From 1025bb79773fe79fed82e738500b4b475340940d Mon Sep 17 00:00:00 2001 From: Yahaya Suleiman Date: Fri, 24 Apr 2026 04:19:41 -0700 Subject: [PATCH] Fix flaky timing assertion in gloo transport multiproc tests Summary: The IoErrors and UnboundIoErrors tests assert that error propagation after SIGKILL completes within `kMultiProcTimeout * 2` (6s). On loaded CI machines, TCP error detection and process scheduling delays can push wall-clock time well beyond this limit (observed ~10.6s), causing spurious failures. Relax the multiplier from 2x to 4x to accommodate CI variability while still catching genuine hangs. ___ overriding_review_checks_triggers_an_audit_and_retroactive_review Oncall Short Name: testing_frameworks Differential Revision: D102326105 --- gloo/test/transport_test.cc | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/gloo/test/transport_test.cc b/gloo/test/transport_test.cc index dbba55904..c60e68d0a 100644 --- a/gloo/test/transport_test.cc +++ b/gloo/test/transport_test.cc @@ -98,7 +98,7 @@ TEST_P(TransportMultiProcTest, IoErrors) { wait(); const auto delta = std::chrono::duration_cast( std::chrono::high_resolution_clock::now() - start); - ASSERT_LT(delta.count(), kMultiProcTimeout.count() * 2); + ASSERT_LT(delta.count(), kMultiProcTimeout.count() * 4); for (auto i = 0; i < processCount; i++) { if (i != 0) { @@ -199,7 +199,7 @@ TEST_P(TransportMultiProcTest, UnboundIoErrors) { wait(); const auto delta = std::chrono::duration_cast( std::chrono::high_resolution_clock::now() - start); - ASSERT_LT(delta.count(), kMultiProcTimeout.count() * 2); + ASSERT_LT(delta.count(), kMultiProcTimeout.count() * 4); for (auto i = 0; i < processCount; i++) { if (i != 0) {