[ale] cyrus - slow reconstruct of mailboxes.db

Chris Ricker kaboom at gatech.edu
Wed May 5 22:26:32 EDT 2004


On Mon, 3 May 2004, J.M. Taylor wrote:

> Anybody run cyrus for delivery on upwards of 10,000 mailboxes?  How long 
> does it take to start up master after a catastrophic failure assuming its 
> running ctl_cyrusdb -r?

A while ;-). What kind of hardware? How often do you checkpoint?

> Does ctl_cyrusdb -r walk every user mailbox and do a thorough reconstruct?

ctl_cyrusdb -r recovers the db logs since the last db checkpoint....  
reconstruct can be used to rebuild user mailboxes.

> Anybody know how to speed this up, or keep a secondary, failover 
> mailboxes.db that can be moved in to replace the corrupt one in event of 
> a catastrophic failure?

You can use ctl_mboxlist to export the db to flat text. When you restore, 
just import the text back, something like:

su - cyrus
ctl_mboxlist -d > /backups/mboxdump.txt
....
ctl_mboxlist -u < /backups/mboxdump.txt

You can back up the actual file as well, but that only works to restore if 
you're on the same version, platform, etc. (just like most binary db 
files), so the export and import is the far safer way to go.

later,
chris



More information about the Ale mailing list