[
https://issues.apache.org/jira/browse/FLINK-38562?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18128097#comment-18128097
]
Sangkyoon Nam commented on FLINK-38562:
---------------------------------------
Hi [~dianfu], I'd like to work on this. A standalone pemja 0.5.7 spike ran an
asyncio loop on a Python thread across interpreter calls, with callbacks from
that thread into a Java queue, and the loop stopped before the interpreter
closed once the results were in.
My default would be {{AsyncWaitOperator}}'s semantics for thread mode:
in-flight elements checkpointed in operator state, timeouts and retries on
processing-time timers through the mailbox, and the same ordered and unordered
modes and capacity as the Java API. Process mode instead drains its Python
queue before the checkpoint barrier, so recovery would differ between the two
modes. I'd keep the async scalar function ({{CommonExecPythonAsyncCalc}}) out
of scope.
Would you rather thread mode match process mode's checkpoint behavior?
> Support thread mode of async function in Python DataStream API
> ---------------------------------------------------------------
>
> Key: FLINK-38562
> URL: https://issues.apache.org/jira/browse/FLINK-38562
> Project: Flink
> Issue Type: Sub-task
> Components: API / Python
> Reporter: Dian Fu
> Priority: Major
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
