You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Remove timer-contention throttles from send allocation tests #588
Two SenderTests allocation comparisons serialize against the rest of the test suite using [NotInParallel] to avoid lazy native timer-queue wait-event allocations. They measure thread-local allocation counters, so this throttle conflicts with the repository's parallel-test requirements. Reuse the explicit native timer-lock warmup introduced by #587 and validate the send-specific scenarios independently.
Reproduction
At main commit 74259d2, tests/Relayer.Tests/SenderTests.cs:416 and :437 apply [NotInParallel] to Direct_send_has_zero_steady_state_allocations_with_the_default_serializer and Overlapping_asynchronous_sends_retain_the_peak_pool_and_allocate_zero_bytes_after_warmup.
Run the two methods with dotnet test -c Release --project tests/Relayer.Tests --treenode-filter "/*/*/SenderTests/(Direct_send_has_zero_steady_state_allocations_with_the_default_serializer|Overlapping_asynchronous_sends_retain_the_peak_pool_and_allocate_zero_bytes_after_warmup)".
Expected and actual
Expected: native timer state is warmed during setup, exact zero-byte assertions remain, and both comparisons run alongside other tests. Actual: both tests request suite serialization. #585 establishes a deterministic native timer-lock contention mechanism for a separate batch comparison; it does not retrospectively attribute a failure in these sender tests. Historical direct-send issue #115 is closed and did not request this remaining parallelism correction.
Acceptance
Reuse the #587 setup helper, remove both throttles, and validate these two scenarios on net10.0 and net11.0 without tolerances or replacement process throttles. Keep native finite send deadlines intact.
This follow-up covers direct-send and overlapping-send comparisons separately from #585's batch-measurement fix. No new production behavior is requested.
Audited against origin/main at 61a142f. Both SenderTests allocation comparisons still carry NotInParallel, so the acceptance criteria remain current. The native blocked_by link to #585 is present; its helper PR #587 remains open and has not reached main. No label or scope change is needed.
Low-queue discovery found no new untracked issue: the recent failed main CI run 43380604838 is covered by closed #570 and merged #571; the recent performance failures are already tracked by #565, #562 and #555. There are no open milestone assignments or defect TODO/FIXME comments to split out.
Activity
thomhurst commentedon Nov 13, 2026
Audited against origin/main at 61a142f. Both SenderTests allocation comparisons still carry NotInParallel, so the acceptance criteria remain current. The native blocked_by link to #585 is present; its helper PR #587 remains open and has not reached main. No label or scope change is needed.
Low-queue discovery found no new untracked issue: the recent failed main CI run 43380604838 is covered by closed #570 and merged #571; the recent performance failures are already tracked by #565, #562 and #555. There are no open milestone assignments or defect TODO/FIXME comments to split out.
test: warm timer contention for parallel sender allocation checks (#588…