[
https://issues.apache.org/jira/browse/HDDS-5393?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17370510#comment-17370510
]
UENISHI Kota commented on HDDS-5393:
------------------------------------
I did a one-day survey on this issue, and it turned out it's not that
straightforward like other read requests. Opening a new key on every overwrite
leads to potentially infinite increase of KeyLocations with a lot of duplicates
on blocks. IMHO it needs some consistent logic to merge those duplicated
KeyLocations with addressing concurrent uploads.
> Return latest version of key location for client on createKey
> -------------------------------------------------------------
>
> Key: HDDS-5393
> URL: https://issues.apache.org/jira/browse/HDDS-5393
> Project: Apache Ozone
> Issue Type: Improvement
> Components: Ozone Client
> Reporter: UENISHI Kota
> Priority: Major
> Fix For: 1.2.0
>
>
> HDDS-5243 was a patch for omitting unnecessary key locations for clients on
> reading. But the same warning of large response size observed in our cluster
> for putting data. The patch can also be ported for putting data, as long as
> until object versioning is supported.
> My hypothesis is: The large message was originally, and possibly maybe due to
> this warning and sudden connection close from client side on reading large
> message in Hadoop IPC layer, from Ozone Manager - which causes hopeless 15
> retries from RetryInvocationHandler. The retries create another entry in
> OpenKeyTable but they never moved to KeyTable because the key never gets
> commited.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
