[
https://issues.apache.org/jira/browse/BOOKKEEPER-816?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15968064#comment-15968064
]
ASF GitHub Bot commented on BOOKKEEPER-816:
-------------------------------------------
Github user athanatos commented on the issue:
https://github.com/apache/bookkeeper/pull/80
@sijie @eolivelli If I'm reading this right, you are trying to use
sync_file_range to avoid calling FileChannel.force, which presumably calls
fsync. You actually can't use sync_file_range to ensure durability -- all it
does it write out the dirty pages, it won't write out allocation information
and (crucially) it won't issue the barriers required to flush volatile caches.
See http://man7.org/linux/man-pages/man2/sync_file_range.2.html,
particularly:
"This system call does not flush disk write caches and thus does not
provide any data integrity on systems with volatile disk write caches."
See also http://www.spinics.net/lists/ceph-devel/msg03995.html wherein
Christoph Hellwig removes the same behavior in Ceph.
> use native fallocate & sync_file_range to improve journal allocation
> --------------------------------------------------------------------
>
> Key: BOOKKEEPER-816
> URL: https://issues.apache.org/jira/browse/BOOKKEEPER-816
> Project: Bookkeeper
> Issue Type: Improvement
> Components: bookkeeper-server
> Reporter: Sijie Guo
> Assignee: Sijie Guo
> Fix For: 4.5.0
>
>
> it'd better to leverage filesystem fallocate & sync_file_range for journal
> performance.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
