proxyd cpu usage

[Bron Gondwana]

On Mon, Dec 07, 2009 at 01:37:28PM -0800, Andrew Morgan wrote:
> On Mon, 7 Dec 2009, John Madden wrote:
> 
> >> At a guess, it sounds like load from LIST.  You should be able to see
> >> what's causing the load if you have several seconds.  For example,
> >> enable telemetry and look for long turn around.  Or use strace (or
> >> equiv).  Have you experimented with foolstupidclients?  What is your
> >> client mix?  How many mailboxes do you have?
> >
> > I set up a second frontend to analyze a single client.  Strace shows the
> > dumb:
> >
> > read(0, "00000002 LIST \"\" INBOX.*\r\n", 4096) = 26
> > read(0, "00000003 LIST \"\" user.*\r\n", 4096) = 25
> > read(0, "00000004 LIST \"\" *\r\n", 4096) = 20
> >
> > I tested fullstupidclients on the production frontend and it has cut
> > load significantly but that LIST "" user.* call is expensive (0.950 secs
> > 1 calls).  I assume that's to poll for shared folders and I'm going to
> > look into whether or not it can be eliminated.  I don't suppose there's
> > another fullstupidclients that improves responsiveness on this call, is
> > there?
> 
> Those LIST queries seem a little odd coming from a normal user account in 
> Cyrus.  Are you logging in as a Cyrus admin account?  Why is the client 
> asking for "user.*"?

By "the client" I suspect we're talking Thunderbird here?  Because that's
one of the namespaces.  Yeah, it's pretty dumb.  And yes, it's for shared
folders.

Bron ( the "*" request is particularly classy! )