> > Do the processes seem to grow on a runaway basis, or do they reach a > > maximum? If it's a per-connection leak, then we should see > > process sizes > > grow towards a maximum that's proportional with the maxconnects value. > > looks like runaway, but slower when limiting the maxconnects value. > - well this is a very limited box ( pII 400, 192 MB, no swap, running > a FreeBSD 5.3 based LiveCD with postfix's mailqeue in a 64MB memory disc, > specs http://mailsat.dstm.de/sysinfo/ ) so it's not easy to figure if > there is an upper limit, because system limits are reached very early.
There seems to be a upper limit. I've attached my swap usage graph of the last three months. This is a 2 GHz Celeron with 512 MB RAM that's mainly used as dbmail server. The zero swap values at the end of october and the beginning of november were because of reboots. dbmail is a cvs version from 04-09-22 so I have dynamic preforking: [LMTP] NCHILDREN=4 MAXCHILDREN=30 MAXCONNECTS=1000 MINSPARECHILDREN=2 MAXSPARECHILDREN=4 Thomas -- http://www.tmueller.com for pgp key (95702B3B)
swap-quarter.png
Description: PNG image
