slangasek | mathiaz: for new stuff, definitely preferred to modify the job directly; for existing stuff, I've tried to respect the existing configs when converting | 00:26 |
---|---|---|
mathiaz | slangasek: great thanks | 00:27 |
mathiaz | slangasek: I'm writting my UDW session about server packages | 00:27 |
mathiaz | slangasek: and plan a topic about upstart jobs | 00:27 |
slangasek | ah :) | 00:27 |
mathiaz | slangasek: do you have specific ideas about upstart features useful for daemons? | 00:27 |
mathiaz | slangasek: I already have: supervision | 00:28 |
mathiaz | slangasek: support for forking daemons | 00:28 |
mathiaz | slangasek: examples of running daemons in foreground | 00:28 |
slangasek | I was intending to put together a talk about authoring upstart jobs for UDW, but I didn't see a call for sessions this time around (just an announcement of when it was that implied the slots were already filled) | 00:28 |
slangasek | mathiaz: automatically cascading restarts? | 00:28 |
mathiaz | slangasek: oh nice | 00:29 |
mathiaz | slangasek: examples? | 00:29 |
slangasek | (service portmap restart -> restarts portmap, gssd, statd) | 00:29 |
slangasek | not sure that's actually /needed/ for either of those services, but in practice it will happen :) | 00:30 |
mathiaz | slangasek: is this already available in lucid? | 00:30 |
slangasek | yes | 00:30 |
slangasek | and in karmic | 00:30 |
mathiaz | slangasek: do you have example of an upstart job that starts a process that forks | 00:30 |
mathiaz | slangasek: IIRC there is the fork keyword available | 00:30 |
mathiaz | slangasek: or expect fork - something like that | 00:30 |
slangasek | mathiaz: many do... ssh, portmap, atd, et al | 00:31 |
ion | Upstart’s fork-following code is quite fragile currently. The next-gen Upstart will have a proper implementation that should handle any situation well. | 00:31 |
slangasek | really? that's not a caveat Keybnz has mentioned :) | 00:31 |
mathiaz | slangasek: any other ideas about Upstart feature relevant for managing daemons? | 00:32 |
ion | He has mentioned it repeatedly. | 00:32 |
slangasek | mathiaz: built in support for oom handling? | 00:33 |
slangasek | ion: to you, apparently, but not anywhere that I've picked up on it... | 00:33 |
slangasek | mathiaz: +nicing +limits | 00:33 |
mathiaz | slangasek: hmmm - any examples of this use? | 00:33 |
slangasek | mathiaz: ssh in lucid; for the other two I don't have live examples, just init(5) | 00:34 |
ion | If the job’s main command consistently does a fork or a double fork and you tell Upstart about it with the expect stanza, things will work. | 00:34 |
slangasek | ion: oh, fragile in that you have to tell upstart what to look for, sure | 00:35 |
ion | If the main command is a shell script that runs stuff and then execs the main daemon *that forks*, that’s likely to break the current implementation. | 00:35 |
mathiaz | slangasek: great - thanks for the input! | 00:35 |
slangasek | ion: right, that was one of the points for my theoretical UDW talk, under the heading "how to write an upstart job that you have to reboot to fix" :) | 00:35 |
mathiaz | slangasek: I also noticed that there wasn't a reload command | 00:35 |
ion | If the shell script does a fork-and-exec for an external command, the code will shit itself. :-) | 00:35 |
mathiaz | slangasek: hm - sorry | 00:36 |
mathiaz | slangasek: a force-reload command | 00:36 |
ion | The proper implementation in Upstart 0.10 or whatever it’ll be called will handle that, too. | 00:36 |
mathiaz | slangasek: as mentionned in the Debian policy | 00:36 |
slangasek | mathiaz: Debian policy describes the behavior of init scripts in /etc/init.d; the upstart-job shim implements force-reload | 00:37 |
slangasek | (though I'm not sure the 'reload' command is the right backend for that, maybe we have a bug there) | 00:38 |
mathiaz | slangasek: is there a way to specify in the upstart job the command to be used for reload? | 00:38 |
mathiaz | slangasek: ie send SIGUSR2 to the process for example? | 00:38 |
slangasek | no | 00:41 |
LLStarks | okay. can somebody tell me why reinstalling packages doesn't actually reinstall files? | 00:49 |
LLStarks | if i reinstall the pulseaudio package after rm'ing the /etc/pulse directory, i expect it to be restored. | 00:49 |
LLStarks | i shouldn't have to purge and then install | 00:49 |
RAOF | LLStarks: /etc/pulse/ contains conffiles (as defined by dpkg), and dpkg makes sure that your changes to conffiles are preserved across installs. | 00:50 |
LLStarks | how do i override? | 00:50 |
RAOF | Well, you've found one way; purging will remove dpkg's memory of your conf file changes. | 00:50 |
LLStarks | well, that can break and uninstall other packages | 00:51 |
LLStarks | is there a more direct way? | 00:51 |
RAOF | dpkg --purge --force-depends should purge just the package you're interested in, leaving everything else installed (but broken). I think. | 00:52 |
RAOF | Then installing again will reconfigure everything. | 00:52 |
geser | IIRC there are some options to dpkg to force reinstall of missing /etc/ files | 00:53 |
geser | LLStarks: try "dpkg --force-confmiss $the.deb" | 00:54 |
LLStarks | thanks. | 00:54 |
=== TLUL is now known as TLUL_teh_lurker | ||
=== TLUL_teh_lurker is now known as TLUL | ||
=== joshua is now known as Guest57151 | ||
=== v is now known as vorian | ||
=== joshua is now known as Guest23314 | ||
=== vish is now known as mac_v | ||
=== mac_v is now known as vish | ||
persia | slangasek: IF you're still interested in a UDW slot, you can have mine, as I'm just reprising a talk I've given lots of times (and I suspect all would be better served by a good understanding of upstart vs. me blathering about stacktraces again) | 10:40 |
slangasek | persia: when is your slot? I haven't done any prep work yet | 10:51 |
persia | Friday | 10:54 |
* persia digs up the exact time | 10:54 | |
persia | 29th January 20:00 UTC. It's 5:00 saturday local for me, so I really won't miss it :) | 10:55 |
=== arand_ is now known as arand | ||
hyperair | is it possible to create nfs shares without requiring root access? | 13:20 |
hyperair | if it was, then i could possibly expand nautilus-share's scope to include NFS support. | 13:20 |
=== ryanakca_ is now known as ryanakca | ||
=== Eren is now known as Guest34070 | ||
=== Nafallo_ is now known as Nafallo | ||
=== yofel_ is now known as yofel | ||
=== andresmujica1 is now known as andresmujica | ||
=== andresmujica1 is now known as andresmujica | ||
geser | what's the solution for the python and linking with ssl issue? | 19:34 |
ScottK | What's the issue? | 19:35 |
geser | I'm trying to merge vim and got the problem that linking with python failed because of -lssl not being there | 19:37 |
geser | I remember that some other builds had similar issues in the past too | 19:38 |
geser | ScottK: looks like vim pulls in the value from LOCALMODLIBS from /usr/lib/python2.6/config/Makefile and -lssl is listed there | 19:53 |
_Groo_ | hi/2 all | 19:55 |
ScottK | geser: Oh. No idea on that one. | 19:55 |
_Groo_ | you guys are updating to rc2 as i can see, gonna have to update my kde multimedia pulseaudio as soon as you upload the new multimedia to lucid | 19:56 |
ion | hi/2 looks like an Erlang function reference. | 19:58 |
_Groo_ | ion: actually its an os/2 old warrior reference | 19:59 |
_Groo_ | ScottK: is rc2 be up 100% today (also with kdebindings for the first time? oO | 20:00 |
ScottK | _Groo_: I know it's building now. I didn't help with the packaging this time around, so I don't know about kdebindings. | 20:01 |
* _Groo_ cross is fingers | 20:01 | |
_Groo_ | ScottK: but the complete package will be uploaded today? | 20:02 |
ScottK | _Groo_: It's all uploaded now. In the process of building. | 20:02 |
_Groo_ | ScottK: sorry i mean published :) do you have a url i could see fo check for building progress? | 20:02 |
ScottK | _Groo_: For kdebindings? | 20:03 |
ScottK | There isn't one url for all of KDE. | 20:03 |
_Groo_ | ScottK: no, i mean, are you using a ppa to test or something like that? i wanna check what packages are built and published already | 20:03 |
ScottK | _Groo_: For Lucid, it's in the Ubuntu archive. | 20:04 |
_Groo_ | ScottK: link pls? | 20:04 |
ScottK | _Groo_: There isn't one for all of KDE. You should probably ask in #kubuntu-devel for someone who was involved in packaging it. | 20:05 |
_Groo_ | ScottK: OMG i forgot that konversation removed that tab lol.. you are right | 20:07 |
qense | What is the difference between Python packages in /usr/share/pyshared/ and those in /usr/share/? | 21:02 |
ScottK | Generally the ones in /usr/share/pyshared are designed to (properly) support more than one Python version on a system. | 21:35 |
Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!