> So if there is anyone on the list that is using dbmail 2.0 > with PostgreSQL (8.x) with 5k+ users, and a data set of around > 20G would you contact me off list if you like. > We have around that for dataset/userbase, but are using dbmail 1.2. I'll answer your questions anyway. First off, we're running postgresql 7.4.7 on a debian system, adaptec hardware scsi raid5, 2.6 GHz P4, 2 G ram. We only offer POP.
> - How/if you backup or replicate your database We dump all tables but messageblks nightly, we dump all tables on friday morn (01:10), and it is in average done about 02:40, compressed size of 4.6GB. We only use pg_dump. > - How you deal with vacuum's (mine seem to take forever) We analyze hourly, vacuum users,pbsp,messages every six hours, vacuum entire db daily. We also reindex weekly. > - How/if you deal with a vacuum FULL. Since we got our maintenance routine down, we have never had to nor felt the need to vacuum full. There was a PG issue with earlier versions that necessitated this, but that was ages ago... > - What kind of schedule you use for dbmail-util We don't use dbmail-util directly for delete or purge. We "update messages set status=003::smallint where status=002::smallint" and "delete from messages where status=003::smallint" once daily (about one and a half hours apart, and about an hour and a half before the daily vacuum). > - The number of connections you have from dbmail and what kind > of load that produces. We have four servers attaching for smtp injection, and two servers attaching for POP retrieval. We remain <1 for load average during most production time. This, of course, goes up some during the vacuums and such, but those are relatively short lived (for example, we vacuum verbose, and the one that runs every six hours runs in 737 seconds). A quick look at the logs shows that for yesterday's log cycle, we injected message ids 49600239 - 49656408, and grepping for "logging out" to catch POP connections, there were ~56000. The postmaster will allow 512 connections, dbmail.conf has a MAXCHILDREN=100 on each of the (two) POP servers. postfix is set to default for all delivery concurrency limits (20). > - What kind of "bad things" happen to DBMail when your database > load gets rather high. The worst thing I've seen is that customer cry like little school girls. I've never seen, experienced, nor heard of lost data, even when the system load "idled" at about 24, and pop connections took upwards of 120 seconds to establish (for those patient enough to wait). If I missed anything, let me know. In fact, at looking over all this data, I may even start doing a full dump nightly (I wasn't even aware it was going that quickly :) Thanks! -- Dave Logan http://www.digitalcoven.com/ "No! Try not! Do. Or do not. There is no try." -- Yoda
