[
https://issues.apache.org/jira/browse/BOOKKEEPER-717?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13891031#comment-13891031
]
Sijie Guo commented on BOOKKEEPER-717:
--------------------------------------
{code}
If, in the first clause, the time since enqueue for the first entry is greater
than maxGroupWaitInMSec, then surely shouldFlush should be set to true?
{code}
No, the change here is to avoid this. for example, in the queue, you have 100
entries, a disk hiccup cause all 100 entries to be time out. if you set
shouldFlush to be true, then you will have 100 individual sync requests, which
each only sync one entry. that's bad for performance. the change is to group
all these 100 entries into one batch and sync them together.
> journal should look forward to group time-out entries
> -----------------------------------------------------
>
> Key: BOOKKEEPER-717
> URL: https://issues.apache.org/jira/browse/BOOKKEEPER-717
> Project: Bookkeeper
> Issue Type: Sub-task
> Components: bookkeeper-server
> Reporter: Sijie Guo
> Assignee: Sijie Guo
> Fix For: 4.3.0
>
> Attachments: BOOKKEEPER-717.diff
>
>
> journal should look a bit forward to group timeout entries, which avoid too
> much fsyncs with single entry impacting journal performance. so bookie could
> have sustained high throughput with low-latency.
--
This message was sent by Atlassian JIRA
(v6.1.5#6160)
