RE: [Xen-API] [RFC][PATCH] Deadlock in RRD handling during master transition

[Dave Scott <Dave.Scott@xxxxxxxxxxxxx>]

Hi Ted,

Thanks for diagnosing this problem and sending the patch.

I think this might be the same deadlock that is fixed by:

http://xenbits.xensource.com/XCP/xen-api.hg?rev/793489e20822;style=rev%3DCA

Although I think the problem is now fixed in trunk, it seems to me that this 
problem might recur: it's very easy to forget you have the RRD mutex held and 
accidentally query the master's db. So I think it's probably best to adopt a 
'belt-and-braces' approach and take your patch too.

Thanks,
Dave

> -----Original Message-----
> From: xen-api-bounces@xxxxxxxxxxxxxxxxxxx [mailto:xen-api-
> bounces@xxxxxxxxxxxxxxxxxxx] On Behalf Of Ted Lin
> Sent: 09 August 2010 15:52
> To: xen-api@xxxxxxxxxxxxxxxxxxx
> Subject: [Xen-API] [RFC][PATCH] Deadlock in RRD handling during master
> transition
> 
> The codebase that this XAPI patch is based on is a bit old (5.5.0;
> George), but I think some form of it still exists in the latest
> codebase.
> 
> What we're seeing is that if the master host goes away during slave RRD
> processing, it will prevent the slave XAPI from transitioning to master.
> The patch adds a try-lock in the RRD backup so if the lock is still
> held, the backup is skipped.
> 
> 
> I'm wondering if this is a safe change (at least for 5.5.0), and if
> anyone has ever run into this before?  (An unreliable way of generating
> the situation is to pull the management link during a migration.)
> 
> 
> 
> Thanks for the help,
> 
> --
> Ted M Lin

_______________________________________________
xen-api mailing list
xen-api@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/mailman/listinfo/xen-api