mandarwagh9 opened a new pull request, #13450:
URL: https://github.com/apache/trafficserver/pull/13450
Fixes #13445.
### Problem
`tests/gold_tests/h2/h2origin.test.py` hand-rolled its "wait for
squid.log" step:
```python
tr = Test.AddTestRun("Wait for the squid.log to be written")
timeout = 30
watcher = tr.Processes.Process("watcher")
watcher.Command = f"sleep {timeout}"
watcher.Ready = When.FileContains(ts.Disk.squid_log.Name, r'14 http/1.1
http/2')
watcher.TimeOut = timeout
...
```
The intent is a 30-second budget, but neither `watcher.TimeOut` nor
`tr.TimeOut` governs the readiness gate. That gate is `StartupTimeout`,
which `Processes.Process()` defaults to 10 seconds:
- `autest/testenities/processes.py`: `startup_timeout=10, # default to 10
second as most things should be ready by this time`
- `autest/runlogic/runlogic.py`: `if ready_item.object.StartupTimeout <
curr_temp: ... "Checking that {typename} is ready within
{ready_item.object.StartupTimeout} seconds ..."`
So the run failed at 10.0s while the watcher was still waiting at 9.5s,
exactly as reported in the issue. On a loaded CI worker the squid.log
entry for the last transaction had not been flushed yet.
Worth noting: simply raising `StartupTimeout` to 30 would still race,
because the watcher's own `sleep 30` would exit at about the same
moment, tripping autest's "Process finished before it was ready" path.
### Fix
Two complementary changes:
1. **Use the existing `AddAwaitFileContainsTestRun` helper**
(`tests/gold_tests/autest-site/when.test.ext`) instead of the
hand-rolled watcher. It sets `StartupTimeout = 30` against a
`sleep 60`, so the 30-second budget is actually honored with no
sleep/gate race. This helper is already used at 20+ call sites;
`logging/log-milestone-fields.test.py::_waitForLog` is the same
shape, `StillRunningAfter` lines included. h2origin was simply a
leftover from before the helper existed.
2. **Set `proxy.config.log.max_secs_per_buffer: 1`** so the log entry
appears promptly rather than waiting out the default 5-second buffer
flush. This addresses the underlying latency the issue describes
rather than only widening the window. 13 other gold tests that assert
on log contents already do this.
The needle, test-run name, and `StillRunningAfter` assertions are
unchanged, so coverage is identical.
### Testing
Verified locally: `yapf` 0.43.0 with `.style.yapf` reports no
formatting changes, and the file compiles. I develop on Windows and
cannot build or run ATS there, so I have not executed the autest
myself — relying on Jenkins for that. The change is confined to test
configuration and contains no production code.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
