[
https://issues.apache.org/jira/browse/ARTEMIS-2211?focusedWorklogId=189902&page=com.atlassian.jira.plugin.system.issuetabpanels:worklog-tabpanel#worklog-189902
]
ASF GitHub Bot logged work on ARTEMIS-2211:
-------------------------------------------
Author: ASF GitHub Bot
Created on: 25/Jan/19 08:53
Start Date: 25/Jan/19 08:53
Worklog Time Spent: 10m
Work Description: franz1981 commented on issue #2479: ARTEMIS-2211 Avoid
duplicate code for ByteBuffer pooling and alignment
URL: https://github.com/apache/activemq-artemis/pull/2479#issuecomment-457500842
@michaelandrepearce @clebertsuconic We can talk about this one on the
issue/forum too: the thing is that currently we have yet several parts on
artemis where we rely on thread pool storage for direct `ByteBuffer`s:
1. write side of paging
2. write side of journal with NIO/MAPPED journal types are used
Considering that the context of execution in artemis is agnostic of the
executing `Thread` and that the same task could move between executing
`Thread`s, it could happen that after 1 minute of inactivity threads could die,
they'll leak the thread pool storage related to them (until the first minor GC).
If a new burst of activity will create new threads and no minor GC has
happened there risk is to have too many direct buffers alive that could lead
the OOM killer of the OS to kill the first process that allocate native memory.
For compaction, this has been addressed (by me) by allocating upfront the
compaction task a single direct buffer and reusing it for any compaction
sub-task, deallocating manually (like a C `free`) at the end: although it could
seems a bad solution, it avoid leaking several MBs of direct memory if the
compaction operations are handled by different threads each time. It would slow
down compaction, but given that compaction is not meant to be an hot path like
journal writes is best to have predictable memory footprint.
This same approach can't be used for the other 2 cases and I was thinking to
provide a lock-free generic implementation of direct buffer pools to address it
(as ASYNCIO does), but maybe there is a simpler and more effective solution:
feel free to give your opinion ;)
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
Issue Time Tracking
-------------------
Worklog Id: (was: 189902)
Time Spent: 40m (was: 0.5h)
> Avoid duplicate code for ByteBuffer pooling and alignment
> ---------------------------------------------------------
>
> Key: ARTEMIS-2211
> URL: https://issues.apache.org/jira/browse/ARTEMIS-2211
> Project: ActiveMQ Artemis
> Issue Type: Improvement
> Reporter: Francesco Nigro
> Assignee: Francesco Nigro
> Priority: Minor
> Fix For: 2.7.0
>
> Time Spent: 40m
> Remaining Estimate: 0h
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
