Fix flaky autests for timeout, sigusr2, and thread_config - #13012
Merged
Conversation
The ssl-delay-server test helper could die unexpectedly when a client disconnects during the handshake delay. SIGPIPE from the broken connection kills the process, or accept() returns EINTR under heavy parallel load. Add SIGPIPE ignore and EINTR retry to keep the server alive for the StillRunningAfter check.
Test 1's Default process had Ready = When.FileExists(diags.log), but by the time Default starts, rotate_diags_log has already moved diags.log to diags.log_old. This creates a deadlock: Default waits for diags.log to exist, but only SIGUSR2 (sent by Default) would cause TS to recreate it. The StartBefore chain already guarantees correct ordering (ts → rotate → Default), so the Ready condition is unnecessary and harmful.
Under ASAN, the ATS process CWD may differ from the expected ts_path. Fall back to matching ts_path in the process command line arguments so the test can find the correct traffic_server process.
Contributor
There was a problem hiding this comment.
Pull request overview
Test-only PR to stabilize three flaky AuTests under parallel ASAN runs by hardening helper behavior and improving AuTest process sequencing / matching.
Changes:
- Update
ssl-delay-serverhelper to ignoreSIGPIPEand retryaccept()onEINTR. - Make
thread_config’s thread-count helper identify the correcttraffic_serverprocess more reliably under ASAN by matching via CWD or command line. - Adjust
sigusr2test process ordering to remove a deadlocking Ready condition and clarify the intended startup chain.
Reviewed changes
Copilot reviewed 3 out of 3 changed files in this pull request and generated 2 comments.
| File | Description |
|---|---|
tests/gold_tests/timeout/ssl-delay-server.cc |
Improves helper robustness against SIGPIPE and EINTR during accept. |
tests/gold_tests/thread_config/check_threads.py |
Improves ATS process identification under ASAN by broadening the matching criteria. |
tests/gold_tests/logging/sigusr2.test.py |
Removes deadlocking Ready gating and documents intended process ordering for SIGUSR2 log rotation. |
accept() returns -1 on error but fd 0 is a valid descriptor (e.g. if stdin is closed). The <= 0 check would incorrectly treat a valid connection as failure.
bneradt
requested changes
Mar 24, 2026
This reverts commit 3db0ce7.
bneradt
approved these changes
Mar 25, 2026
cmcfarlen
pushed a commit
that referenced
this pull request
Jun 9, 2026
* Handle SIGPIPE in ssl-delay-server to prevent helper death when a client disconnects during TLS handshake delay. * Retry accept() on EINTR under heavy parallel load instead of treating the interruption as a fatal error. * Fix accept() error check to use < 0 instead of <= 0 since fd 0 is a valid descriptor when stdin is closed. * Add cmdline matching fallback in check_threads.py for ASAN where the process CWD differs from expected ts_path. (cherry picked from commit ff31470)
bneradt
pushed a commit
to bneradt/trafficserver
that referenced
this pull request
Jun 10, 2026
* Handle SIGPIPE in ssl-delay-server to prevent helper death when a client disconnects during TLS handshake delay. * Retry accept() on EINTR under heavy parallel load instead of treating the interruption as a fatal error. * Fix accept() error check to use < 0 instead of <= 0 since fd 0 is a valid descriptor when stdin is closed. * Add cmdline matching fallback in check_threads.py for ASAN where the process CWD differs from expected ts_path. (cherry picked from commit ff31470)
cmcfarlen
pushed a commit
that referenced
this pull request
Jun 10, 2026
* curl 8.20 test update: curl PROXY destination changes (#13239) curl 8.20 intentionally mirrors --haproxy-clientip into both PROXY addresses to keep the header address family consistent. The TSVConnPPInfo AuTest still expected the older destination address, so jobs with newer curl failed even though ATS preserved the PROXY metadata it received. This relaxes the destination-address expectation to accept either curl behavior while continuing to verify the source address and PROXY metadata. This also wraps the long curl command strings while leaving the test's request flow unchanged. (cherry picked from commit ad0ce02) * fedora:44: Trim remap ACL reload waits (#13237) The remap ACL AuTests run hundreds of reload scenarios in a single case, and the Fedora 44 shard is sensitive to extra reload-wait overhead, causing the tests to hang. Their reload sentinel also counted only explicit reloads, even though the log contains the startup load marker too. This replaces the long-lived sleep Ready helper with a short command that exits once the expected reload marker count is present. This also waits for the startup marker plus the explicit reload count, so each scenario observes the reload it just requested. (cherry picked from commit c52eeda) * Align AuTests with latest proxy-verifier checks (#12986) The latest proxy-verifier now fails a run when a verifier server or client is given proxy-side checks for traffic that ATS never produces. Most tests only needed stale proxy-request or proxy-response nodes removed, but the shared replay cases below need server-specific files so each verifier only owns traffic it can actually observe. - disable_pristine_host_hdr_server_canary_false.replay.yaml keeps the canary server on uuid 1 when pristine_host_hdr stays enabled. - disable_pristine_host_hdr_server_canary_true.replay.yaml keeps the canary server on uuid 1 when the Host header is rewritten. - disable_pristine_host_hdr_server_stable.replay.yaml keeps the stable server on uuid 2, which the canary server never receives. - escalate_original_server_default.replay.yaml keeps only the requests that the default-mode origin server really handles. - escalate_failover_server_default.replay.yaml keeps only the GET requests that default-mode escalation sends to failover. - escalate_original_server_non_get.replay.yaml keeps the origin-side subset when --escalate-non-get-methods is enabled. - escalate_failover_server_non_get.replay.yaml keeps the failover subset, including the escalated HEAD request in that mode. - ja4_fingerprint_basic_server.replay.yaml limits the non-preserve test to its one real request instead of preserve-only checks. - traffic_dump_server.yaml keeps origin verification only for the sessions that really reach origin in the main traffic_dump test. - traffic_dump_ip_filter_server.yaml keeps only the /one request used by the traffic_dump IP filter test. These per-server replays preserve fallback server-response directives and client-side coverage while dropping only the proxy-side checks that latest proxy-verifier now correctly reports as unprocessed. (cherry picked from commit ecf505c) * Fix 10.1.x PV replay expectations Adjust two replay expectations after backporting the proxy-verifier 3.1 cleanup. The 10.1.x branch has fewer escalate transactions, and duplicate header checks now observe the combined header value. * [autest] thread_config: add startup polling and skip test on non-Linux (#12940) * thread_config: add startup polling for thread checks and skip on non-Linux check_threads.py now uses a short bounded poll/retry window so thread-count validation does not fail on startup races; the test is also skipped on non-Linux platforms because per-thread introspection used by this check is not reliably available there. * thread_config: stop retrying when Process.threads() access is denied (cherry picked from commit 6dfaadd) * Fix flaky autests for timeout, sigusr2, and thread_config (#13012) * Handle SIGPIPE in ssl-delay-server to prevent helper death when a client disconnects during TLS handshake delay. * Retry accept() on EINTR under heavy parallel load instead of treating the interruption as a fatal error. * Fix accept() error check to use < 0 instead of <= 0 since fd 0 is a valid descriptor when stdin is closed. * Add cmdline matching fallback in check_threads.py for ASAN where the process CWD differs from expected ts_path. (cherry picked from commit ff31470) * Fix 10.1.x JA4 AuTest setup Initialize the branch-local JA4 test helper's preserve flag after backporting replay changes from master. Without it the test fails during Python load before the actual replay runs. * Proxy Verifier: use concise stack protocol specification (#13003) Proxy Verifier v3.0.0 has a more concise `stack` configurable for `protocol` specification. This makes use of that over the more verbose full `protocol` sequence. * tests/gold_tests/headers tests: use ATSReplayTest (#13033) Move the replay-friendly headers gold tests to ATSReplayTest wrappers and describe their ATS, origin, and client setup in replay YAML. This keeps the cache, range, redirect, HSTS, and alternate handling coverage while making the tests easier to read and maintain, and removes the old gold files left orphaned by the conversion. (cherry picked from commit 49cb7c8) * Adapt headers replay tests for 10.1.x --------- Co-authored-by: bneradt <bneradt@yahooinc.com> Co-authored-by: Mo Chen <mochen@apache.org> Co-authored-by: Bryan Call <bcall@apache.org>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Problem
The
tls_conn_timeoutandthread_configautests fail intermittently under parallel ASAN runs. Thessl-delay-serverhelper dies when a client disconnects during the handshake delay (SIGPIPE) or whenaccept()is interrupted (EINTR), failing theStillRunningAftercheck. Thethread_configtest can't find the correcttraffic_serverprocess under ASAN because the process CWD differs from the expectedts_path.Changes
ssl-delay-server-- ignore SIGPIPE to prevent the helper from dying when a client disconnects during the TLS handshake delay.accept()on EINTR -- under heavy parallel load,accept()can return EINTR; retry instead of treating it as a fatal error.accept()error check -- use< 0instead of<= 0since fd 0 is a valid descriptor when stdin is closed.check_threads.py-- fall back to matchingts_pathin process command line arguments when the CWD doesn't match, which happens under ASAN.Testing