perf(ci): narrow native checks and deduplicate Rust qualification - #439
Conversation
|
Final verification for 2673d7c:
|
|
Resolved a second incoming conflict from #436, preserving its beta102 fixture identity and test-variable cleanup while retaining exact-tag (not newest-release) verification. All 198 script regressions pass. Removed the temporary ci:full label now that the experiment and full PR validation are complete: ordinary PR feedback remains fast and the required merge queue still performs full qualification on the combined candidate. |
|
Merge-candidate run 35109046938 passed all lanes except Windows filesystem durability: unchanged test server::policy::tests::stuck_admitted_durable_work_does_not_delay_snapshot_or_publish_receipt failed started.elapsed() < Duration::from_millis(100). Core tests passed 150/150 and daemon tests 122/123. Classifying this one strict wall-clock assertion as a likely runner-timing flake and requeuing once; retaining the failed evidence and not changing assertions or qualification gates. |
Changes
Measured binary distribution
Run 35100242511 successfully transferred the 236 MiB artifact to six Linux runners: upload/compression 12s; downloads 7–19s; extraction/runtime probes 0–1s. Shard-local builds cost 132–189s each. However the existing Rust producer finishes after 454s, so waiting for it would worsen latency. Keep independent builds; a future build-first producer needs an end-to-end experiment. Detailed evidence: docs/ci-qualification.md.
Validation