Re: High Availability on Yarn

Posted by Jain, Ankit on
URL: http://deprecated-apache-flink-user-mailing-list-archive.369.s1.nabble.com/High-Availability-on-Yarn-tp12938p12984.html

Thanks for your reply Aljoscha.

 

After building better understanding of Yarn and spending copious amount of time on Flink codebase, I think I now get how Flink & Yarn interact – I plan to document this soon in case it could help somebody starting afresh with Flink-Yarn.

 

Regarding Zookeper, in YARN mode there is only one JobManager running, do we still need leader election?

 

If the ApplicationMaster goes down (where JM runs) it is restarted by Yarn RM and while restarting, Flink AM will bring back previous running containers.  So, where does Zookeeper sit in this setup?

 

Thanks

Ankit

 

From: Aljoscha Krettek <[hidden email]>
Date: Wednesday, May 3, 2017 at 2:05 AM
To: "Jain, Ankit" <[hidden email]>
Cc: "[hidden email]" <[hidden email]>, Till Rohrmann <[hidden email]>
Subject: Re: High Availability on Yarn

 

Hi,

As a first comment, the work mentioned in the FLIP-6 doc you linked is still work-in-progress. You cannot use these abstractions yet without going into the code and setting up a cluster “by hand”.

 

The documentation for one-step deployment of a Job to YARN is available here: https://ci.apache.org/projects/flink/flink-docs-release-1.2/setup/yarn_setup.html#run-a-single-flink-job-on-yarn

 

Regarding your third question, ZooKeeper is mostly used for discovery and leader election. That is, JobManagers use it to decide who is the main JM and who are standby JMs. TaskManagers use it to discover the leading JobManager that they should connect to.

 

I’m also cc’ing Till, who should know this stuff better and can maybe explain it in a bit more detail.

 

Best,

Aljoscha

On 1. May 2017, at 18:59, Jain, Ankit <[hidden email]> wrote:

 

Hi fellow users,

We are trying to straighten out high availability story for flink.

 

Our setup includes a long running EMR cluster, job submission is a two-step process – 1) Flink cluster is first created using flink yarn client on the EMR cluster already running 2) Flink job is submitted.

 

I also saw references that with 1.2, these two steps have been combined into 1 – is that change in FlinkYarnSessionCli.java? Can somebody point to documentation please?

 

W/o worrying about Yarn RM (not Flink Yarn RM that seems to be newly introduced) failure for now, I want to understand first how task manager & job manager failures are handled.

 

My questions-

1)       https://cwiki.apache.org/confluence/pages/viewpage.action?pageId=65147077 suggests a new RM has been added and now there is one JobManager for each job. Since Yarn RM will now talk to Flink RM( instead of JobManager previously), will Yarn automatically restart failing Flink RM?

2)       Is there any documentation on behavior of new Flink RM that will come up? How will previously running JobManagers & TaskManagers find out about new RM?

3)       https://ci.apache.org/projects/flink/flink-docs-release-1.3/setup/jobmanager_high_availability.html#configuration requires configuring Zookeeper even for Yarn – Is this needed for handling Task Manager failures or JM or both? Will Yarn not take care of JM failures?

 

It may sound like I am little confused between role of Yarn and Flink components– who has the most burden of HA? Documentation in current state is lacking clarity – I know it is still evolving.

 

Please let me know if somebody can help clear the confusion.

 

Thanks

Ankit