[
https://issues.apache.org/jira/browse/HDDS-1086?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16772427#comment-16772427
]
Hanisha Koneru commented on HDDS-1086:
--------------------------------------
[~shashikant], even with Short Circuit read/ write in Ratis, we would still
have the extra hop from a client talking to a non leader OM.
Lets say OM1 is the Leader OM and we a have client sending requests to OM2.
OM2's raft client would redirect the request to OM1 server. The client will not
know that OM2 is not the leader and keep sending subsequent requests also to
this follower node. This extra hop for every client request can be avoided if
we submit requests directly to Ratis server and handle required features
(Retries, Failover etc.) on OM client.
> Remove RaftClient from OM
> -------------------------
>
> Key: HDDS-1086
> URL: https://issues.apache.org/jira/browse/HDDS-1086
> Project: Hadoop Distributed Data Store
> Issue Type: Sub-task
> Components: HA, OM
> Reporter: Hanisha Koneru
> Assignee: Hanisha Koneru
> Priority: Major
> Attachments: HDDS-1086.001.patch
>
>
> Currently we run RaftClient in OM which takes the incoming client requests
> and submits it to the OM's Ratis server. This hop can be avoided if OM
> submits the incoming client request directly to its Ratis server.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
