[
https://issues.apache.org/jira/browse/HDDS-9392?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17772115#comment-17772115
]
Ivan Andika edited comment on HDDS-9392 at 10/5/23 8:25 AM:
------------------------------------------------------------
Hi [~szetszwo] [~Nicholas Niu] [~ckj], may I ask about your opinions on this?
Whether pipeline mode / even star mode should sort the datanode so that the
primary node is the nearest to client?
was (Author: JIRAUSER298977):
Hi [~szetszwo] [~Nicholas Niu] , may I ask about your opinions on this? Whether
pipeline mode / even star mode should sort the datanode so that the primary
node is the nearest to client?
> Streaming write pipeline should pick the nearest datanode as primary node
> -------------------------------------------------------------------------
>
> Key: HDDS-9392
> URL: https://issues.apache.org/jira/browse/HDDS-9392
> Project: Apache Ozone
> Issue Type: Sub-task
> Components: OM, Ozone Client, SCM
> Affects Versions: 1.3.0
> Reporter: Ivan Andika
> Assignee: Ivan Andika
> Priority: Major
>
> In `BlockDataStreamOutput#setupStream`, if pipeline mode is enabled, we pass
> the Pipeline object to the `RatisHelper#getRoutingTable`.
> In `RatisHelper#getRoutingTable`, we pick the first datanode in the pipeline
> as the primary, which from my understanding should be the Datanode closest to
> the client. However, I'm unable to find the logic that sort the datanodes in
> the pipeline.
> From what I understand, when we open the stream key and request SCM to
> allocate the block (`OMKeyRequest#allocateBlock`), we need to also pass the
> clientMachineString so that so that nearest datanodes will be used instead.
> This should improve the performance of the stream pipeline mode.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
