github-actions[bot] commented on code in PR #65713:
URL: https://github.com/apache/doris/pull/65713#discussion_r3636008547
##########
be/src/runtime/workload_group/workload_group_manager.cpp:
##########
@@ -865,11 +863,13 @@ void
WorkloadGroupMgr::update_queries_limit_(WorkloadGroupPtr wg, bool enable_ha
// If the query is a pure load task, then should not modify its limit.
Or it will reserve
// memory failed and we did not hanle it.
if (!resource_ctx->task_controller()->is_pure_load_task()) {
Review Comment:
[P1] Refresh pure-load trackers after process expansion
Pure stream/routine/group-commit queries set their live tracker and adjusted
baseline to the current `MemInfo::mem_limit()` when they are created, but this
guard excludes them from the only periodic reassignment. If the cgroup grows
from 8 GiB to 32 GiB, the load therefore keeps an 8 GiB task tracker.
`PipelineTask` reservations check that tracker before the refreshed
workload-group/process limits, so valid work above the old cap returns
`QUERY_MEMORY_EXCEEDED`; disabling reserve memory then enables allocator hard
checks on the same stale tracker. A live-only raise is also insufficient
because the workload-group paused path can reapply the stale adjusted value.
This is distinct from the existing requested-limit thread because the current
fix restores only non-pure queries. Please ensure process expansion raises both
existing pure-load trackers and their adjusted/baseline limit to the new
process cap, without applying per-query or slot-policy limits, and add a
pure-load expansion r
egression that exercises a subsequent workload-group-memory pause.
--
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]
