I'm performing a migration from 2.2.8 on an E450 running Solaris 10 Beta Build 58 to 2.3.9 on a T2000 running Solaris 10 latest. The former isn't zoned (it wasn't available in that build), isn't smf'd (ditto) and isn't ZFS'd (ditto).
The new machine has Cyrus in a zone, rooted on ZFS, with the local storage also on ZFS. The mailboxes themselves (35K mailboxes, a couple of terabytes) are on NFS storage (some on Sun, mostly on a Pillar), so I'm testing the new configuration by performing a metadata migration from read-only mounts. I've written a script to perform the migration in an rsync style, paralleled over the T2000, which is able to sync the metadata between the live storage and the new metadata partition in 35 seconds. Just to ensure the best health and hygiene, I'm rebuilding as many databases as I can. The mailboxes file is the result of ctl_mboxlist -d | ctl_mboxlist -u, the sasldb is the result of db_dump -p and db_load, etc, etc. However, the one that seems resistant to this is deliver.db (skiplist format). cvt_cyrusdb ... skiplist ... flat takes five hours before failing, and although ctl_deliver -d appears to run quickly and correctly, I can't see an immediately obvious way to reload it. Obviously, this isn't a big problem: I can just shut down the old Cyrus instance, copy the file over, and all is well. But having databases de-fragmented, built with the same bits that are in production, is a nice confidence booster if it's possible. ian ---- Cyrus Home Page: http://cyrusimap.web.cmu.edu/ Cyrus Wiki/FAQ: http://cyrusimap.web.cmu.edu/twiki List Archives/Info: http://asg.web.cmu.edu/cyrus/mailing-list.html
