mupdate cpu, thread timeouts

Wesley Craig wes at umich.edu
Mon Jul 12 23:10:20 EDT 2010


On 12 Jul 2010, at 15:49, John Madden wrote:
> ...Meaning the error is nothing to worry about?  This is on RHEL  
> 5.5 if that helps.

The "error" is a debug statement.  The CPU usage reflects poorly on  
the pthreads implementation.

> Can any of this be tweaked to, for example, wait longer before  
> thinking the connection to the master has been lost?

You can set mupdate_retry_delay, which controls both how long a NOOP  
is waited for AND how long to sleep between connection retries (+ up  
to 50% more).  The default is 20 seconds.  This is also how long  
imapd et al may be forced to wait if they need to kick the mupdate  
slave and things are hung.  Of course, if things are hung, there are  
perhaps bigger problems.

> It might be nice to be able to re-sync a bunch of slaves while  
> keeping changes locked, or even to do re-syncs with a MVCC sort of  
> model where "here are the changes that were made during your  
> resync" can be sent when re-sync finishes.  I could then throw 8  
> cores at the master and hopefully avoid these thrashing situations.

Where there's a programmer, there's a way...

:wes


More information about the Info-cyrus mailing list