[
https://issues.apache.org/jira/browse/FLINK-41329?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18137845#comment-18137845
]
sepuri sai krishna commented on FLINK-41329:
--------------------------------------------
I have a fix and tests for this locally. Please assign it to me and I will open
a PR.
> Restore tests are absent for all seven Python stream ExecNodes, four of them
> stateful
> -------------------------------------------------------------------------------------
>
> Key: FLINK-41329
> URL: https://issues.apache.org/jira/browse/FLINK-41329
> Project: Flink
> Issue Type: Bug
> Components: Table SQL / Planner
> Affects Versions: 2.4.0
> Reporter: sepuri sai krishna
> Priority: Major
>
> {{RestoreTestCompletenessTest}} skips seven Python stream nodes under the
> comment "Ignoring python
> based exec nodes temporarily", with no ticket behind it:
> {noformat}
> StreamExecPythonCalc
> StreamExecPythonCorrelate
> StreamExecPythonAsyncCalc
> StreamExecPythonOverAggregate
> StreamExecPythonGroupAggregate
> StreamExecPythonGroupTableAggregate
> StreamExecPythonGroupWindowAggregate
> {noformat}
> So none of them has a restore test, and a compiled plan for any of them is
> never restored from a
> savepoint in CI.
> Four of the seven are stateful. The operators behind
> {{StreamExecPythonGroupAggregate}},
> {{StreamExecPythonGroupTableAggregate}}, {{StreamExecPythonOverAggregate}} and
> {{StreamExecPythonGroupWindowAggregate}} all keep keyed state, so a plan or
> state change in those
> four can break restore without any test noticing.
> The other two groups in that same skip set are accounted for: the two window
> nodes are covered by
> {{WindowAggregateEventTimeRestoreTest}}, and the two batch nodes point at
> FLINK-40306. The Python
> block is the only entry with neither a cover nor a ticket.
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
