[Kolab-devel] 10.000 events in a Resource Calendar

[Thomas Jarosch]

Hello Jeroen,

On Tuesday, 22. May 2012 14:57:41 Jeroen van Meeuwen (Kolab Systems) wrote:
> On 2012-05-22 13:46, Thomas Jarosch wrote:
> > You might want to reconsider "freebusyd", which solved that exact
> > problem for years now: https://roundup.kolab.org/issue2554
> > 
> > After we introduced freebusyd, updating the freebusy data
> > was never an issue again.
> > 
> > Freebusyd + caching of the Horde Kolab classes easily handles
> > updates to calendars with 10.000 events in no time at all.
> 
> Very interesting, I wasn't aware of its existence to be honest.
> 
> How much effort would it be to rebase on Cyrus IMAP master, noted it is
> currently targeted to ship CalDAV with RFC822/ical (not xcal, I'm
> afraid) messages in the spool itself, and IMAP NOTIFY notifications are
> going to be merged in by Andreas Osowski soon enough?

I have an updated "notification" patch against cyrus 2.4.x, though it hasn't
seen production use yet since we still run cyrus 2.3.x on our system.

> Could this include inner-mailbox mutations such as delete/add messages
> as well (using any of the aforementioned new logic, or the existing
> replication capabilities in Cyrus IMAP perhaps)? I've barely glanced
> over the patches attached to the ticket and I'm recognizing only delete
> and rename folder hooks.

It theoretically could. Currently I don't see a performance problem
with add or delete of messages, the pfb.php cache is efficient enough.
Only large folder moves or deletes were a problem, so we just
patched notification in for that.

A proper notification bus could of course replace the "notification" hack.


The problem with freebusyd is that we (=Intra2net) don't run a full Kolab 
system, so we can't do any upstream porting. To be precise we just use
cyrus-imapd, the freebusy module and Horde.

freebusyd certainly needs work for full Kolab integration as discussed
in the issue tracker. Before we introduced freebusyd in our system,
our customers regularly killed their servers with huge freebusy updates.
So far the performance problem didn't seem painful enough for
upstream kolab to fix this issue properly, at least nobody
stepped up to take a closer look at freebusyd for years.

Cheers,
Thomas