[ 
https://issues.apache.org/jira/browse/RATIS-1534?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Junhua Zhang resolved RATIS-1534.
---------------------------------
    Resolution: Done

> SegmentedRaftLogWorker should enforce a minimum time interval between flush 
> calls
> ---------------------------------------------------------------------------------
>
>                 Key: RATIS-1534
>                 URL: https://issues.apache.org/jira/browse/RATIS-1534
>             Project: Ratis
>          Issue Type: Improvement
>          Components: server
>            Reporter: mingchao zhao
>            Assignee: Junhua Zhang
>            Priority: Major
>             Fix For: 2.3.0
>
>         Attachments: image-2022-03-03-16-23-32-929.png, screenshot-1.png
>
>          Time Spent: 1h 50m
>  Remaining Estimate: 0h
>
> In our previous test([see streaming test 
> doc|https://docs.google.com/document/d/1mS3GqovQ3D1b7V0L3--VF9xhl5jdId1mSL0cQNb7uHo/edit]),
>  we found that SegmentedRaftLogWorker writing RaftLog seriously affected the 
> performance of streaming. Mainly because the flush action was very frequent.
> So we can change flush to async. This will greatly improve performance. Of 
> course we'd better make this configurable, which is disable by default.
> !screenshot-1.png!
> We previously tried to reduce disk IO by introducing a minimum interval 
> between flushes, the relevant pr is here: 
> [https://github.com/apache/ratis/pull/611]
> However, after subsequent tests, we found that the raft performance has 
> degraded,  because {color:#FF0000}flushIfNecessary(){color} function does 
> {color:#FF0000}more operations{color} than just flushing the stream:
> !image-2022-03-03-16-23-32-929.png|width=741,height=593!
> Since we enforced the minimum time for each flush, the raft log to can't be 
> committed in flush intervals, which reduces the performance of raft.
> So we want to separate out.flush()  from flushIfNecessary() and execute it 
> asynchronously, so that the process of writing to disk will not block the 
> operation of other threads.



--
This message was sent by Atlassian Jira
(v8.20.1#820001)

Reply via email to