Skip to content

[GLUTEN-9456][VL] Cancel load earlier in GlutenDirectBufferedInput destructor #11697

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
rui-mo merged 1 commit into from
Mar 14, 2026
+8 −1
Merged

[GLUTEN-9456][VL] Cancel load earlier in GlutenDirectBufferedInput destructor #11697

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
9 changes: 8 additions & 1 deletion cpp/velox/memory/GlutenDirectBufferedInput.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -48,6 +48,14 @@ class GlutenDirectBufferedInput : public facebook::velox::dwio::common::DirectBu

~GlutenDirectBufferedInput() override {
requests_.clear();
// Cancel all the planned loads as soon as possible to avoid unnecessary IO.
for (auto& load : coalescedLoads_) {
Copy link
Member

@PHILO-HE PHILO-HE Mar 4, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Can we unify the two loops into one?

Copy link
Contributor Author

@rui-mo rui-mo Mar 5, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I intentionally moved this forward to cancel the load as early as possible, trying to reduce wait time. Thanks.

Copy link
Contributor

@FelixYBW FelixYBW Mar 13, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The goal is that we should cancel the unused IO as soon as possible.

In old logic, a I/O thread may still be scheduled when we are waiting for current I/O.

if (load->state() == facebook::velox::cache::CoalescedLoad::State::kPlanned) {
load->cancel();
}
}
// Ensure all the loadings can finish in the TableScan destructor to avoid the issue that the load is still running
// when the VeloxMemoryManager used by the whole task is trying to destruct.
for (auto& load : coalescedLoads_) {
if (load->state() == facebook::velox::cache::CoalescedLoad::State::kLoading) {
folly::SemiFuture<bool> waitFuture(false);
Expand All @@ -56,7 +64,6 @@ class GlutenDirectBufferedInput : public facebook::velox::dwio::common::DirectBu
std::move(waitFuture).via(&exec).wait();
}
}
load->cancel();
}
coalescedLoads_.clear();
}
Expand Down
Loading