[
https://issues.apache.org/jira/browse/HBASE-28809?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17878675#comment-17878675
]
guluo commented on HBASE-28809:
-------------------------------
If the server disables the flush procedure, the client will receive a
DoNotRetryIOException. The client will catch this exception and fall back to
the old implementation.
----------------------
Can optimization be done for this situation?
What I mean is that when the server disables Flush Procedure, the client would
always send two requests for flushing just like you said. Can we optimize it to
only send once?
thanks a lot！
> Client always send flush requests twice when flush procedure is disabled
> ------------------------------------------------------------------------
>
> Key: HBASE-28809
> URL: https://issues.apache.org/jira/browse/HBASE-28809
> Project: HBase
> Issue Type: Improvement
> Components: Client
> Affects Versions: 2.6.0, 4.0.0-alpha-1
> Environment: hbase 2.6.0
> centos7
> Reporter: guluo
> Assignee: guluo
> Priority: Major
>
> When hbase disable flush procedure at server side, whether flush procedure is
> diabled or not at clien side, client always sends flush requests twice.
> In current hbase, There are four situations about flush request:
> 1. Both server and client enables flush Procedure, Client only sends a flush
> request by flush procedure.
> 2. Server disabled flush procedure and client enables flush procedurce,
> Client sends a flush request by flush procedure, after failure ,it will sends
> a flush request by old flush implementation
> 3. Server enables flush procedure and client disables flush procedure, Client
> only send a flush request by flush procedure
> 4. Both server and client disables flush Procedure, Client alwalys sends a
> flush request by flush procedure at first, after failure ,it will sends a
> flush request by old flush implementation
>
> The situation 1/2 is no problem.
> For situation 3, I think client should only send a flush request by old flush
> implementation instead of flush procedure
> For situation 4, I think client should only send a flush request by old flush
> implementation, instead of first using flush procedure, and then switching to
> old flush implementation after failing
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
