[
https://issues.apache.org/jira/browse/ARTEMIS-1108?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15965924#comment-15965924
]
ASF GitHub Bot commented on ARTEMIS-1108:
-----------------------------------------
Github user bgutjahr commented on the issue:
https://github.com/apache/activemq-artemis/pull/1194
I have now setup a GFS2 file system, shared between 2 nodes (with CentOS
6.9, gfs2-utils-3.0.12.1, most according to this description:
https://hungred.com/how-to/setup-gfs2-gfs-linux-centos/). HA failover from life
to backup server worked well multiple times.
BTW: it also worked with NIO. Initially NIO was used because
libartemis-native-64.so version 1.5.4, which I got from mvnrepository.com, has
a dependency to glibc-1.14. But CentOS 6 has glibc-1.12, so the library wasn't
loaded and Artemis fell back to NIO. I had to compile the library on CentOS 6
to get AIO working.
Nevertheless, this was just a quick test that the failover worked. I have
no idea how/where the GFS2 bug affected Artemis.
> Shared-store with AIO fails to lock lockfile on NFS shares
> ----------------------------------------------------------
>
> Key: ARTEMIS-1108
> URL: https://issues.apache.org/jira/browse/ARTEMIS-1108
> Project: ActiveMQ Artemis
> Issue Type: Bug
> Components: Broker
> Affects Versions: 1.5.4, 2.0.0
> Environment: Linux with NFS-mounted shared-store directory
> Reporter: Bernd Gutjahr
> Priority: Minor
>
> When we setup shared-store backup servers with the shared store path pointed
> to an NFS-mounted directory, file locking didn't work until we disabled AIO
> and used NIO instead. Although NFS is not recommended and we don't want to
> use in production, we wanted to use NFS in some test or demo setups just for
> simplicity.
> As it turned out, the reason why it doesn't work with AIO is that flock is
> used instead of lockf. According to the code comments, the class
> AIOFileLockNodeManager had been added due to a GFS2 bug,
> https://bugzilla.redhat.com/show_bug.cgi?id=678585. But that GFS2 bug has
> already been fixed more than 5 years ago, end of 2011. So I think it would be
> time to remove the flock-based workaround.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
