GitHub user kawamuray opened a pull request:
https://github.com/apache/kafka/pull/2352
KAFKA-4614 Forcefully unmap mmap of OffsetIndex to prevent long GC pause
Issue: https://issues.apache.org/jira/browse/KAFKA-4614
Fixes the problem that the broker threads suffered by long GC pause.
When GC thread collects mmap objects which were created for index files, it
unmaps memory mapping so kernel turns to delete a file physically. This work
may transparently read file's metadata from physical disk if it's not available
on cache.
This seems to happen typically when we're using G1GC, due to it's strategy
to left a garbage for a long time if other objects in the same region are still
alive.
See the link for the details.
You can merge this pull request into a Git repository by running:
$ git pull https://github.com/kawamuray/kafka
KAFKA-4614-force-munmap-for-index
Alternatively you can review and apply these changes as the patch at:
https://github.com/apache/kafka/pull/2352.patch
To close this pull request, make a commit to your master/trunk branch
with (at least) the following in the commit message:
This closes #2352
----
commit 8c3a4c9f5a6188c641a45cb3e111094f545f7e66
Author: Yuto Kawamura <kawamuray.dad...@gmail.com>
Date: 2017-01-11T08:10:05Z
KAFKA-4614 Forcefully unmap mmap of OffsetIndex to prevent long GC pause
----
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at infrastruct...@apache.org or file a JIRA ticket
with INFRA.
---
