The fabric8 K8s client is using PATCH to replace get-and-update in v6.6.2.
That's why you also need to give PATCH permission for the K8s service
account.
This would help to decrease the pressure of K8s APIServer. You could find
more information here[1].
[1]. https://issues.apache.org/jira/browse/F
Hi Ethan,
Pekko is basically a fork of Akka before its license change, so the usage is
almost the same. From the exception posted, it looks like you are trying to
connector to a terminated dispatcher, which usually indicates some exceptions
on the JobManager side. You can try checking the JM lo
Never mind. The issue was fix due to the service account permission missing
“patch” verb. Which lead to RPC service not started.
> On Dec 5, 2023, at 1:40 PM, Ethan T Yang wrote:
>
> Hi Flink users,
> After upgrading Flink ( from 1.13.1 -> 1.18.0), I noticed the an issue when
> HA is enabled.(
>From the implementation of DefaultCompletedCheckpointStore, Flink will only
retain the configured amount of checkpoints.
Maybe you could also check the content of jobmanager-leader ConfigMap. It
should have the same number of pointers for the completedCheckpoint.
Best,
Yang
Ivan Yang 于202
Thanks for the reply. Yes, We are seeing all the completedCheckpoint and
they keep growing. We will revisit our k8s set up, configmap etc
> On Jun 23, 2021, at 2:09 AM, Yang Wang wrote:
>
> Hi Ivan,
>
> For completedCheckpoint files will keep growing, do you mean too many
> files exis
Hi Ivan,
For completedCheckpoint files will keep growing, do you mean too many
files exist in the S3 bucket?
AFAIK, if the K8s HA services work normally, only
one completedCheckpoint file will be retained. Once a
new one is generated, the old one will be deleted.
Best,
Yang
Ivan Yang
