Stabilize post and slow_post autests under parallel runs

[bryancall]

Problem

The post-early-return and slow_post/server_abort autests fail intermittently under parallel execution on Fedora 43. The nc -l based mock server only accepts a single TCP connection, which gets consumed by the readiness probe, leaving nothing for ATS to connect to.

Changes

• Stabilize thread_config startup ordering -- gate each check_threads invocation on ATS port readiness using When.PortOpen(ts.Variables.port) to avoid races where validation runs before traffic_server is ready.
• Stabilize post and slow_post autests -- add explicit origin readiness checks in post-early-return and accept the current microserver exception variant in server_abort for Fedora 43 compatibility.
• Revert thread_config PortOpen change per review -- keep ATS startup gating on existing fully-initialized readiness behavior. Move post-early-return upstream readiness to server process Ready fields instead of StartBefore calls.
• Replace nc-based server with mock_origin.py -- server1.sh only accepts a single TCP connection, consumed by the When.PortOpen() readiness probe. mock_origin.py absorbs probes, sends the configured HTTP response, and drains request data before closing to prevent TCP RST that causes 502 on H2 streams.
• Remove unused server1.sh -- replaced by mock_origin.py.

Testing

• 5/5 parallel ASAN runs on hera2
• All 15 CI platforms green

[Copilot]

Pull request overview

This PR improves determinism in the thread_config gold test by ensuring the thread-validation helper (check_threads.py) does not run until the corresponding ATS instance is actually accepting connections on its configured port, eliminating a startup-order race in parallel autest runs.

Changes:

• Add per-run readiness gating so each check_threads.py invocation waits on When.PortOpen(ts.Variables.port) before proceeding.
• Apply the same readiness condition consistently across all thread-configuration variants in the test.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[maskit]

[approve ci autest 1]

[bryancall]

The consistent AuTest 1of4 failure was caused by a fundamental incompatibility between nc -l and When.PortOpen():

nc -l only accepts a single TCP connection. The PortOpen readiness probe connects to check the port is open, consuming that one connection. When ATS then tries to proxy the request to the origin, the port is dead → 502 every time. This is why it failed deterministically on all 4 CI retries.

The fix (pushed in 3cf53d3) replaces server1.sh with a new shared tests/tools/mock_origin.py that:

1. Uses Python socket.listen() which handles multiple accept() calls, so readiness probes are absorbed silently
2. Sends the configured HTTP response after reading request headers
3. Drains remaining request data before closing — without this, closing the connection mid-POST causes a TCP RST that makes ATS return 502 on HTTP/2 streams (this was a second issue discovered during testing)

mock_origin.py is a reusable tool that can replace all the ad-hoc nc -l server scripts in the test suite (post/server1.sh, chunked_encoding/server2-4.sh, post_slow_server/server.sh). It supports configurable status codes, response bodies (content-length or chunked), and response delays via CLI args.

Tested 5/5 parallel ASAN runs passing on a local dev machine. The previous mock_origin.py without the drain fix was failing ~2/3 parallel runs on the HTTP/2 test case.

[Copilot]

Pull request overview

Copilot reviewed 3 out of 3 changed files in this pull request and generated no new comments.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[maskit]

Fatal: ../src/api/InkAPITest.cc:916: failed assertion (event == TS_EVENT_NET_ACCEPT) || (event == TS_EVENT_NET_ACCEPT_FAILED)

I see this error on my PR as well.

[maskit]

Can we remove server1.sh ?

[bryancall]

Yes, it can be removed. I will push a new commit to delete it.