[
https://issues.apache.org/jira/browse/HDDS-5630?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17410838#comment-17410838
]
Neil Joshi commented on HDDS-5630:
----------------------------------
Hi guys - [~kerneltime] , thanks for bringing this up the other day. I'd like
to help and I am looking at the problem you described. On the 'cached client'
patch provided in this Jira, two things to consider:
i.) the OzoneClientProducer is RequestScoped through java CDI, so you can
expect a new instance of the OzoneClientProducer for every s3 request, this
results in the cache created for every request (defeats the purpose of this
cache with this configuration (?))
ii.) it appears a 'Client' class is created for each request, resulting in a
new socket connection open for every request regardless if it is in the cache
(the 'Client' is not closed if found in cache).
> S3G keep connections to OM around leading to slowness
> -----------------------------------------------------
>
> Key: HDDS-5630
> URL: https://issues.apache.org/jira/browse/HDDS-5630
> Project: Apache Ozone
> Issue Type: Bug
> Components: S3
> Reporter: Ritesh H Shukla
> Assignee: Ritesh H Shukla
> Priority: Major
> Attachments: ClientCache20210902.patch
>
>
> S3G currently creates one Ozone Client per requests. The client uses Hadoop
> RPC to communicate with OM. The TCP connections with OM are not cleaned up
> post completion of the S3 requests which leads to large number of TCP
> connections to be open between S3G and OM.
> https://issues.apache.org/jira/browse/HDDS-4440 will eventually replace the
> use of Hadoop RPC but in the mean time it is prudent to look into how the
> current RPC usage can be improved.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
