[
https://issues.apache.org/jira/browse/FLINK-25333?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17460230#comment-17460230
]
Kerem Ulutaş commented on FLINK-25333:
--------------------------------------
[~sjwiesman] I tested changes in your pull request and my application could
recover successfully.
I've tried killing the jobmanager pod, killing the elected jobmanager pod when
there is 1 leader and 1 standby jobmanager, deleting the whole application from
minikube and re-deploying - the application could always recover from last
checkpoint.
One thing to note is, the "Generated hash ..." log lines mentioned in
FLINK-25267 are missing now - I believe this is the direct result of your
changes.
> Harden UID management when constructing statefun universe
> ---------------------------------------------------------
>
> Key: FLINK-25333
> URL: https://issues.apache.org/jira/browse/FLINK-25333
> Project: Flink
> Issue Type: Bug
> Components: Stateful Functions
> Reporter: Seth Wiesman
> Assignee: Seth Wiesman
> Priority: Major
> Labels: pull-request-available
> Fix For: statefun-3.2.0
>
>
> As described in this stack overflow, we do not set UIDs on routers. Because
> of how the stream graph is generated, the uids may be non-deterministic. We
> should manually set all UIDs and enforce this via configuration.
>
> https://stackoverflow.com/questions/70316498/flink-statefun-high-availability-exception-java-lang-illegalstateexception-th
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
