Questions before upgrading vom 2.3 to 2.4

[Pascal Gienger]

Le 05/01/2012 17:00, Sebastian Hagedorn a écrit :
> Hi,
>
> we're currently running a rather outdated configuration, with Cyrus
> 2.3.14 and RHEL 3 i386 on hardware that's almost 7 years old. I'm
> planning an upgrade to 2.4 with RHEL 5 x86_64 and I have a umber of
> questions regarding new features that I played around with using 2.4.13.
> I've searched the archives and found some answers, but not all.

In principle it is simple, but I hit a hard ground today. I did the 
mistake to miscalculate the "converting indexes"....

Our storage was perfectly sized to handle the needed IOPS for cyrus 
operation, but having 8000 index files being rebuild simultaneosuly...

Important:

Set lmtp_mail_timeout (when you use postfix to deliver mail) to a low 
value, like 2seconds, because also cyrus lmtpd triggers the index 
rebuild of a INBOX. So people with less mail in their INBOX (which 
already constructed index and cache) still get their mail in their 
INBOX, and people having MANY mails so that the index rebuild takes time 
will get the mail later due to the early deferred state (after 
lmtp_mail_timeout). Otherwise you'll have many blocked Postfix lmtp and 
no mail gets transferred.

I tackled the situation like this, mail is being delivered normally now 
and there are 5-10 index rebuilds running only which are going fast and 
without noticeable delay. People having large mailboxes will get the 
mail on the 2nd queue run of Postfix (set the delay not too high!).

On our test system all worked so well, with many connections, much load 
to the limit, and no errors - but only on a few mailboxes so index 
rebuild was little compared to the real case. It was my fault, my error. 
Should not have happened, but ok, no mail was lost.


And be familiar with

expunge_mode:
delete_mode:
expunge_days:

and the consequences on cyr_expire!


With meta-partition on a fast SSD device this even would not have 
occured (500 GB SSD needed in our case....)


Lucky eye: Cyrus 2.4.13 is somewhat faster with our SOGo system (i 
presume STATUSCACHE).

Crying eye: Some people suggest moving to Google or Microsoft because 
that would not cause delays at an upgrade like this...

Should you use SOGo other other webmailers, please set

flushseenstate: 1
statuscache: 1


Put your cyrus data directory + your meta partition(s) on a fast 
LUN/Volume, and/or have much RAM for your file system cache (ARC in my 
case).


-- 
Pascal Gienger     Jabber/XMPP/Mail: pascal.gienger at uni-konstanz.de
University of Konstanz, IT Services Department ("Rechenzentrum")
Electronic Communications and Web Services
Building V, Room V404, Phone +49 7531 88 5048, Fax +49 7531 88 3739