提交 179709f8 编写于 作者: H Haojun Liao

fix(query): assign the subplan to belong to task struct

上级 79b0240a
......@@ -159,6 +159,13 @@ typedef struct {
int64_t recoverEndVer;
} SStreamTaskInfo;
typedef struct {
char* tablename;
char* dbname;
int32_t tversion;
SSchemaWrapper* sw;
} SSchemaInfo;
typedef struct SExecTaskInfo {
STaskIdInfo id;
uint32_t status;
......@@ -168,18 +175,12 @@ typedef struct SExecTaskInfo {
int32_t code;
SStreamTaskInfo streamInfo;
struct {
char* tablename;
char* dbname;
int32_t tversion;
SSchemaWrapper* sw;
} schemaVer;
SSchemaInfo schemaInfo;
STableListInfo tableqinfoList; // this is a table list
const char* sql; // query sql string
jmp_buf env; // jump to this position when error happens.
EOPTR_EXEC_MODEL execModel; // operator execution model [batch model|stream model]
SSubplan* pSubplan;
struct SOperatorInfo* pRoot;
} SExecTaskInfo;
......
......@@ -227,14 +227,14 @@ int32_t extractTargetSlotId(const SArray* pColMatchInfo, SExecTaskInfo* pTaskInf
for (int32_t i = 0; i < numOfCols; ++i) {
SColMatchInfo* pColMatch = taosArrayGet(pColMatchInfo, i);
for (int32_t j = 0; j < pTaskInfo->schemaVer.sw->nCols; ++j) {
if (pColMatch->colId == pTaskInfo->schemaVer.sw->pSchema[j].colId &&
for (int32_t j = 0; j < pTaskInfo->schemaInfo.sw->nCols; ++j) {
if (pColMatch->colId == pTaskInfo->schemaInfo.sw->pSchema[j].colId &&
pColMatch->colId == PRIMARYKEY_TIMESTAMP_COL_ID) {
(*pSlotIds)[pColMatch->targetSlotId] = -1;
break;
}
if (pColMatch->colId == pTaskInfo->schemaVer.sw->pSchema[j].colId) {
if (pColMatch->colId == pTaskInfo->schemaInfo.sw->pSchema[j].colId) {
(*pSlotIds)[pColMatch->targetSlotId] = j;
break;
}
......
......@@ -138,7 +138,6 @@ qTaskInfo_t qCreateQueueExecTaskInfo(void* msg, SReadHandle* readers, int32_t* n
}
}
nodesDestroyNode((SNode*)pPlan);
return pTaskInfo;
}
......@@ -165,7 +164,6 @@ qTaskInfo_t qCreateStreamExecTaskInfo(void* msg, SReadHandle* readers) {
return NULL;
}
nodesDestroyNode((SNode*)pPlan);
return pTaskInfo;
}
......@@ -243,19 +241,19 @@ int32_t qGetQueryTableSchemaVersion(qTaskInfo_t tinfo, char* dbName, char* table
ASSERT(tinfo != NULL && dbName != NULL && tableName != NULL);
SExecTaskInfo* pTaskInfo = (SExecTaskInfo*)tinfo;
if (pTaskInfo->schemaVer.sw == NULL) {
if (pTaskInfo->schemaInfo.sw == NULL) {
return TSDB_CODE_SUCCESS;
}
*sversion = pTaskInfo->schemaVer.sw->version;
*tversion = pTaskInfo->schemaVer.tversion;
if (pTaskInfo->schemaVer.dbname) {
strcpy(dbName, pTaskInfo->schemaVer.dbname);
*sversion = pTaskInfo->schemaInfo.sw->version;
*tversion = pTaskInfo->schemaInfo.tversion;
if (pTaskInfo->schemaInfo.dbname) {
strcpy(dbName, pTaskInfo->schemaInfo.dbname);
} else {
dbName[0] = 0;
}
if (pTaskInfo->schemaVer.tablename) {
strcpy(tableName, pTaskInfo->schemaVer.tablename);
if (pTaskInfo->schemaInfo.tablename) {
strcpy(tableName, pTaskInfo->schemaInfo.tablename);
} else {
tableName[0] = 0;
}
......
......@@ -4127,7 +4127,7 @@ static SExecTaskInfo* createExecTaskInfo(uint64_t queryId, uint64_t taskId, EOPT
SExecTaskInfo* pTaskInfo = taosMemoryCalloc(1, sizeof(SExecTaskInfo));
setTaskStatus(pTaskInfo, TASK_NOT_COMPLETED);
pTaskInfo->schemaVer.dbname = strdup(dbFName);
pTaskInfo->schemaInfo.dbname = strdup(dbFName);
pTaskInfo->cost.created = taosGetTimestampMs();
pTaskInfo->id.queryId = queryId;
pTaskInfo->execModel = model;
......@@ -4153,35 +4153,35 @@ int32_t extractTableSchemaInfo(SReadHandle* pHandle, uint64_t uid, SExecTaskInfo
return terrno;
}
pTaskInfo->schemaVer.tablename = strdup(mr.me.name);
pTaskInfo->schemaInfo.tablename = strdup(mr.me.name);
if (mr.me.type == TSDB_SUPER_TABLE) {
pTaskInfo->schemaVer.sw = tCloneSSchemaWrapper(&mr.me.stbEntry.schemaRow);
pTaskInfo->schemaVer.tversion = mr.me.stbEntry.schemaTag.version;
pTaskInfo->schemaInfo.sw = tCloneSSchemaWrapper(&mr.me.stbEntry.schemaRow);
pTaskInfo->schemaInfo.tversion = mr.me.stbEntry.schemaTag.version;
} else if (mr.me.type == TSDB_CHILD_TABLE) {
tDecoderClear(&mr.coder);
tb_uid_t suid = mr.me.ctbEntry.suid;
metaGetTableEntryByUid(&mr, suid);
pTaskInfo->schemaVer.sw = tCloneSSchemaWrapper(&mr.me.stbEntry.schemaRow);
pTaskInfo->schemaVer.tversion = mr.me.stbEntry.schemaTag.version;
pTaskInfo->schemaInfo.sw = tCloneSSchemaWrapper(&mr.me.stbEntry.schemaRow);
pTaskInfo->schemaInfo.tversion = mr.me.stbEntry.schemaTag.version;
} else {
pTaskInfo->schemaVer.sw = tCloneSSchemaWrapper(&mr.me.ntbEntry.schemaRow);
pTaskInfo->schemaInfo.sw = tCloneSSchemaWrapper(&mr.me.ntbEntry.schemaRow);
}
metaReaderClear(&mr);
return TSDB_CODE_SUCCESS;
}
static void cleanupTableSchemaInfo(SExecTaskInfo* pTaskInfo) {
taosMemoryFreeClear(pTaskInfo->schemaVer.dbname);
if (pTaskInfo->schemaVer.sw == NULL) {
static void cleanupTableSchemaInfo(SSchemaInfo* pSchemaInfo) {
taosMemoryFreeClear(pSchemaInfo->dbname);
if (pSchemaInfo->sw == NULL) {
return;
}
taosMemoryFree(pTaskInfo->schemaVer.sw->pSchema);
taosMemoryFree(pTaskInfo->schemaVer.sw);
taosMemoryFree(pTaskInfo->schemaVer.tablename);
taosMemoryFree(pSchemaInfo->tablename);
taosMemoryFree(pSchemaInfo->sw->pSchema);
taosMemoryFree(pSchemaInfo->sw);
}
static int32_t sortTableGroup(STableListInfo* pTableListInfo, int32_t groupNum) {
......@@ -4939,6 +4939,7 @@ int32_t createExecTaskInfoImpl(SSubplan* pPlan, SExecTaskInfo** pTaskInfo, SRead
}
(*pTaskInfo)->sql = sql;
(*pTaskInfo)->pSubplan = pPlan;
(*pTaskInfo)->pRoot = createOperatorTree(pPlan->pNode, *pTaskInfo, pHandle, &(*pTaskInfo)->tableqinfoList, pPlan->pTagCond, pPlan->pTagIndexCond, pPlan->user);
if (NULL == (*pTaskInfo)->pRoot) {
......@@ -4977,7 +4978,9 @@ void doDestroyTask(SExecTaskInfo* pTaskInfo) {
doDestroyTableList(&pTaskInfo->tableqinfoList);
destroyOperatorInfo(pTaskInfo->pRoot);
cleanupTableSchemaInfo(pTaskInfo);
cleanupTableSchemaInfo(&pTaskInfo->schemaInfo);
nodesDestroyNode((SNode*)pTaskInfo->pSubplan);
taosMemoryFreeClear(pTaskInfo->sql);
taosMemoryFreeClear(pTaskInfo->id.str);
......
......@@ -150,7 +150,6 @@ typedef struct SQWTaskCtx {
void *taskHandle;
void *sinkHandle;
SSubplan *plan;
STbVerInfo tbInfo;
} SQWTaskCtx;
......
......@@ -306,11 +306,6 @@ void qwFreeTaskCtx(SQWTaskCtx *ctx) {
dsDestroyDataSinker(ctx->sinkHandle);
ctx->sinkHandle = NULL;
}
if (ctx->plan) {
nodesDestroyNode((SNode*)ctx->plan);
ctx->plan = NULL;
}
}
int32_t qwDropTaskCtx(QW_FPARAMS_DEF) {
......@@ -327,7 +322,6 @@ int32_t qwDropTaskCtx(QW_FPARAMS_DEF) {
atomic_store_ptr(&ctx->taskHandle, NULL);
atomic_store_ptr(&ctx->sinkHandle, NULL);
atomic_store_ptr(&ctx->plan, NULL);
QW_SET_EVENT_PROCESSED(ctx, QW_EVENT_DROP);
......
......@@ -522,8 +522,6 @@ int32_t qwProcessQuery(QW_FPARAMS_DEF, SQWMsg *qwMsg, const char* sql) {
QW_ERR_JRET(code);
}
ctx->plan = plan;
code = qCreateExecTask(qwMsg->node, mgmt->nodeId, tId, plan, &pTaskInfo, &sinkHandle, sql, OPTR_EXEC_MODEL_BATCH);
if (code) {
QW_TASK_ELOG("qCreateExecTask failed, code:%x - %s", code, tstrerror(code));
......@@ -928,8 +926,6 @@ int32_t qwProcessDelete(QW_FPARAMS_DEF, SQWMsg *qwMsg, SDeleteRes *pRes) {
QW_ERR_JRET(code);
}
ctx.plan = plan;
code = qCreateExecTask(qwMsg->node, mgmt->nodeId, tId, plan, &pTaskInfo, &sinkHandle, NULL, OPTR_EXEC_MODEL_BATCH);
if (code) {
QW_TASK_ELOG("qCreateExecTask failed, code:%x - %s", code, tstrerror(code));
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册