Jean-Baptiste Quenot
2009-Apr-21 07:19 UTC
[Puppet Users] Should we really expect puppetd to die randomly? (Was: Puppet watching puppet)
2009/4/6 Mike Renfro <renfro@tntech.edu>:> I''d normally expect that to work, but I just have puppet keep cron > running, and have a periodic cron job that checks if puppet has died, > and if so, restarts it:Interesting, but why would you expect Puppet to die? Would you expect Apache, Nginx or worse MySQL to die randomly like this? I''m a Puppet user since nearly two years now, and in the big picture of my web servers I find that puppetd is not the most reliable piece of software. It dies every day, and my colleagues complain about this regularly because their installed packages are not uptodate as they expect. So I have to start it again and again on all machines. Is Puppetd dying because of network problems? I believe so, but I think it should be fixed instead of finding creative ways to keep puppetd running, especially since I request it to run every 5 minutes as a daemon. Is anyone using puppetd in a WAN setup with default Webrick server successfully? Shall I switch the HTTP server to Mongrel to gain reliability? I''ll test this setup. But if puppetd fails on the client side, I''m not certain that changing the server''s HTTP server would actually prevent the client to fail at all... I''m a bit eager with this, and I''m really looking forward to find a solution, community-wise. Your comments are welcome. -- Jean-Baptiste Quenot http://jbq.caraldi.com/ --~--~---------~--~----~------------~-------~--~----~ You received this message because you are subscribed to the Google Groups "Puppet Users" group. To post to this group, send email to puppet-users@googlegroups.com To unsubscribe from this group, send email to puppet-users+unsubscribe@googlegroups.com For more options, visit this group at http://groups.google.com/group/puppet-users?hl=en -~----------~----~----~----~------~----~------~--~---
Ohad Levy
2009-Apr-21 08:03 UTC
[Puppet Users] Re: Should we really expect puppetd to die randomly? (Was: Puppet watching puppet)
We are using puppetd though cron, and it seems to me much more reliable and much less resource hungry.. if you have more than a few clients, switch to mongrel/passenger. WAN is not a real issue if its stable, of course it takes longer runs but... cheers, Ohad On Tue, Apr 21, 2009 at 3:19 PM, Jean-Baptiste Quenot <caraldi@gmail.com>wrote:> > 2009/4/6 Mike Renfro <renfro@tntech.edu>: > > > I''d normally expect that to work, but I just have puppet keep cron > > running, and have a periodic cron job that checks if puppet has died, > > and if so, restarts it: > > Interesting, but why would you expect Puppet to die? Would you expect > Apache, Nginx or worse MySQL to die randomly like this? > > I''m a Puppet user since nearly two years now, and in the big picture > of my web servers I find that puppetd is not the most reliable piece > of software. It dies every day, and my colleagues complain about this > regularly because their installed packages are not uptodate as they > expect. So I have to start it again and again on all machines. Is > Puppetd dying because of network problems? I believe so, but I think > it should be fixed instead of finding creative ways to keep puppetd > running, especially since I request it to run every 5 minutes as a > daemon. > > Is anyone using puppetd in a WAN setup with default Webrick server > successfully? Shall I switch the HTTP server to Mongrel to gain > reliability? I''ll test this setup. But if puppetd fails on the > client side, I''m not certain that changing the server''s HTTP server > would actually prevent the client to fail at all... > > I''m a bit eager with this, and I''m really looking forward to find a > solution, community-wise. Your comments are welcome. > -- > Jean-Baptiste Quenot > http://jbq.caraldi.com/ > > > >--~--~---------~--~----~------------~-------~--~----~ You received this message because you are subscribed to the Google Groups "Puppet Users" group. To post to this group, send email to puppet-users@googlegroups.com To unsubscribe from this group, send email to puppet-users+unsubscribe@googlegroups.com For more options, visit this group at http://groups.google.com/group/puppet-users?hl=en -~----------~----~----~----~------~----~------~--~---
Trevor Vaughan
2009-Apr-21 11:30 UTC
[Puppet Users] Re: Should we really expect puppetd to die randomly? (Was: Puppet watching puppet)
I haven''t had any problems with the client stability. But, I''m also not using the ''file'' type to copy files over the ''puppet://'' protocol. It might be related, but it might not. Are you hitting the OOM killer? It sounds like you might be. I *would* expect Apache and MySQL to die like this (and I''ve seen it) if they''re hitting resource limits and the OS decides that they need to die. The last time I saw it with Apache was due to a badly written plugin that ate all of the memory in the system which caused Apache to be nuked. Trevor On Tue, Apr 21, 2009 at 03:19, Jean-Baptiste Quenot <caraldi@gmail.com> wrote:> > 2009/4/6 Mike Renfro <renfro@tntech.edu>: > >> I''d normally expect that to work, but I just have puppet keep cron >> running, and have a periodic cron job that checks if puppet has died, >> and if so, restarts it: > > Interesting, but why would you expect Puppet to die? Would you expect > Apache, Nginx or worse MySQL to die randomly like this? > > I''m a Puppet user since nearly two years now, and in the big picture > of my web servers I find that puppetd is not the most reliable piece > of software. It dies every day, and my colleagues complain about this > regularly because their installed packages are not uptodate as they > expect. So I have to start it again and again on all machines. Is > Puppetd dying because of network problems? I believe so, but I think > it should be fixed instead of finding creative ways to keep puppetd > running, especially since I request it to run every 5 minutes as a > daemon. > > Is anyone using puppetd in a WAN setup with default Webrick server > successfully? Shall I switch the HTTP server to Mongrel to gain > reliability? I''ll test this setup. But if puppetd fails on the > client side, I''m not certain that changing the server''s HTTP server > would actually prevent the client to fail at all... > > I''m a bit eager with this, and I''m really looking forward to find a > solution, community-wise. Your comments are welcome. > -- > Jean-Baptiste Quenot > http://jbq.caraldi.com/ > > > >--~--~---------~--~----~------------~-------~--~----~ You received this message because you are subscribed to the Google Groups "Puppet Users" group. To post to this group, send email to puppet-users@googlegroups.com To unsubscribe from this group, send email to puppet-users+unsubscribe@googlegroups.com For more options, visit this group at http://groups.google.com/group/puppet-users?hl=en -~----------~----~----~----~------~----~------~--~---
Bjørn Dyre Dyresen
2009-Apr-21 11:34 UTC
[Puppet Users] Re: Should we really expect puppetd to die randomly? (Was: Puppet watching puppet)
We had this problem when we hit the scaling wall running webrick. We then moved our puppetmaster to a beefy server (dual quadcore with 16 GB ram). Here we run six puppetmasterds with mongrel and stored config. On the same server we run nginx. With this setup we can loop over a lot of servers restarting puppet client without any trouble at all. I tried doing more than 100 at the time without any trouble. So it certainly performs. After setting this up the problem with dead puppet clients seemed to vanish. I have no idea how many servers you are running, but when pulling every fifth minute they sure should get some traffic. You will hit the scaling wall earlier. If you have 50 servers you need a setup that other people would use for a two/three hundred server setup. Are you really sure you need to pull that often? 2009/4/21 Ohad Levy <ohadlevy@gmail.com>> We are using puppetd though cron, and it seems to me much more reliable and > much less resource hungry.. > > if you have more than a few clients, switch to mongrel/passenger. > WAN is not a real issue if its stable, of course it takes longer runs > but... > > cheers, > Ohad > > > On Tue, Apr 21, 2009 at 3:19 PM, Jean-Baptiste Quenot <caraldi@gmail.com>wrote: > >> >> 2009/4/6 Mike Renfro <renfro@tntech.edu>: >> >> > I''d normally expect that to work, but I just have puppet keep cron >> > running, and have a periodic cron job that checks if puppet has died, >> > and if so, restarts it: >> >> Interesting, but why would you expect Puppet to die? Would you expect >> Apache, Nginx or worse MySQL to die randomly like this? >> >> I''m a Puppet user since nearly two years now, and in the big picture >> of my web servers I find that puppetd is not the most reliable piece >> of software. It dies every day, and my colleagues complain about this >> regularly because their installed packages are not uptodate as they >> expect. So I have to start it again and again on all machines. Is >> Puppetd dying because of network problems? I believe so, but I think >> it should be fixed instead of finding creative ways to keep puppetd >> running, especially since I request it to run every 5 minutes as a >> daemon. >> >> Is anyone using puppetd in a WAN setup with default Webrick server >> successfully? Shall I switch the HTTP server to Mongrel to gain >> reliability? I''ll test this setup. But if puppetd fails on the >> client side, I''m not certain that changing the server''s HTTP server >> would actually prevent the client to fail at all... >> >> I''m a bit eager with this, and I''m really looking forward to find a >> solution, community-wise. Your comments are welcome. >> -- >> Jean-Baptiste Quenot >> http://jbq.caraldi.com/ >> >> >> > > > >--~--~---------~--~----~------------~-------~--~----~ You received this message because you are subscribed to the Google Groups "Puppet Users" group. To post to this group, send email to puppet-users@googlegroups.com To unsubscribe from this group, send email to puppet-users+unsubscribe@googlegroups.com For more options, visit this group at http://groups.google.com/group/puppet-users?hl=en -~----------~----~----~----~------~----~------~--~---
Stasheck
2009-Apr-21 11:35 UTC
[Puppet Users] Re: Should we really expect puppetd to die randomly? (Was: Puppet watching puppet)
Ohad, I remember you were saiyng something about running puppet from inetd - can you share some info? I really think that cron+inetd combination would suit my company better than puppetd (which is restarted every hour via cron) /br Stanislaw On Apr 21, 10:03 am, Ohad Levy <ohadl...@gmail.com> wrote:> We are using puppetd though cron, and it seems to me much more reliable and > much less resource hungry.. > > if you have more than a few clients, switch to mongrel/passenger. > WAN is not a real issue if its stable, of course it takes longer runs but... > > cheers, > Ohad > > On Tue, Apr 21, 2009 at 3:19 PM, Jean-Baptiste Quenot <cara...@gmail.com>wrote: > > > > > 2009/4/6 Mike Renfro <ren...@tntech.edu>: > > > > I''d normally expect that to work, but I just have puppet keep cron > > > running, and have a periodic cron job that checks if puppet has died, > > > and if so, restarts it: > > > Interesting, but why would you expect Puppet to die? Would you expect > > Apache, Nginx or worse MySQL to die randomly like this? > > > I''m a Puppet user since nearly two years now, and in the big picture > > of my web servers I find that puppetd is not the most reliable piece > > of software. It dies every day, and my colleagues complain about this > > regularly because their installed packages are not uptodate as they > > expect. So I have to start it again and again on all machines. Is > > Puppetd dying because of network problems? I believe so, but I think > > it should be fixed instead of finding creative ways to keep puppetd > > running, especially since I request it to run every 5 minutes as a > > daemon. > > > Is anyone using puppetd in a WAN setup with default Webrick server > > successfully? Shall I switch the HTTP server to Mongrel to gain > > reliability? I''ll test this setup. But if puppetd fails on the > > client side, I''m not certain that changing the server''s HTTP server > > would actually prevent the client to fail at all... > > > I''m a bit eager with this, and I''m really looking forward to find a > > solution, community-wise. Your comments are welcome. > > -- > > Jean-Baptiste Quenot > >http://jbq.caraldi.com/--~--~---------~--~----~------------~-------~--~----~ You received this message because you are subscribed to the Google Groups "Puppet Users" group. To post to this group, send email to puppet-users@googlegroups.com To unsubscribe from this group, send email to puppet-users+unsubscribe@googlegroups.com For more options, visit this group at http://groups.google.com/group/puppet-users?hl=en -~----------~----~----~----~------~----~------~--~---
Mike Renfro
2009-Apr-21 13:56 UTC
[Puppet Users] Re: Should we really expect puppetd to die randomly? (Was: Puppet watching puppet)
Jean-Baptiste Quenot wrote:> 2009/4/6 Mike Renfro <renfro@tntech.edu>: > >> I''d normally expect that to work, but I just have puppet keep cron >> running, and have a periodic cron job that checks if puppet has died, >> and if so, restarts it: > > Interesting, but why would you expect Puppet to die? Would you expect > Apache, Nginx or worse MySQL to die randomly like this?At the time I put together that recipe, I didn''t particularly expect puppet to die randomly, but I knew there was a similar practice for cfengine users. And though I wouldn''t *expect* any of those other services to die randomly, I''d certainly monitor all of them with Nagios and write puppet manifests to restart them if they''re found to be not running. I have had some puppetd''s die and get restarted from cron. On the few dozen systems I have with puppetd, I don''t think it happens even once per day for the entire group. -- Mike Renfro / R&D Engineer, Center for Manufacturing Research, 931 372-3601 / Tennessee Technological University -- renfro@tntech.edu --~--~---------~--~----~------------~-------~--~----~ You received this message because you are subscribed to the Google Groups "Puppet Users" group. To post to this group, send email to puppet-users@googlegroups.com To unsubscribe from this group, send email to puppet-users+unsubscribe@googlegroups.com For more options, visit this group at http://groups.google.com/group/puppet-users?hl=en -~----------~----~----~----~------~----~------~--~---
Nigel Kersten
2009-Apr-21 15:17 UTC
[Puppet Users] Re: Should we really expect puppetd to die randomly? (Was: Puppet watching puppet)
On Tue, Apr 21, 2009 at 6:56 AM, Mike Renfro <renfro@tntech.edu> wrote:> > Jean-Baptiste Quenot wrote: >> 2009/4/6 Mike Renfro <renfro@tntech.edu>: >> >>> I''d normally expect that to work, but I just have puppet keep cron >>> running, and have a periodic cron job that checks if puppet has died, >>> and if so, restarts it: >> >> Interesting, but why would you expect Puppet to die? Would you expect >> Apache, Nginx or worse MySQL to die randomly like this?Why not? Isn''t that why we have sitters and puppet ensuring services are running? :) my 2c is that we had persistent performance and reliability problems with puppetd as a daemon, and switched to cron/launchd jobs that simply run puppetd --onetime --no-daemonize instead.> > At the time I put together that recipe, I didn''t particularly expect > puppet to die randomly, but I knew there was a similar practice for > cfengine users. And though I wouldn''t *expect* any of those other > services to die randomly, I''d certainly monitor all of them with Nagios > and write puppet manifests to restart them if they''re found to be not > running. > > I have had some puppetd''s die and get restarted from cron. On the few > dozen systems I have with puppetd, I don''t think it happens even once > per day for the entire group. > > -- > Mike Renfro / R&D Engineer, Center for Manufacturing Research, > 931 372-3601 / Tennessee Technological University -- renfro@tntech.edu > > > >-- Nigel Kersten nigelk@google.com System Administrator Google, Inc. --~--~---------~--~----~------------~-------~--~----~ You received this message because you are subscribed to the Google Groups "Puppet Users" group. To post to this group, send email to puppet-users@googlegroups.com To unsubscribe from this group, send email to puppet-users+unsubscribe@googlegroups.com For more options, visit this group at http://groups.google.com/group/puppet-users?hl=en -~----------~----~----~----~------~----~------~--~---
Jean-Baptiste Quenot
2009-Apr-21 15:30 UTC
[Puppet Users] Re: Should we really expect puppetd to die randomly? (Was: Puppet watching puppet)
2009/4/21 Trevor Vaughan <peiriannydd@gmail.com>:> > I haven''t had any problems with the client stability. > > But, I''m also not using the ''file'' type to copy files over the > ''puppet://'' protocol.I''m using the puppet:// protocol extensively yes. As I noticed it was very slow especially with recurse=>true, I tend to put my files in Debian packages when possible.> Are you hitting the OOM killer? It sounds like you might be.I don''t think so because there is nothing in syslog. The last message from puppetd is "Starting catalog run". And then it dies. -- Jean-Baptiste Quenot http://jbq.caraldi.com/ --~--~---------~--~----~------------~-------~--~----~ You received this message because you are subscribed to the Google Groups "Puppet Users" group. To post to this group, send email to puppet-users@googlegroups.com To unsubscribe from this group, send email to puppet-users+unsubscribe@googlegroups.com For more options, visit this group at http://groups.google.com/group/puppet-users?hl=en -~----------~----~----~----~------~----~------~--~---
Jean-Baptiste Quenot
2009-Apr-21 15:34 UTC
[Puppet Users] Re: Should we really expect puppetd to die randomly? (Was: Puppet watching puppet)
2009/4/21 Bjørn Dyre Dyresen <bjorn@dyresen.net>:> We had this problem when we hit the scaling wall running webrick. We then > moved our puppetmaster to a beefy server (dual quadcore with 16 GB ram). > Here we run six puppetmasterds with mongrel and stored config. On the same > server we run nginx. With this setup we can loop over a lot of servers > restarting puppet client without any trouble at all. I tried doing more than > 100 at the time without any trouble. So it certainly performs. > > After setting this up the problem with dead puppet clients seemed to vanish. > I have no idea how many servers you are running, but when pulling every > fifth minute they sure should get some traffic. You will hit the scaling > wall earlier. If you have 50 servers you need a setup that other people > would use for a two/three hundred server setup. Are you really sure you need > to pull that often?If scaling was involved I would expect Puppetd to be slow, not to die. I have about 12 servers that hit a single puppetmasterd through the Internet, with a lot of puppet:// resources, and every 5 minutes. Anyway, I''ll try that setup if it worked for you, especially as you mention that "dead puppet clients" disappeared. Thanks! -- Jean-Baptiste Quenot http://jbq.caraldi.com/ --~--~---------~--~----~------------~-------~--~----~ You received this message because you are subscribed to the Google Groups "Puppet Users" group. To post to this group, send email to puppet-users@googlegroups.com To unsubscribe from this group, send email to puppet-users+unsubscribe@googlegroups.com For more options, visit this group at http://groups.google.com/group/puppet-users?hl=en -~----------~----~----~----~------~----~------~--~---
Robin Lee Powell
2009-Apr-21 17:48 UTC
[Puppet Users] Re: Should we really expect puppetd to die randomly? (Was: Puppet watching puppet)
On Tue, Apr 21, 2009 at 05:30:54PM +0200, Jean-Baptiste Quenot wrote:> > > Are you hitting the OOM killer? It sounds like you might be. > > I don''t think so because there is nothing in syslog. The last > message from puppetd is "Starting catalog run". And then it dies.The OOM killer isn''t a message from puppetd, and probably isn''t even in syslog; it''s a kernel message. On my box I''d expect to see it in /var/log/kern.log -Robin -- They say: "The first AIs will be built by the military as weapons." And I''m thinking: "Does it even occur to you to try for something other than the default outcome?" See http://shrunklink.com/cdiz http://www.digitalkingdom.org/~rlpowell/ *** http://www.lojban.org/ --~--~---------~--~----~------------~-------~--~----~ You received this message because you are subscribed to the Google Groups "Puppet Users" group. To post to this group, send email to puppet-users@googlegroups.com To unsubscribe from this group, send email to puppet-users+unsubscribe@googlegroups.com For more options, visit this group at http://groups.google.com/group/puppet-users?hl=en -~----------~----~----~----~------~----~------~--~---
Luke Kanies
2009-Apr-21 19:04 UTC
[Puppet Users] Re: Should we really expect puppetd to die randomly? (Was: Puppet watching puppet)
On Apr 21, 2009, at 2:19 AM, Jean-Baptiste Quenot wrote:> > 2009/4/6 Mike Renfro <renfro@tntech.edu>: > >> I''d normally expect that to work, but I just have puppet keep cron >> running, and have a periodic cron job that checks if puppet has died, >> and if so, restarts it: > > Interesting, but why would you expect Puppet to die? Would you expect > Apache, Nginx or worse MySQL to die randomly like this? > > I''m a Puppet user since nearly two years now, and in the big picture > of my web servers I find that puppetd is not the most reliable piece > of software. It dies every day, and my colleagues complain about this > regularly because their installed packages are not uptodate as they > expect. So I have to start it again and again on all machines. Is > Puppetd dying because of network problems? I believe so, but I think > it should be fixed instead of finding creative ways to keep puppetd > running, especially since I request it to run every 5 minutes as a > daemon.I agree, these problems shouldn''t exist, and when they do they should be fixed. The problem is that we can''t seem to find anyone who has the time or interest in debugging the problem, and we not-too-surprisingly can''t reproduce it ourselves. Everyone who has the problem just shrugs and switches to cron, and pretty much stops responding to our emails asking for help fixing the problem. What we really need is someone who''s having these problems and is willing to spend some time investigating. I expect it''s not that complicated, it just needs to be hunted down. And, for the record, I think this got a lot better in 0.24.8 (or maybe 0.24.7?). We found a few cases where errors could propagate far enough to kill the client, and I think we''ve eradicated nearly all of them.> Is anyone using puppetd in a WAN setup with default Webrick server > successfully? Shall I switch the HTTP server to Mongrel to gain > reliability? I''ll test this setup. But if puppetd fails on the > client side, I''m not certain that changing the server''s HTTP server > would actually prevent the client to fail at all...We need to add logs or something to clarify this, but you should *definitely* switch away from Webrick if you''re having any kind of performance or reliability problems. Or if you''re not. -- Death and taxes are inevitable; at least death doesn''t get worse every year. --------------------------------------------------------------------- Luke Kanies | http://reductivelabs.com | http://madstop.com --~--~---------~--~----~------------~-------~--~----~ You received this message because you are subscribed to the Google Groups "Puppet Users" group. To post to this group, send email to puppet-users@googlegroups.com To unsubscribe from this group, send email to puppet-users+unsubscribe@googlegroups.com For more options, visit this group at http://groups.google.com/group/puppet-users?hl=en -~----------~----~----~----~------~----~------~--~---
Ohad Levy
2009-Apr-22 03:41 UTC
[Puppet Users] Re: Should we really expect puppetd to die randomly? (Was: Puppet watching puppet)
Hi, You can find a "working" daemon here: http://github.com/ohadlevy/puppet/tree/puppetlisten under the ext/puppetlisten subdir. It doesn''t use inetd as ruby socket implementation does not work inet, I''m considering rewriting it in c but no real reason / time for now... search the mailing list for previous usage instruction (but its kinda straight forward..) Ohad On Tue, Apr 21, 2009 at 7:35 PM, Stasheck <stanislaw.kaminski@gmail.com>wrote:> > Ohad, I remember you were saiyng something about running puppet from > inetd - can you share some info? I really think that cron+inetd > combination would suit my company better than puppetd (which is > restarted every hour via cron) > > /br > Stanislaw > > On Apr 21, 10:03 am, Ohad Levy <ohadl...@gmail.com> wrote: > > We are using puppetd though cron, and it seems to me much more reliable > and > > much less resource hungry.. > > > > if you have more than a few clients, switch to mongrel/passenger. > > WAN is not a real issue if its stable, of course it takes longer runs > but... > > > > cheers, > > Ohad > > > > On Tue, Apr 21, 2009 at 3:19 PM, Jean-Baptiste Quenot <cara...@gmail.com > >wrote: > > > > > > > > > 2009/4/6 Mike Renfro <ren...@tntech.edu>: > > > > > > I''d normally expect that to work, but I just have puppet keep cron > > > > running, and have a periodic cron job that checks if puppet has died, > > > > and if so, restarts it: > > > > > Interesting, but why would you expect Puppet to die? Would you expect > > > Apache, Nginx or worse MySQL to die randomly like this? > > > > > I''m a Puppet user since nearly two years now, and in the big picture > > > of my web servers I find that puppetd is not the most reliable piece > > > of software. It dies every day, and my colleagues complain about this > > > regularly because their installed packages are not uptodate as they > > > expect. So I have to start it again and again on all machines. Is > > > Puppetd dying because of network problems? I believe so, but I think > > > it should be fixed instead of finding creative ways to keep puppetd > > > running, especially since I request it to run every 5 minutes as a > > > daemon. > > > > > Is anyone using puppetd in a WAN setup with default Webrick server > > > successfully? Shall I switch the HTTP server to Mongrel to gain > > > reliability? I''ll test this setup. But if puppetd fails on the > > > client side, I''m not certain that changing the server''s HTTP server > > > would actually prevent the client to fail at all... > > > > > I''m a bit eager with this, and I''m really looking forward to find a > > > solution, community-wise. Your comments are welcome. > > > -- > > > Jean-Baptiste Quenot > > >http://jbq.caraldi.com/ > > >--~--~---------~--~----~------------~-------~--~----~ You received this message because you are subscribed to the Google Groups "Puppet Users" group. To post to this group, send email to puppet-users@googlegroups.com To unsubscribe from this group, send email to puppet-users+unsubscribe@googlegroups.com For more options, visit this group at http://groups.google.com/group/puppet-users?hl=en -~----------~----~----~----~------~----~------~--~---
Joel Krauska
2009-Apr-22 07:23 UTC
[Puppet Users] Re: Should we really expect puppetd to die randomly? (Was: Puppet watching puppet)
Luke, I was the guy who brought up puppet client fails last June at the SF meetup. The memory footprint of the puppetd client in idle mode (between triggers) was enough of a reason to dump the always-on client, but I also found that they crashed regularly whenever there was any difficulty contacting the puppet master. (too many concurrent, etc) I just run my clients in cron like this: /usr/bin/perl -e "sleep int(rand(60*30));" ; /usr/sbin/puppetd --test 1> /tmp/puppetd.log I would be more than happy to help debug why the client crashes. What do you suggest I run it with? Is there a ruby debugger? Unfortunately I''ll probably not want to upgrade to the newest puppet for this test harness, but I''d be happy to put 0.24.4 up for some crash and burns. --joel On Tue, Apr 21, 2009 at 12:04 PM, Luke Kanies <luke@madstop.com> wrote:> > On Apr 21, 2009, at 2:19 AM, Jean-Baptiste Quenot wrote: > >> >> 2009/4/6 Mike Renfro <renfro@tntech.edu>: >> >>> I''d normally expect that to work, but I just have puppet keep cron >>> running, and have a periodic cron job that checks if puppet has died, >>> and if so, restarts it: >> >> Interesting, but why would you expect Puppet to die? Would you expect >> Apache, Nginx or worse MySQL to die randomly like this? >> >> I''m a Puppet user since nearly two years now, and in the big picture >> of my web servers I find that puppetd is not the most reliable piece >> of software. It dies every day, and my colleagues complain about this >> regularly because their installed packages are not uptodate as they >> expect. So I have to start it again and again on all machines. Is >> Puppetd dying because of network problems? I believe so, but I think >> it should be fixed instead of finding creative ways to keep puppetd >> running, especially since I request it to run every 5 minutes as a >> daemon. > > I agree, these problems shouldn''t exist, and when they do they should > be fixed. > > The problem is that we can''t seem to find anyone who has the time or > interest in debugging the problem, and we not-too-surprisingly can''t > reproduce it ourselves. Everyone who has the problem just shrugs and > switches to cron, and pretty much stops responding to our emails > asking for help fixing the problem. > > What we really need is someone who''s having these problems and is > willing to spend some time investigating. I expect it''s not that > complicated, it just needs to be hunted down. > > And, for the record, I think this got a lot better in 0.24.8 (or maybe > 0.24.7?). We found a few cases where errors could propagate far > enough to kill the client, and I think we''ve eradicated nearly all of > them. > >> Is anyone using puppetd in a WAN setup with default Webrick server >> successfully? Shall I switch the HTTP server to Mongrel to gain >> reliability? I''ll test this setup. But if puppetd fails on the >> client side, I''m not certain that changing the server''s HTTP server >> would actually prevent the client to fail at all... > > We need to add logs or something to clarify this, but you should > *definitely* switch away from Webrick if you''re having any kind of > performance or reliability problems. Or if you''re not. > > -- > Death and taxes are inevitable; at least death doesn''t get worse every > year. > --------------------------------------------------------------------- > Luke Kanies | http://reductivelabs.com | http://madstop.com > > > > >--~--~---------~--~----~------------~-------~--~----~ You received this message because you are subscribed to the Google Groups "Puppet Users" group. To post to this group, send email to puppet-users@googlegroups.com To unsubscribe from this group, send email to puppet-users+unsubscribe@googlegroups.com For more options, visit this group at http://groups.google.com/group/puppet-users?hl=en -~----------~----~----~----~------~----~------~--~---
Jean-Baptiste Quenot
2009-Apr-22 07:25 UTC
[Puppet Users] Re: Should we really expect puppetd to die randomly? (Was: Puppet watching puppet)
2009/4/21 Luke Kanies <luke@madstop.com>:> And, for the record, I think this got a lot better in 0.24.8 (or maybe > 0.24.7?). We found a few cases where errors could propagate far > enough to kill the client, and I think we''ve eradicated nearly all of > them.I''ll try to switch to 0.24.8, as I''m using 0.24.4 right now. Thanks for your valuable answers. If the problem persists I''ll be running the daemons in debug mode to see what''s wrong. -- Jean-Baptiste Quenot http://jbq.caraldi.com/ --~--~---------~--~----~------------~-------~--~----~ You received this message because you are subscribed to the Google Groups "Puppet Users" group. To post to this group, send email to puppet-users@googlegroups.com To unsubscribe from this group, send email to puppet-users+unsubscribe@googlegroups.com For more options, visit this group at http://groups.google.com/group/puppet-users?hl=en -~----------~----~----~----~------~----~------~--~---
Jean-Baptiste Quenot
2009-Apr-23 13:54 UTC
[Puppet Users] Re: Should we really expect puppetd to die randomly? (Was: Puppet watching puppet)
2009/4/22 Jean-Baptiste Quenot <caraldi@gmail.com>:> 2009/4/21 Luke Kanies <luke@madstop.com>: > >> And, for the record, I think this got a lot better in 0.24.8 (or maybe >> 0.24.7?). We found a few cases where errors could propagate far >> enough to kill the client, and I think we''ve eradicated nearly all of >> them. > > I''ll try to switch to 0.24.8, as I''m using 0.24.4 right now. Thanks > for your valuable answers. If the problem persists I''ll be running > the daemons in debug mode to see what''s wrong.Seems good so far, none of my Puppet clients died since I upgraded to 0.24.8, still with default Webrick. Good job! I''m happy :-) -- Jean-Baptiste Quenot http://jbq.caraldi.com/ --~--~---------~--~----~------------~-------~--~----~ You received this message because you are subscribed to the Google Groups "Puppet Users" group. To post to this group, send email to puppet-users@googlegroups.com To unsubscribe from this group, send email to puppet-users+unsubscribe@googlegroups.com For more options, visit this group at http://groups.google.com/group/puppet-users?hl=en -~----------~----~----~----~------~----~------~--~---
threetee
2009-Apr-24 13:25 UTC
[Puppet Users] Re: Should we really expect puppetd to die randomly? (Was: Puppet watching puppet)
On Apr 21, 12:19 am, Jean-Baptiste Quenot <cara...@gmail.com> wrote:> 2009/4/6 Mike Renfro <ren...@tntech.edu>: > > > I''d normally expect that to work, but I just have puppet keep cron > > running, and have a periodic cron job that checks if puppet has died, > > and if so, restarts it: > > Interesting, but why would you expect Puppet to die? Would you expect > Apache, Nginx or worse MySQL to die randomly like this? > > I''m a Puppet user since nearly two years now, and in the big picture > of my web servers I find that puppetd is not the most reliable piece > of software. It dies every day, and my colleagues complain about this > regularly because their installed packages are not uptodate as they > expect. So I have to start it again and again on all machines. Is > Puppetd dying because of network problems? I believe so, but I think > it should be fixed instead of finding creative ways to keep puppetd > running, especially since I request it to run every 5 minutes as a > daemon. > > Is anyone using puppetd in a WAN setup with default Webrick server > successfully? Shall I switch the HTTP server to Mongrel to gain > reliability? I''ll test this setup. But if puppetd fails on the > client side, I''m not certain that changing the server''s HTTP server > would actually prevent the client to fail at all... > > I''m a bit eager with this, and I''m really looking forward to find a > solution, community-wise. Your comments are welcome. > -- > Jean-Baptiste Quenothttp://jbq.caraldi.com/This has already been mentioned in this thread, but I''ll add my $0.02 anyway. I use puppet to manage about 500 servers, and I started having the same scalability problems everyone else is having with Webrick: puppetd started randomly dying on clients, IIRC when I reached around 50 nodes. I initially thought about working around it by putting puppet into cron, I didn''t want to sacrifice some of the features that puppetd provides (scheduling, ability to monitor puppetd status from a central location). I was also concerned that puppet might still have problems pulling files from the fileserver with a Webrick-based solution. I kicked Webrick to the curb and replaced it with Passenger/ Rack on my puppetmaster, using the instructions at http://reductivelabs.com/trac/puppet/wiki/UsingPassenger. This dropped the puppetd mortality rate on my client nodes to nearly zero, and I get to keep using puppetd on the clients. :) --~--~---------~--~----~------------~-------~--~----~ You received this message because you are subscribed to the Google Groups "Puppet Users" group. To post to this group, send email to puppet-users@googlegroups.com To unsubscribe from this group, send email to puppet-users+unsubscribe@googlegroups.com For more options, visit this group at http://groups.google.com/group/puppet-users?hl=en -~----------~----~----~----~------~----~------~--~---
Nigel Kersten
2009-Apr-24 13:45 UTC
[Puppet Users] Re: Should we really expect puppetd to die randomly? (Was: Puppet watching puppet)
On Fri, Apr 24, 2009 at 6:25 AM, threetee <threetee@gmail.com> wrote:> > On Apr 21, 12:19 am, Jean-Baptiste Quenot <cara...@gmail.com> wrote: >> 2009/4/6 Mike Renfro <ren...@tntech.edu>: >> >> > I''d normally expect that to work, but I just have puppet keep cron >> > running, and have a periodic cron job that checks if puppet has died, >> > and if so, restarts it: >> >> Interesting, but why would you expect Puppet to die? Would you expect >> Apache, Nginx or worse MySQL to die randomly like this? >> >> I''m a Puppet user since nearly two years now, and in the big picture >> of my web servers I find that puppetd is not the most reliable piece >> of software. It dies every day, and my colleagues complain about this >> regularly because their installed packages are not uptodate as they >> expect. So I have to start it again and again on all machines. Is >> Puppetd dying because of network problems? I believe so, but I think >> it should be fixed instead of finding creative ways to keep puppetd >> running, especially since I request it to run every 5 minutes as a >> daemon. >> >> Is anyone using puppetd in a WAN setup with default Webrick server >> successfully? Shall I switch the HTTP server to Mongrel to gain >> reliability? I''ll test this setup. But if puppetd fails on the >> client side, I''m not certain that changing the server''s HTTP server >> would actually prevent the client to fail at all... >> >> I''m a bit eager with this, and I''m really looking forward to find a >> solution, community-wise. Your comments are welcome. >> -- >> Jean-Baptiste Quenothttp://jbq.caraldi.com/ > > This has already been mentioned in this thread, but I''ll add my $0.02 > anyway. I use puppet to manage about 500 servers, and I started having > the same scalability problems everyone else is having with Webrick: > puppetd started randomly dying on clients, IIRC when I reached around > 50 nodes. I initially thought about working around it by putting > puppet into cron, I didn''t want to sacrifice some of the features that > puppetd provides (scheduling, ability to monitor puppetd status from a > central location). I was also concerned that puppet might still have > problems pulling files from the fileserver with a Webrick-based > solution. I kicked Webrick to the curb and replaced it with Passenger/ > Rack on my puppetmaster, using the instructions at > http://reductivelabs.com/trac/puppet/wiki/UsingPassenger. This dropped > the puppetd mortality rate on my client nodes to nearly zero, and I > get to keep using puppetd on the clients. :)I have some python code that fires off a bunch of simultaneous threads that mimic puppet clients supplying facts and requesting a compiled manifest (planning to release this btw), and I''ve been amazed at just how much better Passenger/Rack performs compared to Pound/Nginx/Mongrel. -- Nigel Kersten nigelk@google.com System Administrator Google, Inc. --~--~---------~--~----~------------~-------~--~----~ You received this message because you are subscribed to the Google Groups "Puppet Users" group. To post to this group, send email to puppet-users@googlegroups.com To unsubscribe from this group, send email to puppet-users+unsubscribe@googlegroups.com For more options, visit this group at http://groups.google.com/group/puppet-users?hl=en -~----------~----~----~----~------~----~------~--~---
Ohad Levy
2009-Apr-30 06:29 UTC
[Puppet Users] Re: Should we really expect puppetd to die randomly? (Was: Puppet watching puppet)
On Fri, Apr 24, 2009 at 9:45 PM, Nigel Kersten <nigelk@google.com> wrote:> ... > > manifest (planning to release this btw), and I''ve been amazed at just > how much better Passenger/Rack performs compared to > Pound/Nginx/Mongrel. >How much did you tweak the passenger options? (e.g. PassengerMaxRequests, PassengerPoolIdleTime etc) cheers, Ohad --~--~---------~--~----~------------~-------~--~----~ You received this message because you are subscribed to the Google Groups "Puppet Users" group. To post to this group, send email to puppet-users@googlegroups.com To unsubscribe from this group, send email to puppet-users+unsubscribe@googlegroups.com For more options, visit this group at http://groups.google.com/group/puppet-users?hl=en -~----------~----~----~----~------~----~------~--~---
Nigel Kersten
2009-Apr-30 17:14 UTC
[Puppet Users] Re: Should we really expect puppetd to die randomly? (Was: Puppet watching puppet)
On Wed, Apr 29, 2009 at 11:29 PM, Ohad Levy <ohadlevy@gmail.com> wrote:> > > On Fri, Apr 24, 2009 at 9:45 PM, Nigel Kersten <nigelk@google.com> wrote: >> >> ... >> manifest (planning to release this btw), and I''ve been amazed at just >> how much better Passenger/Rack performs compared to >> Pound/Nginx/Mongrel. > > How much did you tweak the passenger options? (e.g. PassengerMaxRequests, > PassengerPoolIdleTime etc)I''m actually looking at this at the moment. I might start a new thread....> > cheers, > Ohad > > > >-- Nigel Kersten nigelk@google.com System Administrator Google, Inc. --~--~---------~--~----~------------~-------~--~----~ You received this message because you are subscribed to the Google Groups "Puppet Users" group. To post to this group, send email to puppet-users@googlegroups.com To unsubscribe from this group, send email to puppet-users+unsubscribe@googlegroups.com For more options, visit this group at http://groups.google.com/group/puppet-users?hl=en -~----------~----~----~----~------~----~------~--~---