/*
* ExecutorStart hook: start up tracking if needed
*/
static void
pgss_ExecutorStart(QueryDesc *queryDesc, int eflags)
{
if (prev_ExecutorStart)
prev_ExecutorStart(queryDesc, eflags);
else
standard_ExecutorStart(queryDesc, eflags);
if (pgss_enabled())
{
/*
* Set up to track total elapsed time in ExecutorRun. Make sure the
* space is allocated in the per-query context so it will go away at
* ExecutorEnd.
*/
if (queryDesc->totaltime == NULL)
{
MemoryContext oldcxt;
oldcxt = MemoryContextSwitchTo(queryDesc->estate->es_query_cxt);
queryDesc->totaltime = InstrAlloc(1, INSTRUMENT_ALL);
MemoryContextSwitchTo(oldcxt);
}
}
}
/*
* ExecutorStart hook: start up logging if needed
*/
static void
explain_ExecutorStart(QueryDesc *queryDesc, int eflags)
{
/* decide */
collect = collecting_enabled();
if (collect) {
gettimeofday(&start_time, NULL);
}
if (prev_ExecutorStart)
prev_ExecutorStart(queryDesc, eflags);
else
standard_ExecutorStart(queryDesc, eflags);
/* Enable the histogram whenever the histogram is dynamic or (bins>0). */
if (collect)
{
/*
* Set up to track total elapsed time in ExecutorRun. Make sure the
* space is allocated in the per-query context so it will go away at
* ExecutorEnd.
*/
if (queryDesc->totaltime == NULL)
{
MemoryContext oldcxt;
oldcxt = MemoryContextSwitchTo(queryDesc->estate->es_query_cxt);
queryDesc->totaltime = InstrAlloc(1, INSTRUMENT_ALL);
MemoryContextSwitchTo(oldcxt);
}
}
}
static void
collectLocalMemoryStats(QueryDesc *queryDesc, int eflags)
{
BackendMemoryStat *myStat;
MemoryContextIteratorState state;
if (prev_ExecutorStart)
prev_ExecutorStart(queryDesc, eflags);
else
standard_ExecutorStart(queryDesc, eflags);
if (MyBackendProcNo < 0)
{
on_proc_exit(cleanupMyStat, 0);
MyBackendProcNo = MyProc->pgprocno;
Assert(MyBackendProcNo >= 0);
}
if (checkTick() == false)
return;
Assert(MyBackendProcNo < PROCARRAY_MAXPROCS);
myStat = NthBMS(MyBackendProcNo);
/*
* do not wait if reader currently locks our slot
*/
if (LWLockConditionalAcquire(myStat->lock, LW_EXCLUSIVE) == false)
return;
myStat->pid = MyProc->pid;
myStat->nContext = 0;
state.context = TopMemoryContext;
state.level = 0;
/*
* walk through all memory context and fill stat table in shared memory
*/
do {
MemoryContextStat *mcs = myStat->stats + myStat->nContext;
int namelen = strlen(state.context->name);
if (namelen > NAMEDATALEN - 1)
namelen = NAMEDATALEN - 1;
memcpy(mcs->name.data, state.context->name, namelen);
mcs->name.data[namelen] = '\0';
mcs->level = state.level;
getMemoryContextStat(state.context, &mcs->stat);
myStat->nContext++;
iterateMemoryContext(&state);
} while (state.context && myStat->nContext < N_MC_STAT);
LWLockRelease(myStat->lock);
}
static void
grab_ExecutorStart(QueryDesc * queryDesc, int eflags)
{
if (prev_ExecutorStart)
prev_ExecutorStart(queryDesc, eflags);
else
standard_ExecutorStart(queryDesc, eflags);
if (queryDesc->totaltime == NULL) {
MemoryContext oldcxt;
oldcxt = MemoryContextSwitchTo(queryDesc->estate->es_query_cxt);
queryDesc->totaltime = InstrAlloc(1, INSTRUMENT_ALL);
MemoryContextSwitchTo(oldcxt);
}
}
/*
* ExecutorStart hook: start up logging if needed
*/
static void
explain_ExecutorStart(QueryDesc *queryDesc, int eflags)
{
/*
* For rate sampling, randomly choose top-level statement. Either all
* nested statements will be explained or none will.
*/
if (auto_explain_log_min_duration >= 0 && nesting_level == 0)
current_query_sampled = (random() < auto_explain_sample_rate *
MAX_RANDOM_VALUE);
if (auto_explain_enabled() && current_query_sampled)
{
/* Enable per-node instrumentation iff log_analyze is required. */
if (auto_explain_log_analyze && (eflags & EXEC_FLAG_EXPLAIN_ONLY) == 0)
{
if (auto_explain_log_timing)
queryDesc->instrument_options |= INSTRUMENT_TIMER;
else
queryDesc->instrument_options |= INSTRUMENT_ROWS;
if (auto_explain_log_buffers)
queryDesc->instrument_options |= INSTRUMENT_BUFFERS;
}
}
if (prev_ExecutorStart)
prev_ExecutorStart(queryDesc, eflags);
else
standard_ExecutorStart(queryDesc, eflags);
if (auto_explain_enabled() && current_query_sampled)
{
/*
* Set up to track total elapsed time in ExecutorRun. Make sure the
* space is allocated in the per-query context so it will go away at
* ExecutorEnd.
*/
if (queryDesc->totaltime == NULL)
{
MemoryContext oldcxt;
oldcxt = MemoryContextSwitchTo(queryDesc->estate->es_query_cxt);
queryDesc->totaltime = InstrAlloc(1, INSTRUMENT_ALL);
MemoryContextSwitchTo(oldcxt);
}
}
}
/*
* ExecutorStart hook: start up logging if needed
*/
static void
explain_ExecutorStart(QueryDesc *queryDesc, int eflags)
{
if (auto_explain_enabled())
{
/* Enable per-node instrumentation iff log_analyze is required. */
if (auto_explain_log_analyze && (eflags & EXEC_FLAG_EXPLAIN_ONLY) == 0)
{
if (auto_explain_log_timing)
queryDesc->instrument_options |= INSTRUMENT_TIMER;
else
queryDesc->instrument_options |= INSTRUMENT_ROWS;
if (auto_explain_log_buffers)
queryDesc->instrument_options |= INSTRUMENT_BUFFERS;
}
}
if (prev_ExecutorStart)
prev_ExecutorStart(queryDesc, eflags);
else
standard_ExecutorStart(queryDesc, eflags);
if (auto_explain_enabled())
{
/*
* Set up to track total elapsed time in ExecutorRun. Make sure the
* space is allocated in the per-query context so it will go away at
* ExecutorEnd.
*/
if (queryDesc->totaltime == NULL)
{
MemoryContext oldcxt;
oldcxt = MemoryContextSwitchTo(queryDesc->estate->es_query_cxt);
queryDesc->totaltime = InstrAlloc(1, INSTRUMENT_ALL);
MemoryContextSwitchTo(oldcxt);
}
}
}
/*
* PgPaxosExecutorStart blocks until the table is ready to read.
*/
static void
PgPaxosExecutorStart(QueryDesc *queryDesc, int eflags)
{
PlannedStmt *plannedStmt = queryDesc->plannedstmt;
List *rangeTableList = plannedStmt->rtable;
CmdType commandType = queryDesc->operation;
if (IsPgPaxosActive() && HasPaxosTable(rangeTableList))
{
char *sqlQuery = (char *) queryDesc->sourceText;
char *groupId = NULL;
bool topLevel = true;
/* disallow transactions during paxos commands */
PreventTransactionChain(topLevel, "paxos commands");
groupId = DeterminePaxosGroup(rangeTableList);
if (commandType == CMD_INSERT || commandType == CMD_UPDATE ||
commandType == CMD_DELETE)
{
PrepareConsistentWrite(groupId, sqlQuery);
}
else
{
PrepareConsistentRead(groupId);
}
queryDesc->snapshot->curcid = GetCurrentCommandId(false);
}
/* call into the standard executor start, or hook if set */
if (PreviousExecutorStartHook != NULL)
{
PreviousExecutorStartHook(queryDesc, eflags);
}
else
{
standard_ExecutorStart(queryDesc, eflags);
}
}
/*
* multi_ExecutorStart is a hook called at at the beginning of any execution
* of any query plan.
*
* If a distributed relation is the target of the query, perform some validity
* checks. If a legal statement, start the distributed execution. After that
* the to-be-executed query is replaced with the portion executing solely on
* the master.
*/
void
multi_ExecutorStart(QueryDesc *queryDesc, int eflags)
{
PlannedStmt *planStatement = queryDesc->plannedstmt;
if (HasCitusToplevelNode(planStatement))
{
MultiPlan *multiPlan = GetMultiPlan(planStatement);
MultiExecutorType executorType = MULTI_EXECUTOR_INVALID_FIRST;
Job *workerJob = multiPlan->workerJob;
ExecCheckRTPerms(planStatement->rtable, true);
executorType = JobExecutorType(multiPlan);
if (executorType == MULTI_EXECUTOR_ROUTER)
{
Task *task = NULL;
List *taskList = workerJob->taskList;
TupleDesc tupleDescriptor = ExecCleanTypeFromTL(
planStatement->planTree->targetlist, false);
List *dependendJobList PG_USED_FOR_ASSERTS_ONLY = workerJob->dependedJobList;
/* router executor can only execute distributed plans with a single task */
Assert(list_length(taskList) == 1);
Assert(dependendJobList == NIL);
task = (Task *) linitial(taskList);
/* we need to set tupleDesc in executorStart */
queryDesc->tupDesc = tupleDescriptor;
/* drop into the router executor */
RouterExecutorStart(queryDesc, eflags, task);
}
else
{
PlannedStmt *masterSelectPlan = MasterNodeSelectPlan(multiPlan);
CreateStmt *masterCreateStmt = MasterNodeCreateStatement(multiPlan);
List *masterCopyStmtList = MasterNodeCopyStatementList(multiPlan);
RangeTblEntry *masterRangeTableEntry = NULL;
StringInfo jobDirectoryName = NULL;
/*
* We create a directory on the master node to keep task execution results.
* We also register this directory for automatic cleanup on portal delete.
*/
jobDirectoryName = JobDirectoryName(workerJob->jobId);
CreateDirectory(jobDirectoryName);
ResourceOwnerEnlargeJobDirectories(CurrentResourceOwner);
ResourceOwnerRememberJobDirectory(CurrentResourceOwner, workerJob->jobId);
/* pick distributed executor to use */
if (eflags & EXEC_FLAG_EXPLAIN_ONLY)
{
/* skip distributed query execution for EXPLAIN commands */
}
else if (executorType == MULTI_EXECUTOR_REAL_TIME)
{
MultiRealTimeExecute(workerJob);
}
else if (executorType == MULTI_EXECUTOR_TASK_TRACKER)
{
MultiTaskTrackerExecute(workerJob);
}
/* then create the result relation */
ProcessUtility((Node *) masterCreateStmt,
"(temp table creation)",
PROCESS_UTILITY_QUERY,
NULL,
None_Receiver,
NULL);
/* make the temporary table visible */
CommandCounterIncrement();
if (!(eflags & EXEC_FLAG_EXPLAIN_ONLY))
{
CopyQueryResults(masterCopyStmtList);
}
/*
* Update the QueryDesc's snapshot so it sees the table. That's not
* particularly pretty, but we don't have much of a choice. One might
* think we could unregister the snapshot, push a new active one,
* update it, register it, and be happy. That only works if it's only
* registered once though...
*/
queryDesc->snapshot->curcid = GetCurrentCommandId(false);
/*
* Set the OID of the RTE used in the master select statement to point
* to the now created (and filled) temporary table. The target
* relation's oid is only known now.
*/
masterRangeTableEntry =
(RangeTblEntry *) linitial(masterSelectPlan->rtable);
masterRangeTableEntry->relid =
RelnameGetRelid(masterRangeTableEntry->eref->aliasname);
/*
* Replace to-be-run query with the master select query. As the
* planned statement is now replaced we can't call GetMultiPlan() in
* the later hooks, so we set a flag marking this as a distributed
* statement running on the master. That e.g. allows us to drop the
* temp table later.
*
* We copy the original statement's queryId, to allow
* pg_stat_statements and similar extension to associate the
* statement with the toplevel statement.
*/
masterSelectPlan->queryId = queryDesc->plannedstmt->queryId;
queryDesc->plannedstmt = masterSelectPlan;
eflags |= EXEC_FLAG_CITUS_MASTER_SELECT;
}
}
/* if the execution is not done for router executor, drop into standard executor */
if (queryDesc->estate == NULL ||
!(queryDesc->estate->es_top_eflags & EXEC_FLAG_CITUS_ROUTER_EXECUTOR))
{
standard_ExecutorStart(queryDesc, eflags);
}
}