You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@flink.apache.org by "Shaswata Jash (JIRA)" <ji...@apache.org> on 2018/03/15 02:50:00 UTC
[jira] [Commented] (FLINK-8304) Document Kubernetes and Flink HA
setup
[ https://issues.apache.org/jira/browse/FLINK-8304?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16399822#comment-16399822 ]
Shaswata Jash commented on FLINK-8304:
--------------------------------------
[~uce]
Hello Ufuk,
Please feel free to correct my present understanding : My understanding is - for Kubernetes cluster, we don't require standby jobManager to be setup to achieve HA . Because, Kubernetes master itself will take care to bring up the failed JobManager if replica is set. Thus, in kubernetes cluster we just need to ensure HA mode is set along with providing information about zookeeper and shared storage location (e.g. Cinder, glusterfs etc. persistent volume) where JobManager can store job related meta-data. My understanding is- why it should be any different than achieving [HA in YARN managed cluster|https://ci.apache.org/projects/flink/flink-docs-release-1.4/ops/jobmanager_high_availability.html#yarn-cluster-high-availability] ("When running a highly available YARN cluster, we don’t run multiple JobManager (ApplicationMaster) instances, *but only one*, which is restarted by YARN on failures.").
Also, I was not able to find out the configuration parameter ("high-availability.jobmanager.port") that you mentioned in your description of this issue in the official list of [flink 1.4 configuration|https://ci.apache.org/projects/flink/flink-docs-release-1.4/ops/config.html] . Is it available as part of any unreleased flink version?
> Document Kubernetes and Flink HA setup
> --------------------------------------
>
> Key: FLINK-8304
> URL: https://issues.apache.org/jira/browse/FLINK-8304
> Project: Flink
> Issue Type: Improvement
> Components: Documentation
> Reporter: Ufuk Celebi
> Priority: Major
>
> Currently the Flink on Kubernetes documentation does not mention anything about running Flink in HA mode.
> We should add at least the following two things:
> - Currently, there cannot be a standby JobManager pod due to the way Flink HA works
> - `high-availability.jobmanager.port` has to be set to a port that is exposed via Kubernetes
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)