#OSGEO IRC Log - 2009-11-24

For logs after Feb 3, 2007, all times are GMT-8. Prior logs are GMT-9.
Back to Logs
00:41:38 : * mauricio_ is away: Gone away for now
00:58:48 : * mauricio_ is back.
02:17:38 : * mauricio_ is away: Gone away for now
02:46:38 : * mauricio_ is back.
04:37:28 : * mauricio_ is away: Gone away for now
04:40:38 : * mauricio_ is back.
07:22:23 marisn: Hello. Is it just for me trac.osgeo.org sloooow?
07:24:16 wildintellect: theres a good chance - the server team has been troubleshooting for a few days now
07:24:22 FrankW: we are experiencing system problems.
07:24:36 FrankW: I'm trying to take a bit of time to investigate before restarting apache.
07:27:43 marisn: a memleak in Trac Python Apache stack?
07:30:01 FrankW: marisn: that is unclear. It does not seem particularly like a memory leak.
07:30:18 FrankW: I personally suspect we have a problem with our trac/postgres integration.
07:30:41 FrankW: We have 205 postmaster processes running - presumably interfaces between 30+ trac databases and many httpd processes.
07:30:53 : * marisn got strange OOM issue on one of Gentoo servers some weeks a go. No idea why.
07:31:04 marisn: FrankW: what You use for Trac backend?
07:31:18 marisn: I mean - Postgres? MySQL?
07:31:23 FrankW: postgres
07:35:59 : * marisn has no idea how to get usage stats from Postgres. pgstatsinfo?
07:44:54 FrankW: marisn: osgeo.org services should be good again.
07:45:58 marisn: still simply rebooting is not a longterm solution...
07:46:23 marisn: yes. it flies.
07:47:19 FrankW: No, this is the second time in a couple hours I've had to restart apache.
07:47:32 FrankW: We are working on a migration strategy to new server( s ).
07:47:49 FrankW: But I really think if we understand the issue with the current server the hardware would be sufficient.
07:55:56 marisn: couple hours?!? this sounds really bad...
07:56:26 marisn: try to rise trac log verbosity?
07:59:58 marisn: also are trac logs clean? http://trac.edgewall.org/ticket/4465
08:00:00 sigq: Title: #4465 ( Timeout reached while waiting for a connection ) – The Trac Project ( at trac.edgewall.org )
08:00:30 FrankW: I'll check that.
08:03:21 marisn: FrankW: I'm sorry for putting my nose in not mine business
08:04:30 FrankW: hints are appreciated, though we have several folks digging into this for a while.
08:05:32 : * marisn would start with software upgrade, check apache process count == db accepted connection count, and top : )
08:15:26 TylerM: boy, pidgin seems so unstable :(
08:15:39 TylerM: or is it just when running with skype?
08:15:50 crschmidt: The latter.
08:16:47 TylerM: k thx.. will try again.. I was sure it couldn't suck so bad generally
08:25:20 pramsey: FrankW: could put a pgpool in between the httpd and postgres backends
08:25:41 FrankW: pramsey: care to explain further?
08:25:44 pramsey: also, is the #of backends == #of apache frontends? or >?
08:26:06 FrankW: I believe the number of backends ends up being #httpd * #tracinstances.
08:26:11 FrankW: yes, bad arrangement!
08:26:30 pramsey: so trac makes it's own connections...
08:26:36 pramsey: are the trac connections persistent?
08:27:20 FrankW: I believe they are persistent - presumably retained by the trac mod_python module.
08:27:36 marisn: #tracinstances should be controlled by apache if frefork is used
08:27:49 marisn: s/frefork/prefork
08:27:52 pramsey: rather then getting a new backend for each connection, pgpool will operate as a pool, so you'd only have one connection per database
08:27:58 pramsey: #tracinstances, presumably
08:28:21 FrankW: that would be quite reasonable.
08:28:31 FrankW: Is http://pgpool.projects.postgresql.org/ the right place to read about pgpool?
08:28:32 pramsey: so all the tracs would connect to pgpool, and pgpool could connect to pgsql
08:28:33 sigq: Title: pgpool main page ( at pgpool.projects.postgresql.org )
08:28:44 pramsey: yes
08:29:34 FrankW: This wasn't really how I had planned to spend this day. :- )
08:30:32 pramsey: then don't, the world won't end if trac service is slow : )
08:31:04 FrankW: system functioning is significantly hampered.
08:31:14 FrankW: And it makes www.osgeo.org, svn and trac nearly useless.
08:34:49 marisn: pramsey: IMHO Trac by default uses permanent connections to Postgress.
08:35:26 marisn: also being unable to report/search bugs is bad for hosted projects ; )
08:36:30 pramsey: well, that explains 200+ connections, and possibly the research exhaustion too
08:36:45 FrankW: research exhaustion?
08:36:50 pramsey: resource exhaustion
08:36:56 FrankW: ah,yes.
08:37:00 pramsey: I'm losing my ability to form correct words as I get older.
08:37:20 pramsey: Increasingly my emails read like they are written by ESL students.
08:37:25 FrankW: I am also suffering from some research exhaustion on this problem - I wasn't sure if you were making some sort of related joke.
08:37:55 pramsey: 200+ * 8Mb in work memory == lots of memory used up
08:37:58 FrankW: I am also often not a credit to my "english as a first language" status.
08:38:46 TylerM: frankw - you think it's trac bogging down other http stuff then?
08:39:03 TylerM: we could look at moving drupal over to osgeo2
08:39:11 : * TylerM is pretty getting ahead of himself again
08:39:17 FrankW: I suspect the outragous number of postmaster processes is causing problems.
08:39:17 pramsey: slow down, do the pgpool thing
08:39:48 TylerM: if it means the site becomes useable, moving it would be worth it ; )
08:39:53 marisn: pramsey: will it help if for some reason old connections remain open ( are not properly closed )?
08:39:53 TylerM: sure enough.. carry on ;- )
08:40:22 pramsey: if by help you mean hurt, yes. that's more resources used but not helping
08:40:52 FrankW: I must confess I'm not getting an immediate sense of comfort from the pgpool web pages.
08:41:26 TylerM: peer1 at 8am eastern: "We were able to login into your server and fix the backup's issue that was affecting it and backups are again running successfully."
08:41:33 marisn: FrankW: how about httpd processes? also lot's of them?
08:41:45 pramsey: http://pgfoundry.org/projects/pgpool/
08:41:46 sigq: Title: PgFoundry: pgpool: Project Info ( at pgfoundry.org )
08:41:56 FrankW: 17 httpd processes active at this moment.
08:42:11 FrankW: TylerM: any progress on our ability to file a ticket with peer1?
08:42:13 marisn: active or total count?
08:42:24 FrankW: [root@osgeo1 ~]# ps -ael | grep httpd |wc
08:42:24 FrankW: 17 238 1224
08:42:40 FrankW: you can also inspect http://trac.osgeo.org/server-status if you want a view of what's up.
08:42:41 sigq: Title: Apache Status ( at trac.osgeo.org )
08:42:45 FrankW: Note that the server is well behaved right now.
08:43:04 FrankW: wow, only 48 postmaster processes.
08:43:35 FrankW: I'm starting to doubt my theory that connections are never closed off.
08:43:56 : * FrankW sighs.
08:45:28 marisn: any chances on upgrading trac?
08:45:41 FrankW: that would be very tramatic.
08:45:54 FrankW: we run an old old version with some local customization for ldap integration.
08:46:09 : * marisn have read about that
08:48:04 FrankW: TylerM: any answer on the peer1 issue?
08:50:43 TylerM: FrankW: discussing it with them right now
08:50:49 FrankW: cool, thanks
08:51:01 TylerM: i was wondering if jason knew how to get in
08:52:17 TylerM: the password email reminder tool didn't work for me - so they did it manually for me
08:52:22 TylerM: back in a minute
08:55:19 : * TylerM waits for the email...
08:55:28 TylerM: i wonder how well postmaster is performing
08:55:43 TylerM: ~yawn~
10:58:22 TylerM: http://www.agu.org/meetings/fm09/ down for you?
10:59:01 marisn: works ( 1 kb/s )
10:59:05 bitnerd: TylerM, seems up but deadly slow
10:59:13 sigq: Title: 2009 AGU Fall Meeting ( at www.agu.org )
10:59:47 TylerM: thx bitnerd
11:00:12 bitnerd: sigq would seem to agree with two minutes to get the title
14:24:53 bitner: cheers