[
https://issues.apache.org/jira/browse/FLINK-24237?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Chesnay Schepler updated FLINK-24237:
-------------------------------------
Description:
A recent [blog post|https://mux.com/blog/5-years-of-flink-at-mux/] by Mux
mentions that they have disabled DNS caching in their docker image because with
caching the Flink processes had trouble talking to over services when
deployments move between nodes:
??The JVM will cache DNS entries forever by default. This is undesirable in
Kubernetes deployments where there’s an expectation that DNS entries can and do
change frequently as pod deployments move between nodes. We’ve seen Flink
applications suddenly unable to talk to other services in the cluster after
pods are upgraded.??
We should investigate this, in particular whether it would make sense to set it
to a low value instead of disabling it entirely.
was:
A recent [blog post|https://mux.com/blog/5-years-of-flink-at-mux/] by Mux
mentions that they have disabled DNS caching in their docker image because
without it the Flink processes had trouble talking to over services when
deployments move between nodes:
??The JVM will cache DNS entries forever by default. This is undesirable in
Kubernetes deployments where there’s an expectation that DNS entries can and do
change frequently as pod deployments move between nodes. We’ve seen Flink
applications suddenly unable to talk to other services in the cluster after
pods are upgraded.??
We should investigate this, in particular whether it would make sense to set it
to a low value instead of disabling it entirely.
> Consider disabling or lowering DNS caching timeout in docker image
> ------------------------------------------------------------------
>
> Key: FLINK-24237
> URL: https://issues.apache.org/jira/browse/FLINK-24237
> Project: Flink
> Issue Type: Improvement
> Components: Deployment / Kubernetes
> Reporter: Chesnay Schepler
> Priority: Major
> Fix For: 1.15.0
>
>
> A recent [blog post|https://mux.com/blog/5-years-of-flink-at-mux/] by Mux
> mentions that they have disabled DNS caching in their docker image because
> with caching the Flink processes had trouble talking to over services when
> deployments move between nodes:
> ??The JVM will cache DNS entries forever by default. This is undesirable in
> Kubernetes deployments where there’s an expectation that DNS entries can and
> do change frequently as pod deployments move between nodes. We’ve seen Flink
> applications suddenly unable to talk to other services in the cluster after
> pods are upgraded.??
> We should investigate this, in particular whether it would make sense to set
> it to a low value instead of disabling it entirely.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
