dongjoon-hyun commented on PR #854:
URL: 
https://github.com/apache/spark-kubernetes-operator/pull/854#issuecomment-5758938083

   Thank you for the thorough review, @peter-toth! All six are addressed in 
2da0874.
   
   **1. Hold taken on a failed read.** 
`SparkAppContext.getCurrentAttemptDriverPodStrictly()` now surfaces a failed 
verification, and `getCurrentAttemptDriverPod()` keeps its lenient contract by 
wrapping it, so existing callers are unchanged. The suspend guard in 
`AppInitStep` reads the strict variant and, when the verification fails, 
returns `completeAndDefaultRequeue()` without publishing anything: an unknown 
answer is no longer read as "no driver requested", and no longer stretches the 
next look to the hold interval. `suspendedAppWithUnverifiableDriverIsNotHeld` 
covers it. The cluster half waits for the `getResourceStrictly` follow-up on 
#853, as you suggested.
   
   **2. Message claims a pending event that may never have existed.** 
`KueueWorkloadUtils.releaseWorkload` now reports whether a `Workload` was 
actually deleted, and the suffix is conditional on that. A resource created 
with `suspend: true` and a queue name therefore gets the plain message, which 
`suspendedAppWithQueueNameDoesNotCreateKueueWorkload` now asserts in full.
   
   **3. Hardcoded republish interval.** Added 
`spark.kubernetes.operator.reconciler.suspendHoldRequeueIntervalSeconds` 
(`Long`, default `1800`, dynamic override) next to 
`MISSING_DRIVER_REQUEUE_INTERVAL_SECONDS`, and its description points out that 
it must stay below the API server `--event-ttl`. 
`tests/e2e/helm/events-config-values.yaml` sets it to 10 seconds, so the 
`suspend-events` group asserts `count > 1` again for both resource kinds and 
the republishing is back under E2E coverage.
   
   **4. Nothing pins the documented 30 minutes.** Both test classes now build 
the expected progress from a `Duration.ofMinutes(30)` literal rather than from 
the value under test. Verified that it bites: setting the option default to 120 
seconds fails the suspend-hold tests.
   
   **5.** Re-wrapped, the paragraph is back to 100 or under.
   
   **6.** Reworded to "a resource held by `spec.suspend` or by Kueue, whose 
status does not report the hold", which covers the `ScheduledToRestart` case as 
well.
   


-- 
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]


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to