[
https://issues.apache.org/jira/browse/ARTEMIS-2193?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16714406#comment-16714406
]
Francesco Nigro commented on ARTEMIS-2193:
------------------------------------------
[~kkaczkow] I have implemented a test with OpenWire, no transactions/durability
and reusing your same xml configuration with no success.
In addition (to check for any leaks) I have used 100*1024 bytes Bytes messages
with 1 producer 1 consumer where the consumer consumes 1 message every 100 ms
using an heap of 30 MB and 15 MB of address size on #.
The test has finished after letting running for the whole weekend without
getting any OOM.
My suppositions now are 2:
- your disk is too slow (and large messages are using the disk with OpenWire)
making the store requests to pile up
- the heap is just not enough for the topic size when are getting filled with
messages (before starting to drop messages)
So I believe that you can try 2 separate ways to verify it:
1) turn journal-datasync = false, journal-buffer-timeout = 0 and using MAPPED
(or NIO) as journal type: it would emulate a much faster disk
2) just increase the heap size (we need to figure it out by how much)
> Artemis fails on OutOfMemoryError with fast producers
> -----------------------------------------------------
>
> Key: ARTEMIS-2193
> URL: https://issues.apache.org/jira/browse/ARTEMIS-2193
> Project: ActiveMQ Artemis
> Issue Type: Bug
> Components: Broker, OpenWire
> Affects Versions: 2.7.0, 2.6.3
> Reporter: Konrad Kaczkowski
> Priority: Critical
> Labels: performance
> Attachments: artemis.log, broker.xml, jvm-gc.log, resources_usage.png
>
>
> The broker fails on OOM Error in a couple of hours of testing with fast
> producers:
> * The Artemis broker with heap size of 2Gb, global-max-size of 400Mb, and
> global address-setting to apply DROP policy whenever memory exceeds 150Mb
> (max-size-bytes).
> - Two fast producers sending 100 KB messages to 2000 topics every 100 ms
> each
> - Client used: ActiveMQ 5.15.2 JMS
> - Tested on Artemis 2.6.3 and 2.7.0-SNAPSHOT
> - Testing time: 2 - 10 h until OOME occurs
> The heap size grows continuously until it runs of out memory. I tested
> DROP/PAGE/FAIL strategies with no difference. Attached are the log and the
> broker.xml configuration.
>
> Expectation: Having sharp memory limits the broker should handle with
> dropping messages on time.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
