315225 2026-05-20 13:17:05 -0700 [Site Isolation] Layout test cross-origin-iframe-request-id-uniqueness.html times out under mac-wk2-stress; skipped pending speedup or harness fix 2026-05-20 13:20:09 -0700 1 1 1 Unclassified WebKit Web Inspector WebKit Nightly Build All All NEW InRadar P2 Normal --- 310164 1 bburg webkit-unassigned inspector-bugzilla-changes webkit-bug-importer oldest_to_newest 2212646 0 bburg 2026-05-20 13:17:05 -0700 The new layout test http/tests/site-isolation/inspector/network/cross-origin-iframe-request-id-uniqueness.html, added by webkit.org/b/310164 to verify deterministic Network IDs across cross-origin iframes under Site Isolation, is reliably slow and frequently times out under mac-wk2-stress on EWS. Cause: the test spawns several processes (one main page + two cross-origin iframe WebContent processes under Site Isolation, plus an Inspector process) and waits for inspector-side Frame.ResourceWasAdded events for two separate fetches. Under stress mode's parallel workers, inter-worker contention on the shared HTTP server and process launches pushes the test past the timeout, even with [ Slow ]. Sequential bisect (1 worker) passes 30/30; the test logic itself is not racy. Recent EWS run showing the Pass/Timeout pattern: https://ews-build.webkit.org/#/builders/8/builds/146566 Marking the test as [ Skip ] in platform/mac-wk2/TestExpectations to unblock the proxy-page-agent stack (PRs #60842, #61172, #61177). Re-enable when one of: - the test is restructured to launch fewer processes, or - the test runner / EWS stress configuration handles process-heavy Inspector tests without parallel-worker contention. Observed locally: 29-48% timeout rate with parallel workers; 0% sequential. 2212647 1 webkit-bug-importer 2026-05-20 13:17:11 -0700 <rdar://problem/177554308> 2212648 2 bburg 2026-05-20 13:20:09 -0700 Quick archaeology of related-but-not-identical prior art (none addresses process-heavy SI Inspector tests directly): - webkit.org/b/313791 -- 5 http/tests/inspector constant TIMEOUTs were caused by a build-system symlink issue copying WebInspectorUI resources, not test logic. Different root cause. - webkit.org/b/313860 -- TestWebKitAPI.WKInspectorExtension flaky timeout fixed by addressing the underlying regression rather than annotating the test. Different surface. - webkit.org/b/303862 -- Site Isolation mixed-content test failures fixed by replacing determinism-breaking frame load callbacks with setTimeout fallbacks. The right long-term fix for this test is most likely structural -- consolidate the two iframe scenarios into a single iframe (one cross-origin fetch + one same-origin fetch from the same iframe) so we only spawn one extra WebContent process instead of two, or rewrite the assertion at a layer that does not require launching processes at all (e.g., a unit test of the request-ID assignment logic in the dispatcher).