Skip to content

Fix: drain SUB workers during scheduler test teardown #1668

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
ChaoWao merged 1 commit into from
Aug 4, 2026
Merged

Fix: drain SUB workers during scheduler test teardown #1668

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
13 changes: 11 additions & 2 deletions tests/ut/cpp/hierarchical/test_scheduler.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2370,6 +2370,8 @@ struct GroupSchedulerFixture : public ::testing::Test {
worker_a.drain();
worker_b.drain();
worker_c.drain();
sub_worker_a.drain();
sub_worker_b.drain();
sched.stop();
manager.stop();
allocator.shutdown();
Expand Down Expand Up @@ -2842,8 +2844,8 @@ TEST_F(GroupSchedulerFixture, ConsecutiveGroupsReserveOnlyBlockedHeadTargets) {

TEST_F(GroupSchedulerFixture, TearDownDrainsCurrentAndQueuedDispatches) {
// The verification is the teardown itself: work is left deliberately in
// both states — running and queued-but-undispatched — so teardown drains
// a worker mid-task and one whose dispatch has not happened yet.
// both states — running and queued-but-undispatched — across NEXT_LEVEL
// and SUB workers, so every registered worker type must drain cleanly.
{
std::lock_guard<std::mutex> scheduler_pause(sched.loop_mutex());
(void)orch.submit_next_level_group(
Expand All @@ -2856,12 +2858,19 @@ TEST_F(GroupSchedulerFixture, TearDownDrainsCurrentAndQueuedDispatches) {
);
(void)orch.submit_next_level(C(86), single_tensor_args(0x10A, TensorArgType::OUTPUT), cfg, 0);
(void)orch.submit_next_level(C(87), single_tensor_args(0x10B, TensorArgType::OUTPUT), cfg, 2);
(void)orch.submit_sub_group(
C(88), {single_tensor_args(0x10C, TensorArgType::OUTPUT), single_tensor_args(0x10D, TensorArgType::OUTPUT)}
);
}

worker_a.wait_running();
worker_b.wait_running();
sub_worker_a.wait_running();
sub_worker_b.wait_running();
EXPECT_TRUE(worker_a.is_running.load(std::memory_order_acquire));
EXPECT_TRUE(worker_b.is_running.load(std::memory_order_acquire));
EXPECT_TRUE(sub_worker_a.is_running.load(std::memory_order_acquire));
EXPECT_TRUE(sub_worker_b.is_running.load(std::memory_order_acquire));
}

// The shape that hangs when the idle edge is missing: a second task queued for
Expand Down
Loading