[
https://issues.apache.org/jira/browse/FLINK-12884?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17232493#comment-17232493
]
Kevin Kwon commented on FLINK-12884:
------------------------------------
Just in my opinion, I think the e2e test should mostly focus on killing the job
manager since Zookeeper was used for checkpoint metadata storage aside from
leader election (which is innately handled by Kubernetes' pod spawning)
> FLIP-144: Native Kubernetes HA Service
> --------------------------------------
>
> Key: FLINK-12884
> URL: https://issues.apache.org/jira/browse/FLINK-12884
> Project: Flink
> Issue Type: New Feature
> Components: Deployment / Kubernetes, Runtime / Coordination
> Reporter: MalcolmSanders
> Assignee: Yang Wang
> Priority: Major
> Fix For: 1.12.0
>
>
> Currently flink only supports HighAvailabilityService using zookeeper. As a
> result, it requires a zookeeper cluster to be deployed on k8s cluster if our
> customers needs high availability for flink. If we support
> HighAvailabilityService based on native k8s APIs, it will save the efforts of
> zookeeper deployment as well as the resources used by zookeeper cluster. It
> might be especially helpful for customers who run small-scale k8s clusters so
> that flink HighAvailabilityService may not cause too much overhead on k8s
> clusters.
> Previously [FLINK-11105|https://issues.apache.org/jira/browse/FLINK-11105]
> has proposed a HighAvailabilityService using etcd. As [~NathanHowell]
> suggested in FLINK-11105, since k8s doesn't expose its own etcd cluster by
> design (see [Securing etcd
> clusters|https://kubernetes.io/docs/tasks/administer-cluster/configure-upgrade-etcd/#securing-etcd-clusters]),
> it also requires the deployment of etcd cluster if flink uses etcd to
> achieve HA.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
