TingYao Huang created YUNIKORN-3468:
---------------------------------------
Summary: Flaky quota-preemption e2e: 5s pod-running wait (twin
uses 30s)
Key: YUNIKORN-3468
URL: https://issues.apache.org/jira/browse/YUNIKORN-3468
Project: Apache YuniKorn
Issue Type: Bug
Components: test - e2e
Reporter: TingYao Huang
Fails intermittently on the oldest matrix image (`kindest/node v1.24.17`;
v1.34–v1.37 pass). First seen: Actions run `35429951267`.
[FAIL] quota_preemption_test.go:513 — context deadline exceeded
Root cause — the setup step is under-budgeted:
- `quota_preemption_test.go:512` → `WaitForNPodsBySelectorRunning(dev,
"app=app-a", 3, 5*time.Second)` waits for 3 pods *{*}Running{*}* in *{*}5s{*}*
(the only use of this helper in the repo).
- Its twin spec `Quota_Preemption_Delay_Timer_Reset_On_Delay_Update` does the
identical step (`:431` + `:433`) with *{*}30s{*}*.
5s isn't enough on the slow v1.24.17 runner, especially while the previous spec
is still tearing down on the shared scheduler (preempted pod retrying Bind into
a terminating namespace; queues cycling Active↔Draining). Not related to PR
#1097 (touches only `pkg/cache`).
refer e2e result: [e2e
log|https://github.com/apache/yunikorn-k8shim/actions/runs/35429951267/job/105992565214?pr=1097]
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
