Thread: svr1 mail timeouts ...
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 'k, after getting some real information from Stephan, took a look at the machine ... Postfix sets a default max # of concurrent incoming smtpd connections to 100, right now, its sitting at 91 ... I suspect that what is/was happening was that it was hitting 100 and stalling on the other connections, including the nagios check ... I've just raised it to 256, and we'll see if that doesn't eliminate the timeouts ... - ---- Marc G. Fournier Hub.Org Networking Services (http://www.hub.org) Email . scrappy@hub.org MSN . scrappy@hub.org Yahoo . yscrappy Skype: hub.org ICQ . 7615664 -----BEGIN PGP SIGNATURE----- Version: GnuPG v2.0.4 (FreeBSD) iD8DBQFG0IIa4QvfyHIvDvMRAkIxAJ4q0OTXOLdqhj3y0Cz6vVVurSeQhwCghbVI bf+GtRQeUN/zhSKp7xVeQAU= =ZppX -----END PGP SIGNATURE-----
On Sat, Aug 25, 2007 at 04:25:14PM -0300, Marc G. Fournier wrote: > 'k, after getting some real information from Stephan, took a look at the > machine ... Postfix sets a default max # of concurrent incoming smtpd > connections to 100, right now, its sitting at 91 ... > > I suspect that what is/was happening was that it was hitting 100 and stalling > on the other connections, including the nagios check ... > > I've just raised it to 256, and we'll see if that doesn't eliminate the > timeouts ... Interesting (and it hasn't bitched in a while, so it looks like the right direction). Any idea what's using up all those connections? 91 *incoming* connections is very much, especially since the box doesn't exactly *receive* a lot of mails (it sends a whole lot, but that's a different limit). (For example, I have a fairly high-traffic mail server that certainly receives at least an order of magnitude more email than svr1 does, and it's running at 5-6 receiving processes normally) Perhaps it's something a config change can rid us of? //Magnus
Magnus Hagander wrote: > On Sat, Aug 25, 2007 at 04:25:14PM -0300, Marc G. Fournier wrote: >> 'k, after getting some real information from Stephan, took a look at the >> machine ... Postfix sets a default max # of concurrent incoming smtpd >> connections to 100, right now, its sitting at 91 ... >> >> I suspect that what is/was happening was that it was hitting 100 and stalling >> on the other connections, including the nagios check ... >> >> I've just raised it to 256, and we'll see if that doesn't eliminate the >> timeouts ... > > Interesting (and it hasn't bitched in a while, so it looks like the right > direction). yeah I installed a fair bit of monitoring on that very subject over the weekend and it shows that we are hovering around 120 connections(less on the weekend) on average now. > > Any idea what's using up all those connections? 91 *incoming* connections > is very much, especially since the box doesn't exactly *receive* a lot of > mails (it sends a whole lot, but that's a different limit). I have not looked into the configuration at all - but are we doing something like greylisting/connection tarpitting there ? Stefan
On Mon, Aug 27, 2007 at 12:15:37PM +0200, Stefan Kaltenbrunner wrote: > >Any idea what's using up all those connections? 91 *incoming* connections > >is very much, especially since the box doesn't exactly *receive* a lot of > >mails (it sends a whole lot, but that's a different limit). > > I have not looked into the configuration at all - but are we doing > something like greylisting/connection tarpitting there ? Graylisting it does, but that doesn't tie up connection slots. Tarpitting would, but I don't think we use that. Marc can confirm that though. //Magnus
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 - --On Monday, August 27, 2007 10:21:00 +0200 Magnus Hagander <magnus@hagander.net> wrote: > Any idea what's using up all those connections? 91 *incoming* connections > is very much, especially since the box doesn't exactly *receive* a lot of > mails (it sends a whole lot, but that's a different limit). Of course it receives alot ... as an example ... pgsql-announce has 31388 subscribers to it ... when David sends out the weekly news, how many of those 31388 are "over quota", or "out of office", or "user unknown", or "connection timed out", or ... :) All those bounces come back to the main server, and, in alot of cases, get processed by majordomo's "bounce detection" algorithms ... And then there are the 177 314 "Recipient address rejected" email attempts *into* the server so far today, for ppl like "EarlineloamyDarby@postgresql.org" that come flooding in ... It all adds up ... - ---- Marc G. Fournier Hub.Org Networking Services (http://www.hub.org) Email . scrappy@hub.org MSN . scrappy@hub.org Yahoo . yscrappy Skype: hub.org ICQ . 7615664 -----BEGIN PGP SIGNATURE----- Version: GnuPG v2.0.4 (FreeBSD) iD8DBQFG0vfr4QvfyHIvDvMRAro1AJ9J1Hoc9E3VoCwxhwBh2dXnHySTlQCgyXc5 E8YhAlXZMRMCrLlOfizRP4Y= =4BMM -----END PGP SIGNATURE-----
Marc G. Fournier wrote: > > >> Any idea what's using up all those connections? 91 *incoming* connections >> is very much, especially since the box doesn't exactly *receive* a lot of >> mails (it sends a whole lot, but that's a different limit). > > Of course it receives alot ... as an example ... > > pgsql-announce has 31388 subscribers to it ... when David sends out the weekly > news, how many of those 31388 are "over quota", or "out of office", or "user > unknown", or "connection timed out", or ... :) > > All those bounces come back to the main server, and, in alot of cases, get > processed by majordomo's "bounce detection" algorithms ... > > And then there are the 177 314 "Recipient address rejected" email attempts > *into* the server so far today, for ppl like "EarlineloamyDarby@postgresql.org" > that come flooding in ... > > It all adds up ... D'oh. I only considered protocol errors, not that kind of bounces. That explains it. //Magnus