[ci] Run flakiness detection under concurrent load (#100525)
The flakiness detection jobs ran three attempts sequentially with concurrency set to one, while regular test jobs run up to eight suites at once. The response-read flake addressed by #100514 passed all three checks for #100447, which prompted this change. The detector now queues three runs per suite in one runner and uses the job's configured concurrency. Each run has an isolated application directory and separate results, traces, and failure tracking. Retries and result caching remain disabled so every run executes and any failure fails the job. The testing guide documents how to keep concurrent runs independent and allocate listening ports dynamically.
Showing
This diff is collapsed.
Please
register
or
sign in
to comment