Caesar | Hey Keybuk so what is the expected behaviour when an Upstart job gets removed? | 00:41 |
---|---|---|
Caesar | i.e. unlinked | 00:41 |
Keybuk | depends | 01:57 |
Keybuk | is the job still running? | 01:58 |
markl_ | should rc-sysinit.conf start on startup? | 05:26 |
markl_ | mine says start on filesystem and net-device-up IFACE=lo | 05:27 |
markl_ | i'll bet this is why i'm not getting a runlevel, the filesystem part is "failing" | 05:28 |
sadmac | markl_: fix the filesystem part | 05:28 |
markl_ | what if i use mount -a instead of mountall --daemon? | 05:31 |
markl_ | mountall seems to be failing badly | 05:31 |
sadmac | markl_: you should debug mountall. Keybuk and ion are probably the only two people in the world qualified to speak on it. | 05:35 |
Keybuk | change the exec line in /etc/init/mountall.conf to have --debug | 05:41 |
Keybuk | what error message do you see on the splash screen? | 05:41 |
Keybuk | it should let you skip the failing filesystems or drop to a shell | 05:41 |
markl_ | heh | 05:50 |
Keybuk | sadmac: slangasek knows the code as well ;) | 05:51 |
markl_ | well it's a little more complicated than that; i'm making an openvz template. which means no console at all; i really need to get upstart to write its debug info to a file | 05:51 |
Keybuk | oh | 05:51 |
Keybuk | upgrade your kernel | 05:51 |
Keybuk | 10.04 LTS needs 2.6.32 minimum | 05:51 |
Keybuk | mountall isn't the only thing that will fail with an earlier version, it's just the first thing that you hit | 05:51 |
Keybuk | mountall, upstart, udev, ureadahead, libdrm, X, plymouth, etc. all depend on features not found in older kernels | 05:52 |
markl_ | ah fun | 05:54 |
Keybuk | (and yes, I know OpenVZ is a huge pile of messy patches and there's no newer kernel version for it - I consider this an OpenVZ problem - and not the only problem with OpenVZ either) | 05:54 |
markl_ | what are some of the others, just out of curiosity? | 05:56 |
Keybuk | the fact the kernel has several well known security vulnerabilities? | 05:56 |
Keybuk | the fact it doesn't provide a POSIX userland? | 05:57 |
markl_ | do you have info on the current vulnerabilities? they typically keep up with the red hat kernels | 05:59 |
markl_ | not a whole lot on google about it | 06:01 |
markl_ | except hypervm, lots about that | 06:03 |
sadmac | markl_: why not kvm or linux containers then? | 06:06 |
markl_ | maybe some day, openvz is just too cool and mature. | 06:09 |
markl_ | i like the efficiency, and lxc is just not usable yet | 06:09 |
markl_ | it does suck that the vz kernel patch is so massive that linus would never accept it | 06:09 |
markl_ | isn't kvm like xen? where you have to boot up a whole new kernel? | 06:10 |
sadmac | markl_: yes. what's the problem? | 06:10 |
markl_ | http://www.hpl.hp.com/techreports/2007/HPL-2007-59R1.html | 06:11 |
sadmac | ok | 06:12 |
markl_ | http://www.brianmadden.com/blogs/videos/archive/2009/08/20/Hypervisors-vs.-OS-Virtualization-Smackdown_2C00_-a-video-from-BriForum-2009.aspx | 06:13 |
markl_ | xen-like systems are good when you don't care about efficiency | 06:13 |
markl_ | but if you need a console, or to mix different operating systems, then it is necessary | 06:13 |
sadmac | markl_: that's a strong statement | 06:13 |
sadmac | markl_: I can tell you that 5% overhead is usually expected on the RHEL virt stack | 06:15 |
markl_ | plus a ton of wasted memory | 06:16 |
sadmac | not for long | 06:16 |
sadmac | the page de-duplication stuff should fix that | 06:16 |
markl_ | yeah, i use it when i need to bring up a windows machine, it is cool for that | 06:18 |
sadmac | what do you need all these containers for then? | 06:19 |
markl_ | i keep each service on its own vm; test environments, different build enviroments for different clients | 06:20 |
markl_ | i can still use different distros at least | 06:20 |
markl_ | and it's great for bringing up a new vm quickly, since installing is just a tar command (or snapshot if you use a san) | 06:21 |
sadmac | that applies to most virt technologies | 06:22 |
sadmac | (well, not a tar command, but san snapshotting or copying the image file) | 06:23 |
markl_ | give it a try next time you are bored and looking for a toy to play with | 06:23 |
markl_ | one guy in the channel had like 1000 vm's going on his machine | 06:24 |
markl_ | hosting companies have been using it most over the past years | 06:24 |
markl_ | anyway, a real upstart question; should i be able to put "echo" in the script/end script section of one of the .conf files? | 06:24 |
markl_ | e.g. instead of mountall | 06:25 |
markl_ | e.g. echo "foo" > /tmp/foo | 06:25 |
markl_ | ok that seems to work. plymouth starts up but that is the only service | 06:27 |
sadmac | markl_: from what Keybuk just said I doubt this will ever work, or that it will make your life easier once it does. | 06:36 |
sadmac | markl_: unless you want to forward port it might be time to look for an older OS or a different container system | 06:37 |
markl_ | this is a fun way to at least get familiar with upstart. how does "emits" work? it looks like many of these scripts seem to do an exec | 06:40 |
sadmac | markl_: emits means a job may emit other events by way of its scripts. I think its used to prevent some odd races but in general its just nice to let upstart know. | 06:42 |
Keybuk | emits doesn't do anything | 06:43 |
Keybuk | it's just documentation | 06:43 |
Keybuk | there so you can run perl over /etc/init and make dotty | 06:43 |
sadmac | Keybuk: ah. didn't it prevent some sort of potential hang at one point? | 06:43 |
Keybuk | no | 06:45 |
Keybuk | markl_: I have the OpenVZ guys some hints for getting an "ubuntu lite" booted on it over 6 months ago | 06:46 |
Keybuk | s/have/gave/ | 06:46 |
markl_ | they are way too into rhel/centos unfortunately | 06:46 |
Keybuk | RHEL 6 uses Upstart | 06:47 |
Keybuk | and tbh, it's not Upstart that's the issue here | 06:47 |
markl_ | cool | 06:47 |
Keybuk | it's De-HALificiation | 06:47 |
Keybuk | the kernel feature that let us remove HAL was only added in recent releases | 06:47 |
Keybuk | so any distro without HAL is going to hit the exact same error you hit | 06:47 |
Keybuk | (iirc. mountall fails because of an assertion error opening a "udev" netlink socket?) | 06:47 |
markl_ | well the question i have is that rc-sysinit is waiting for "filesystem" | 06:48 |
markl_ | does that mean that whatever mountall.conf execs has to return a 0 code? | 06:48 |
markl_ | so that the "emits filesystem" happens | 06:48 |
markl_ | ? | 06:48 |
Keybuk | no, whatever mountall.conf execs has to communicate with Upstart via D-Bus and ask for a "filesystem" event to be emitted | 06:48 |
Keybuk | ignore the "emits" lines | 06:48 |
Keybuk | (upstart does) | 06:49 |
Keybuk | they mean as much as the "author" and "description" lines ;) | 06:49 |
markl_ | ah ok gotcha | 06:49 |
Keybuk | usually replacing mountall.conf with something like | 06:49 |
Keybuk | script | 06:49 |
Keybuk | initctl emit virtual-fielsystems | 06:49 |
Keybuk | initctl emit loacl-filesystems | 06:49 |
Keybuk | initctl emit remote-filesystems | 06:49 |
Keybuk | initctl emit all-swaps | 06:49 |
Keybuk | initctl emit all-filesystem | 06:49 |
Keybuk | end script | 06:49 |
Keybuk | does most of the job | 06:49 |
markl_ | ok looks like i need to read up on initctl, ty!! | 06:49 |
Keybuk | (but with better spelling) | 06:49 |
markl_ | heh | 06:49 |
Keybuk | but you also need to obviously mount things | 06:49 |
Keybuk | mount -a will probably work inside a vz, since you don't have to wait for the hardware | 06:50 |
markl_ | yeah i just need /proc | 06:50 |
Keybuk | though bear in mind you need everything from /lib/init/fstab as well | 06:50 |
markl_ | ok cool, good to know | 06:51 |
Keybuk | that should at least get you up to getty ;-) | 06:52 |
markl_ | does anything read /lib/init/fstab or is it there for documentation? | 06:52 |
markl_ | heh :) | 06:52 |
markl_ | i think i could do what i need now if i ifup the network and start ssh by hand. but that would be taking the easy way out :) | 06:53 |
markl_ | hmm, when i try to execute mount it bombs out of the mountall.conf | 06:57 |
markl_ | is there a restriction on what commands i can run in there? | 06:57 |
markl_ | ah, i see! if you execute a command that doesn't return a 0, upstart will stop executing your script | 07:08 |
markl_ | hmm well /proc is already mounted before mountall is even called | 07:17 |
markl_ | so i'll just make a mountall_vz.conf that does what you suggested up there. | 07:17 |
markl_ | i am not a fan of PFM | 07:17 |
markl_ | how important are "expect daemon" and "task" | 07:20 |
markl_ | and does "initctl emit" require a dbus process? it looks like 10.04 doesn't install any sort of dbus | 07:31 |
markl_ | hmm: # initctl emit virtual-filesystems | 07:47 |
markl_ | initctl: Event failed | 07:47 |
markl_ | ok on that note, bed time | 07:54 |
markl_ | morning | 15:24 |
markl_ | does initctl without --system require the lo interface to be up? | 15:42 |
markl_ | also, what does it mean if initctl freezes when doing something like: initctl emit filesystem | 15:42 |
markl_ | ok i'm starting to dig into the source to figure out "initctl: Event failed". Does this mean that the init process sent back a response? How can I get details on why it failed? | 18:30 |
=== rberger_ is now known as rberger | ||
sadmac | markl_: it means that the event was supposed to make some job start, and that job failed to start' | 18:34 |
markl_ | ah ok so it is not just delivering the event notification, it actually gets a result back. cool | 19:01 |
markl_ | ok how does the init process know what to do with an event named "filesystem" ? | 19:10 |
markl_ | does it just run all of the confs that are set to start on filesystem? | 19:11 |
Keybuk | yes | 19:23 |
markl_ | Keybuk/sadmac: good morning, long time no chat! | 19:26 |
markl_ | what time zone are you guys in? | 19:26 |
Keybuk | markl_: I'm in US/Pacific at the moment; normally UK though | 19:27 |
markl_ | cool, i live in utah, not too far away | 19:27 |
markl_ | no ocean in between, at least | 19:27 |
markl_ | it is interesting that your election is getting so much coverage over here | 19:28 |
jMCg | A couple of months ago I talked with @developer about upstart and Linux Capabilities. Now I run into pam_cap -- since doing a quick and dirty hack with su/sudo wouldn't do it: http://dpaste.com/186954/ -- I was wondering how the plan to implement caps in upstart looks exactly, since this pam_caps seems like an easy way around... | 19:38 |
Keybuk | there isn't a plan | 19:39 |
Keybuk | that isn't to say there couldn't be a plan | 19:39 |
sadmac | Keybuk: in frisco again? | 19:42 |
Keybuk | sadmac: I'm in SF yes | 19:43 |
sadmac | Keybuk: you seem to get over there an awful lot | 19:43 |
Keybuk | part of the job | 19:47 |
ion | keybuk: Do you know whether you’ll get more time for Upstart in the next cycle? | 19:49 |
Keybuk | ion: I can't see how I could possibly get any *more* time for Upstart in the next cycle | 19:53 |
Keybuk | since I've arranged to be 100% working on Upstart for it | 19:53 |
ion | Heh | 19:54 |
* sadmac can't get the caffene into his head any faster | 19:54 | |
sadmac | I don't remember the last time I was awake | 19:55 |
sadmac | Keybuk: what's in SF that Canonical keeps sending you to go see? | 19:55 |
Keybuk | sadmac: this visit was the Linux Foundation Collaboration Summit | 19:56 |
Keybuk | that was last week | 19:56 |
Keybuk | then there was a volcano | 19:56 |
Keybuk | so now I'm not going home until the 29th April | 19:56 |
sadmac | Keybuk: not into ash? | 19:56 |
Keybuk | I avoid cloud wherever possible ;) | 19:57 |
sadmac | Keybuk: I'm going be in SF in late June I think. | 19:58 |
sadmac | Keybuk: perhaps you'll find a reason to be there as well. There MAY be auto racing, and I MAY be dressed as The Stig. No promises. | 19:58 |
Keybuk | lol | 19:59 |
jMCg | There, I added a rant on lunchpad... Or dinnerpad..mmm... | 20:15 |
jMCg | https://answers.launchpad.net/upstart/+question/45946 | 20:16 |
Keybuk | a rant on twitter is usually equally effective | 20:33 |
Keybuk | as is shouting into the wind on a rainy winter's night | 20:33 |
Keybuk | (I don't think anyone reads answers.lp.net :p) | 20:34 |
Keybuk | jMCg: https://bugs.edge.launchpad.net/upstart/+filebug | 20:38 |
markl_ | if i want to register an event "foo" in /etc/init, do i just add a "start on foo" line? | 23:37 |
markl_ | appears to work this time, ok cool | 23:37 |
markl_ | ok what is the point of plymouth, just to show a fancy graphical screen on bootup? | 23:51 |
Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!