activemq-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Ganesh (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (AMQ-4705) Add keep alive support to shared file locker
Date Tue, 12 May 2015 18:02:59 GMT

    [ https://issues.apache.org/jira/browse/AMQ-4705?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14540364#comment-14540364
] 

Ganesh commented on AMQ-4705:
-----------------------------

Checking of the file exists and comparing the lastModified date  (time since epoch) of the
lock file should take care of the following cases. 

1. Single broker configuration -The lock file is deleted - The file.exists() returns false
and the broker stops
2. A master slave broker configuration - The master starts up, creates the lock file and locks
it. The slave broker is constantly polling to see if it can obtain a lock. Now, the lock file
is deleted, the slave starts up as the master (polling times specified by the lockKeepAlivePeriod
and lockAcquireSleepInterval) and the master compares the date modified on the lock file and
then shuts down.

Please see attachment .


> Add keep alive support to shared file locker
> --------------------------------------------
>
>                 Key: AMQ-4705
>                 URL: https://issues.apache.org/jira/browse/AMQ-4705
>             Project: ActiveMQ
>          Issue Type: Bug
>          Components: Message Store
>    Affects Versions: 5.8.0
>            Reporter: Gary Tully
>            Assignee: Gary Tully
>              Labels: kahadb, netapp, nfsv4, shared-file-lock
>             Fix For: 5.9.0
>
>
> issue on nsfv4 with a master slave configuration, where both the slave and the master
could obtain a lock.
> The following events occurred:
> * master locks the file - does no more i/o to it – it's passive wrt the lock
> * slave asks every 10 seconds if it can get the lock nfs come back and say no, someone
has it
> * nfs dies not nicely
>  ** nfsv4 is stateful - no callback for locks. 
>  ** It has a grace period of 30 seconds to let all clients that had locks reclaim them
as locked
> * master does not realize it needs to reclaim the lock and continues under the assumption
it has the lock.
> * After 30 sec grace period, slave comes in and asks for the lock and it receives it.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Mime
View raw message