[
https://issues.apache.org/jira/browse/FLINK-11105?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16859030#comment-16859030
]
Nathan Howell commented on FLINK-11105:
---------------------------------------
Regarding Kubernetes support, managed Kubernetes offerings such as GKE and EKS
do not expose etcd. The same functionality can be implement using only
Kubernetes APIs - a mix of coordination/v1beta1 Lease and ConfigMap resources,
or purely with ConfigMaps on older versions of Kubernetes... I think Lease was
introduced in 1.13 or 1.14. Atomic replace operations and polling are
sufficient to implement cooperative leader election, checkpoint counters, etc.
> Add a new implementation of the HighAvailabilityServices using etcd
> -------------------------------------------------------------------
>
> Key: FLINK-11105
> URL: https://issues.apache.org/jira/browse/FLINK-11105
> Project: Flink
> Issue Type: New Feature
> Components: Runtime / Coordination
> Reporter: Yang Wang
> Assignee: MalcolmSanders
> Priority: Major
>
> In flink, we use HighAvailabilityServices to do many things, e.g. RM/JM
> leader election and retrieval. ZooKeeperHaServices is an implementation of
> HighAvailabilityServices using Apache ZooKeeper. It is very easy to integrate
> with hadoop ecosystem. However, the cloud native and micro service are become
> more and more popular. We just need to follow the step and add a new
> implementation EtcdHaService using etcd.
> Now flink has supported to run StandaloneSession on kubernetes and FLINK-9953
> start to make an native integration with kubernetes. If we have the
> EtcdHaService, both of them will benefit from this and we will not have
> deploy a zookeeper service on kubernetes cluster.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
