Francesco Nigro commented on ARTEMIS-1865:

Given that currently it just use one file with [5 different 
 used for:
 * shared state ([LIVE, FAILINGBACK, STARTED, 
 - 1 byte -
 * live lock - 1 byte -
 * backup lock - 1 byte -
 * shared nodeId (UUID) - 16 byte -

I think that it could be splitted into 3 files:
 * one with the shared state and nodeId (server.state)
 * one with the live lock (live.lock)
 * one with the backup lock (backup.lock)

That should avoid strange locking issues to happen due to not obvious CIFS 

The only problems I see are related to the retro-compatibility and 
cross-compatibility (ie a new version can't work with an old one etc etc).

I don't have any box to reproduce the CIFS behaviour ATM, but it would be good 
to create 2 simple programs that try to acquire locks on different regions on 
the same remote file and validate that's the real problem.

AFAIK NFS doesn't seem to have similar issues.



> Shared Store Cluster Doesn't Work on CIFS
> -----------------------------------------
>                 Key: ARTEMIS-1865
>                 URL: https://issues.apache.org/jira/browse/ARTEMIS-1865
>             Project: ActiveMQ Artemis
>          Issue Type: Bug
>          Components: Broker
>    Affects Versions: 2.5.0
>         Environment: RHEL 6.2
>            Reporter: Ilkka Virolainen
>            Priority: Minor
>         Attachments: broker_a.xml, broker_b.xml
> When Artemis is configured as a shared store master/slave -pair with the 
> journal saved on a CIFS share only the first instance is able to start. The 
> instance started later will fail the acquire a lock to journal/server.lock 
> file and will start in an invalid state. Similar shared store master/slave 
> -configuration works correctly with 5.14.5.

This message was sent by Atlassian JIRA

Reply via email to