[
https://issues.apache.org/jira/browse/YARN-6426?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15952779#comment-15952779
]
Arun Suresh commented on YARN-6426:
-----------------------------------
[~roniburd], thanks for the patch.
I am sure compressing would help reduce the size. But there are a couple of
things to take into account though:
Upgrades for example. For 3.0.0-alpha3, it might not be a big problem (since we
don't guarantee upgrades from 2.x to 3.x). But if we targeting 2.8.1 / 2.9.0,
we need to figure out a way for:
# The 2.8.1 / 2.9.0 RM to be able to read uncompressed AppStateDate znodes
after upgrade.
# The 2.8.0 RM should be able to ignore compressed AppStateData it gets from ZK
after a downgrade.
> Compress ZK YARN keys to scale up (especially AppStateData
> ----------------------------------------------------------
>
> Key: YARN-6426
> URL: https://issues.apache.org/jira/browse/YARN-6426
> Project: Hadoop YARN
> Issue Type: Improvement
> Components: yarn
> Affects Versions: 3.0.0-alpha2
> Reporter: Roni Burd
> Assignee: Roni Burd
> Labels: patch
> Attachments: zkcompression.patch
>
>
> ZK today stores the protobuf files uncompressed. This is not an issue except
> that if a customer job has thousands of files, AppStateData will store the
> user context as a string with multiple URLs and it is easy to get to 1MB or
> more.
> This can put unnecessary strain on ZK and make the process slow.
> The proposal is to simply compress protobufs before sending them to ZK
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
