Remote mailbox creation error / unified murder

Wesley Craig wes at umich.edu
Mon Jul 2 13:23:25 EDT 2007


On 02 Jul 2007, at 06:53, Janne Peltonen wrote:
> But when try to imapsync the old server's data to the new one, with  
> only
> one multi-threading mupdate master process, I keep getting 'connection
> to the master timed out' errors from the mupdate slaves.

During your imapsync process, I might disable murder.  Of course,  
it's a great load test!  Is mupdate logging anything when the clients  
time out?

> mupdate_connections_max: 1024
> mupdate_workers_max: 500
> mupdate_workers_maxspare: 100
> mupdate_workers_minspare: 20
> mupdate_workers_start: 50
>
> I'll have some 50000 (fifty thousand) users with some 600000 (six
> hundred thousand) mailboxes, with max abt 6000 users active at any  
> given
> time. And the murder has 24 nodes.

UMich's system is somewhat bigger than this.  We invoke mupdate with  
this line:

         mupdate         cmd="mupdate -m" listen="mupdate" prefork=1  
maxfds=1024

and imapd.conf:

	mupdate_connections_max: 1000
	mupdate_workers_max: 200

Of course, while our backends and replicas are running 2.3.x, our  
mupdate master is running 2.2.x.  We're in the process of moving to  
2.3.x. for the mupdate master and frontends.

:wes


More information about the Info-cyrus mailing list