[
https://issues.apache.org/jira/browse/FLINK-35562?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18088521#comment-18088521
]
Martijn Visser commented on FLINK-35562:
----------------------------------------
Another occurrence:
https://dev.azure.com/apache-flink/apache-flink/_build/results?buildId=75906&view=results
(leg: test_cron_azure table, 2026-06-12).
Root cause identified: {{RestoreTestBase#testRestore}}
({{AfterRestoreSource.INFINITE}} branch) waits on
{{CompletableFuture.allOf(...).get()}} with no timeout; the futures only
complete when the sink observer sees an *exact* match of the expected results.
For {{PROCTIME()}} window programs, window boundaries come from the wall clock,
so records can split across windows differently than when the savepoint and
expected data were captured. The expected collection is then never matched and
the fork hangs until the 900 s watchdog kill. The thread dump from build 75906
confirms it: the JUnit worker parked in {{CompletableFuture.get}} at
{{RestoreTestBase.java:434}}, MiniCluster job alive and fully idle. Same
mechanism as FLINK-34404 ({{GroupWindowAggregateProcTimeRestoreTest}} shares
this base). Fix incoming: bound the wait and fail with actual-vs-expected
output naming the program.
> WindowTableFunctionProcTimeRestoreTest produced no output for 900 seconds
> -------------------------------------------------------------------------
>
> Key: FLINK-35562
> URL: https://issues.apache.org/jira/browse/FLINK-35562
> Project: Flink
> Issue Type: Bug
> Components: Build System / CI
> Affects Versions: 1.20.0
> Reporter: Weijie Guo
> Priority: Major
>
> https://dev.azure.com/apache-flink/apache-flink/_build/results?buildId=60172&view=logs&j=32715a4c-21b8-59a3-4171-744e5ab107eb&t=ff64056b-5320-5afe-c22c-6fa339e59586&l=11885
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
