
@@ -1787,6 +1787,34 @@ static void mgmtGetChildTableMeta(SQueuedMsg *pMsg) {
   rpcSendResponse(&rpcRsp);
 }
 
+void mgmtDropAllChildTablesInVgroups(SVgObj *pVgroup) {
+  void *  pIter = NULL;
+  int32_t numOfTables = 0;
+  SChildTableObj *pTable = NULL;
+
+  mPrint("vgId:%d, all child tables will be dropped from sdb", pVgroup->vgId);
+
+  while (1) {
+    pIter = mgmtGetNextChildTable(pIter, &pTable);
+    if (pTable == NULL) break;
+
+    if (pTable->vgId == pVgroup->vgId) {
+      SSdbOper oper = {
+        .type = SDB_OPER_LOCAL,
+        .table = tsChildTableSdb,
+        .pObj = pTable,
+      };
+      sdbDeleteRow(&oper);
+      numOfTables++;
+    }
+    mgmtDecTableRef(pTable);
+  }
+
+  sdbFreeIter(pIter);
+
+  mPrint("vgId:%d, all child tables is dropped from sdb", pVgroup->vgId);
+}
+
 void mgmtDropAllChildTables(SDbObj *pDropDb) {
   void *  pIter = NULL;
   int32_t numOfTables = 0;

src/mnode/src/mgmtVgroup.c

@@ -747,11 +747,14 @@ void mgmtDropAllDnodeVgroups(SDnodeObj *pDropDnode) {
   SVgObj *pVgroup = NULL;
   int32_t numOfVgroups = 0;
 
+  mPrint("dnode:%d, all vgroups will be dropped from sdb", pDropDnode->dnodeId);
+
   while (1) {
     pIter = mgmtGetNextVgroup(pIter, &pVgroup);
     if (pVgroup == NULL) break;
 
     if (pVgroup->vnodeGid[0].dnodeId == pDropDnode->dnodeId) {
+      mgmtDropAllChildTablesInVgroups(pVgroup);
       SSdbOper oper = {
         .type = SDB_OPER_LOCAL,
         .table = tsVgroupSdb,
@@ -759,12 +762,13 @@ void mgmtDropAllDnodeVgroups(SDnodeObj *pDropDnode) {
       };
       sdbDeleteRow(&oper);
       numOfVgroups++;
-      continue;
     }
     mgmtDecVgroupRef(pVgroup);
   }
 
   sdbFreeIter(pIter);
+
+  mPrint("dnode:%d, all vgroups is dropped from sdb", pDropDnode->dnodeId);
 }
 
 void mgmtUpdateAllDbVgroups(SDbObj *pAlterDb) {

tests/script/general/db/dropdnodes.sim

+system sh/stop_dnodes.sh
+
+system sh/deploy.sh -n dnode1 -i 1
+system sh/cfg.sh -n dnode1 -c walLevel            -v 2
+system sh/deploy.sh -n dnode2 -i 2
+system sh/cfg.sh -n dnode1 -c walLevel            -v 2
+
+print ==========  prepare data
+system sh/exec.sh -n dnode1 -s start
+system sh/exec.sh -n dnode2 -s start
+
+sleep 3000
+sql connect 
+
+sql create dnode $hostname2
+sleep 2000
+
+sql create database db maxTables 4
+sql use db
+
+print ==========  step1
+sql create table mt (ts timestamp, tbcol int) TAGS(tgcol int)
+sql create table db.t1 using db.mt tags(1)
+sql create table db.t2 using db.mt tags(2)
+sql create table db.t3 using db.mt tags(3)
+sql create table db.t4 using db.mt tags(4)
+sql create table db.t5 using db.mt tags(5)
+sql create table db.t6 using db.mt tags(6)
+sql create table db.t7 using db.mt tags(7)
+sql create table db.t8 using db.mt tags(8)
+sql create table db.t9 using db.mt tags(9)
+sql create table db.t10 using db.mt tags(10)
+sql create table db.t11 using db.mt tags(11)
+sql create table db.t12 using db.mt tags(12)
+sql create table db.t13 using db.mt tags(13)
+sql create table db.t14 using db.mt tags(14)
+sql create table db.t15 using db.mt tags(15)
+sql create table db.t16 using db.mt tags(16)
+
+sql insert into db.t1 values(now, 1)
+sql insert into db.t2 values(now, 1)
+sql insert into db.t3 values(now, 1)
+sql insert into db.t4 values(now, 1)
+sql insert into db.t5 values(now, 1)
+sql insert into db.t6 values(now, 1)
+sql insert into db.t7 values(now, 1)
+sql insert into db.t8 values(now, 1)
+sql insert into db.t9 values(now, 1)
+sql insert into db.t10 values(now, 1)
+sql insert into db.t11 values(now, 1)
+sql insert into db.t12 values(now, 1)
+sql insert into db.t13 values(now, 1)
+sql insert into db.t14 values(now, 1)
+sql insert into db.t15 values(now, 1)
+sql insert into db.t16 values(now, 1)
+
+print ========== step2
+sql show tables
+print $rows
+if $rows != 16 then
+  return -1
+endi  
+
+sql select * from mt
+print $rows
+if $rows != 16 then
+  return -1
+endi  
+
+print ========== step3
+
+system sh/exec.sh -n dnode2 -s stop -x SIGINT
+sleep 500
+sql drop dnode $hostname2
+sleep 2000
+
+print ========== step3
+
+sql show tables
+print $rows
+if $rows != 8 then
+  return -1
+endi  
+
+sql select * from mt
+print $rows
+if $rows != 8 then
+  return -1
+endi  
+
+sql select * from db.t5
+if $rows != 1 then
+  return -1
+endi  
+
+sql select * from db.t13
+if $rows != 1 then
+  return -1
+endi  
+
+sql_error select * from db.t1
+sql_error select * from db.t9
+
+system sh/exec.sh -n dnode1 -s stop -x SIGINT
\ No newline at end of file

tests/script/jenkins/basic.txt

@@ -61,6 +61,7 @@ cd ../../../debug; make
 ./test.sh -f general/db/delete_writing1.sim
 ./test.sh -f general/db/delete_writing2.sim
 ./test.sh -f general/db/delete.sim
+./test.sh -f general/db/dropdnodes.sim
 ./test.sh -f general/db/len.sim
 ./test.sh -f general/db/repeat.sim
 ./test.sh -f general/db/tables.sim

tests/script/jenkins/sync.txt

+cd ../../debug; cmake ..
+cd ../../debug; make
+cd ../../../debug; cmake ..
+cd ../../../debug; make
+
+./test.sh -u -f unique/cluster/balance1.sim
+./test.sh -u -f unique/cluster/balance2.sim
+./test.sh -u -f unique/cluster/balance3.sim
+./test.sh -u -f unique/cluster/cache.sim
+
+./test.sh -u -f unique/dnode/balance1.sim
+./test.sh -u -f unique/dnode/balance2.sim
+./test.sh -u -f unique/dnode/balance3.sim
+./test.sh -u -f unique/dnode/balancex.sim
+
+./test.sh -u -f unique/mnode/mgmt22.sim
+./test.sh -u -f unique/mnode/mgmt23.sim
+./test.sh -u -f unique/mnode/mgmt24.sim
+./test.sh -u -f unique/mnode/mgmt25.sim
+./test.sh -u -f unique/mnode/mgmt26.sim
+./test.sh -u -f unique/mnode/mgmt33.sim
+./test.sh -u -f unique/mnode/mgmt34.sim
+./test.sh -u -f unique/mnode/mgmtr2.sim