[
https://issues.apache.org/jira/browse/FLINK-35145?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Zhanghao Chen updated FLINK-35145:
----------------------------------
Description:
Currently, cluster termination may be blocked forever as there's no timeout for
that. For example, for an Application cluster with ZK HA enabled, when ZK
cluster is down, the cluster will reach termination status, but the termination
process will be blocked when trying to clean up HA data on ZK, where the ZK
client will retry connecting to ZK forever. Similar phenomenon can be observed
when an HDFS outage occurs.
I propose adding a timeout for the cluster termination process in
ClusterEntryPoint#
shutDownAsync method.
was:
Currently, cluster termination may be blocked forever as there's no timeout for
that. For example, for an Application cluster with ZK HA enabled, when ZK
cluster is down, the cluster will reach termination status, but the termination
process will be blocked when trying to clean up HA data on ZK, where the ZK
client will retry connecting to ZK forever. Similar phenomenon can be observed
when an HDFS/S3 outage occurs.
I propose adding a timeout for the cluster termination process in
ClusterEntryPoint#
shutDownAsync method.
> Add timeout for cluster termination
> -----------------------------------
>
> Key: FLINK-35145
> URL: https://issues.apache.org/jira/browse/FLINK-35145
> Project: Flink
> Issue Type: Improvement
> Components: Runtime / Coordination
> Affects Versions: 1.20.0
> Reporter: Zhanghao Chen
> Priority: Major
> Fix For: 1.20.0
>
>
> Currently, cluster termination may be blocked forever as there's no timeout
> for that. For example, for an Application cluster with ZK HA enabled, when ZK
> cluster is down, the cluster will reach termination status, but the
> termination process will be blocked when trying to clean up HA data on ZK,
> where the ZK client will retry connecting to ZK forever. Similar phenomenon
> can be observed when an HDFS outage occurs.
> I propose adding a timeout for the cluster termination process in
> ClusterEntryPoint#
> shutDownAsync method.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
