pds-server's unit test cache::tests::evicted_under_concurrent_writes failed in the gates of a pull that touches no Rust (harness: a start over its bound is taken again before it fails, "harness: a start over its bound is taken again before it fails"), with "every store let go once idle", and passed when the pull was queued again. The batch ran on runner.knotless.dev beside another gates run.
What you expected
The test holds the cache to letting every store go once idle; it should pass or fail by what the cache does, not by how long the runner takes to schedule the writers. A wait for "idle" measured in wall time on a loaded machine is the likely cause.
Where
The gates workflow (cargo test), on runner.knotless.dev.
What happened
pds-server's unit testcache::tests::evicted_under_concurrent_writesfailed in the gates of a pull that touches no Rust (harness: a start over its bound is taken again before it fails, "harness: a start over its bound is taken again before it fails"), with "every store let go once idle", and passed when the pull was queued again. The batch ran on runner.knotless.dev beside another gates run.What you expected
The test holds the cache to letting every store go once idle; it should pass or fail by what the cache does, not by how long the runner takes to schedule the writers. A wait for "idle" measured in wall time on a loaded machine is the likely cause.
Where
The
gatesworkflow (cargo test), on runner.knotless.dev.