github-actions[bot] commented on code in PR #70826:
URL: https://github.com/apache/doris/pull/70826#discussion_r4649424624
##########
fe/fe-core/src/main/java/org/apache/doris/qe/StmtExecutor.java:
##########
@@ -2173,12 +2217,24 @@ public List<ResultRow> executeInternalQuery() {
LOG.warn("Failed to run internal SQL: {}", originStmt, e);
throw new RuntimeException("Failed to execute internal SQL. "
+ Util.getRootCauseMessage(e), e);
}
+ if (context.supportHandleByFe()) {
+ Optional<ResultSet> resultSet =
planner.handleQueryInFe(parsedStmt);
+ if (resultSet.isPresent()) {
+ isHandleQueryInFe = true;
+ if (context.getSessionVariable().enableProfile() &&
profile != null) {
+
profile.getSummaryProfile().setExecutedByFrontend(true);
+ }
+ return resultSet.get().getResultRows().stream()
+ .map(ResultRow::new)
Review Comment:
[P3] Account for rows returned by the FE shortcut
This return bypasses context.updateReturnRows(), which the existing BE batch
path performs. The finally block still emits an internal audit event, and
AuditLogHelper takes ReturnRows from that context counter. For example, the
internal query for SHOW CHARSET WHERE TRUE now returns one ResultRow but
records ReturnRows=0, as do other internal queries folded to nonempty FE
results. Increment the counter by the FE result row count before returning so
audit accounting remains consistent with the BE path.
##########
fe/fe-core/src/main/java/org/apache/doris/qe/StmtExecutor.java:
##########
@@ -1255,6 +1283,22 @@ public void cancel(Status cancelReason, boolean
needWaitCancelComplete) {
}
}
+ public void setCoord(Coordinator coordinator) {
+ coord = coordinator;
+ Status cancelReason = pendingCoordinatorCancelReason.get();
+ if (coordinator != null && cancelReason != null) {
+ coordinator.cancel(cancelReason);
Review Comment:
[P2] Stop cancelled work before starting its coordinator
If KILL QUERY reaches an INSERT while it is planning (or an analysis task is
cancelled during executeInternalQuery planning), this new replay marks the
not-yet-started coordinator CANCELLED. Both Coordinator.shouldQueue() and
NereidsCoordinator.shouldQueue() then skip workload-queue admission because
isQueryCancelled() is true, but exec() still builds and dispatches the BE
fragments without a status check. At replay time the remote execution
contexts/tasks do not exist, so cancel() cannot cancel those future fragments.
With workload/query queues enabled, cancelled scans can therefore start outside
the configured concurrency limit; before this change they still went through
the queue. Reject pending cancellation on the execution thread before starting
work, or gate coordinator dispatch on its terminal status.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]