[
https://issues.apache.org/jira/browse/HDDS-8897?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17736837#comment-17736837
]
Stephen O'Donnell commented on HDDS-8897:
-----------------------------------------
[~adoroszlai] has posted a PR on this change that may help things, and we have
also committed a few other PRs in the last few days to reduce the
synchronization in the pipeline manager code paths. I have also posted
https://github.com/apache/ozone/pull/4968 to remove the node search from the
locked path in pipeline manager, and after it is committed we could change the
WritableECProvider to use this, hopefully reducing the overhead.
It should be possible to get the performance we need without moving to async
allocation and closing, which will add complexity and may well cause other edge
cases.
t would be great if you could capture a profile of performance problem as I
suggested earlier, as it would let us see for sure the slow part of the code.
> ec container has too high latency when allocate blocks
> ------------------------------------------------------
>
> Key: HDDS-8897
> URL: https://issues.apache.org/jira/browse/HDDS-8897
> Project: Apache Ozone
> Issue Type: Improvement
> Reporter: GuoHao
> Assignee: Stephen O'Donnell
> Priority: Blocker
> Labels: pull-request-available
> Attachments: image-2023-06-21-21-58-15-052.png,
> image-2023-06-21-21-58-50-518.png, image-2023-06-21-23-53-56-534.png
>
>
> ec container has too high latency when allocate blocks
>
> I use ec in a high concurrency situation, I configure the metric on scm, and
> as concurrency gradually increases, the latency of the allocate block
> increases.
>
> allocate block ops：
> !image-2023-06-21-21-58-50-518.png!
>
> average latency：
> !image-2023-06-21-21-58-15-052.png!
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
