| jefimenko | where are the docs for upstart | 09:37 |
|---|---|---|
| jefimenko | i have a daemon that runs on a typical /etc/init.d script | 09:37 |
| jefimenko | to ensure better availability in case it does crash | 09:37 |
| plautrba | jefimenko: man 5 init | 09:38 |
| jefimenko | i'd like it to restart automatically | 09:38 |
| jefimenko | plautrba: thanks, i've been playing around with things after reading those docs | 10:22 |
| jefimenko | how can i source a config file | 10:22 |
| jefimenko | like a /etc/default/servicename | 10:22 |
| jefimenko | to set environment variables, etc. | 10:23 |
| derosa | Hi, I have a service that must be run at shutdown/reboot. Before upstart I just used a init script in rc[06]. As my sript depends on dbus, the rc[06] solution is no longer valid, as dbus is stopped before my script runs. I tried a little config file that just runs the script when dbus is about to be stopped (start on stopping dbus), but it doesn't seem to be running at all. I appened the "--debug" flag to the kernel but it doesn't help to me. Any hints? | 12:21 |
| derosa | Hi, I have a service that must be run at shutdown/reboot. Before upstart I just used a init script in rc[06]. As my sript depends on dbus, the rc[06] solution is no longer valid, as dbus is stopped before my script runs. I tried a little config file that just runs the script when dbus is about to be stopped (start on stopping dbus), but it doesn't seem to be running at all. I appened the "--debug" flag to the kernel but it doesn't help to me. Any hints? | 12:31 |
| Keybuk | you mean "stop on stopping dbus" surely? | 12:46 |
| derosa | My "service" is in fact just a command that doen's daemonize, so I thought it would be starting | 12:56 |
| derosa | I thought "stop" would only stop something that has previously been started | 12:58 |
| Keybuk | ah right | 13:04 |
| Keybuk | then yes, use "start on" | 13:04 |
| Keybuk | you'll also want to add "task" - that tells Upstart to block D-Bus from stopping until your command completes | 13:04 |
| derosa | like this? | 13:08 |
| derosa | description "nfssyncd final" | 13:08 |
| derosa | start on stopping dbus | 13:08 |
| derosa | task | 13:08 |
| derosa | exec nfssyncd-plymouth-final-sync | 13:08 |
| Keybuk | right | 13:11 |
| derosa | so, that should block until my cmd finiches, right? | 13:13 |
| derosa | *finishes | 13:13 |
| Keybuk | right | 13:13 |
| derosa | it blocks when testing with "service dbus stop", but it doesn't when rebooting with "reboot" or using GDM applet | 13:15 |
| Keybuk | there should be no difference between the two | 13:18 |
| Keybuk | it may be that on reboot, other things that your task needs aren't around anymore | 13:18 |
| derosa | I'll take a deeper look, thank you | 13:20 |
| AnAnt | Hello, I have the following code in /etc/init/usb-blaster.conf: http://pastebin.com/SnPtBur2 | 13:39 |
| AnAnt | the problem is that sometimes the symlink is created, and sometimes not | 13:39 |
| AnAnt | what could be the possible problem ? | 13:39 |
| Keybuk | AnAnt: two "exec" lines | 13:41 |
| Keybuk | only the first will be run | 13:41 |
| Keybuk | between script...end script is just shell | 13:41 |
| Keybuk | just drop both "exec" | 13:41 |
| AnAnt | oh ! silly me | 13:41 |
| AnAnt | thanks ! | 13:41 |
| mezcalero | Keybuk: btw, it's not a very grown-up move not to accept the comment i posted a week ago on your latest blog story, while accepting later ones... | 13:54 |
| Keybuk | mezcalero: ? | 13:55 |
| mezcalero | Lennart says: | 13:55 |
| mezcalero | Your comment is awaiting moderation. | 13:55 |
| mezcalero | May 31, 2010 at 5:35 pm | 13:55 |
| Keybuk | hmm, I don't see anything in the pending queue | 13:55 |
| Keybuk | let me check spam | 13:55 |
| Keybuk | ah, found it | 13:56 |
| Keybuk | it's not a very grown-up move to accuse someone of foul play without even asking them ;-) | 13:56 |
| * Keybuk clears a couple of others stuck in spam too | 13:58 | |
| mezcalero | I still a kid inside ;-) | 13:58 |
| mezcalero | add an "am" there | 13:59 |
| magcius | uggh | 16:17 |
| magcius | somehow I knew this would happen | 16:17 |
| Keybuk | magcius: I knew it would to | 16:18 |
| magcius | Keybuk: thoughts? | 16:19 |
| Keybuk | the only thing we can do is fight | 16:19 |
| magcius | Keybuk: the thing is that the article is half wrong | 16:19 |
| Keybuk | the one about the aliens? | 16:19 |
| magcius | Keybuk: nah, systemd | 16:19 |
| magcius | oh wait | 16:19 |
| magcius | yeah, that's the one too | 16:19 |
| magcius | Keybuk: it says a lot about manual configuration of rules and stuff | 16:20 |
| Keybuk | Lennart doesn't "get" Upstart | 16:20 |
| magcius | Keybuk: but, as I understand it, that's all low-level stuff to make it easy for the system to parse at boot-time, right? There's a higher level | 16:20 |
| magcius | language/compiler thing, right? | 16:20 |
| magcius | Keybuk: sshhh, he's in this room. | 16:21 |
| Keybuk | I know ;-) | 16:21 |
| magcius | Keybuk: and I'm surprised at that. | 16:21 |
| magcius | Keybuk: and we're now falling back to letting D-Bus starting things too | 16:21 |
| Keybuk | D-Bus bus activation is an important part of the plan | 16:22 |
| magcius | yep | 16:22 |
| Keybuk | it does for Linux what Mach IPC does for Apple | 16:22 |
| magcius | oh of course | 16:22 |
| Keybuk | (though Apple seem to change their mind on a release by release basis about whether frameworks should use Mach IPC or UNIX sockets) | 16:22 |
| magcius | Keybuk: so, from what I've heard, the rules files were to make it easy to parse, and were done by hand at first | 16:22 |
| Keybuk | the current upstart config? right | 16:22 |
| magcius | Keybuk: wait, I thought there was a compiler that took a map and generated the rules | 16:23 |
| Keybuk | I don't know what you mean by that | 16:24 |
| magcius | Keybuk: for some reason, I thought there was a higher-level language, and a compiler that took conditions and optimized for them | 16:24 |
| Keybuk | in Upstart? no | 16:24 |
| Keybuk | that'd be the wrong approach anyway | 16:24 |
| magcius | Keybuk: like, on an SSD, the compiler would generate different rules than an 6100 RPM HDD | 16:24 |
| magcius | Keybuk: so how do you handle system services? | 16:25 |
| magcius | Keybuk: like Apache, etc. etc. | 16:25 |
| Keybuk | I'm not sure I know what you're asking | 16:26 |
| magcius | Keybuk: I must be thinking of another init system | 16:27 |
| magcius | Keybuk: but if I have a service like httpd to start at boot time | 16:27 |
| magcius | Keybuk: and I install the package, what rules get put into place to ensure that it does? | 16:27 |
| Keybuk | a configuration file in /etc/init | 16:28 |
| magcius | Keybuk: and is this a sh script that uses the Debian start-stop-daemon stuff? | 16:28 |
| Keybuk | no | 16:28 |
| magcius | ok, so, there's a /etc/init/httpd.rule or something | 16:28 |
| Keybuk | right | 16:29 |
| magcius | and it has "d-bus-started" as a dependency? | 16:29 |
| Keybuk | Upstart isn't a dependency-based system | 16:29 |
| Keybuk | (and why would a web server need D-Bus?! :p) | 16:29 |
| magcius | Keybuk: does anything start before d-bus? | 16:29 |
| Keybuk | lots of things | 16:30 |
| magcius | ok | 16:30 |
| Keybuk | some things start in parallel with D-Bus too | 16:30 |
| mezcalero | magcius: for example if you plug in mod_dnssd into apache it would talk to avahi via dbus to register its services | 16:30 |
| magcius | Keybuk: the systemd blog post mentioned things like the conditionals in Upstart. Do you know how systemd handles these things WITHOUT them? | 16:30 |
| magcius | ah | 16:30 |
| Keybuk | magcius: through launchd-like functionality | 16:31 |
| Keybuk | one service makes an open() or connect() call to the other, which blocks | 16:31 |
| magcius | mezcalero: are you talking about Upstart or systemd? | 16:31 |
| mezcalero | magcius: that sentence was about neither | 16:31 |
| Keybuk | the init daemon has prepared the socket in advance, so when it polls, starts the service and passes the socket along to it | 16:31 |
| Keybuk | thus services are started by demand | 16:31 |
| mezcalero | Keybuk: actually often enough it wouldn't even block, i.e. in the case of syslog | 16:32 |
| Keybuk | Dave calls this pay-as-you-go | 16:32 |
| mezcalero | Keybuk: it would just enqueue and go on | 16:32 |
| Keybuk | mezcalero: indeed | 16:32 |
| mezcalero | Keybuk: and no, services are not started on-demand, generally | 16:32 |
| mezcalero | you can do that | 16:32 |
| mezcalero | makes little sense though except in very few cases such as ssh or cups | 16:32 |
| mezcalero | in the general case this is used to parallelize startup | 16:33 |
| mezcalero | not to do things on-demand | 16:33 |
| mezcalero | doing things on-demand simply delays things | 16:33 |
| magcius | mezcalero: uh, why would you ever want to not start sshd on boot? | 16:33 |
| Keybuk | mezcalero: I was referring to launchd there | 16:33 |
| mezcalero | Keybuk: same for launchd | 16:33 |
| Keybuk | I'm not sure how you're going to do things with systemd in Fedora | 16:33 |
| magcius | exactly | 16:33 |
| mezcalero | magcius: why would you? only every other month people to tend to log into their laptops with ssh | 16:33 |
| Keybuk | magcius: it's not a matter of "not starting" ... sshd is "possible to start" on every boot | 16:34 |
| magcius | mezcalero: so | 16:34 |
| mezcalero | magcius: so it is sufficient to have the port open, so that people can connect and then start it on demand | 16:34 |
| magcius | mezcalero: if I want to ssh into my home computer from a remote location | 16:34 |
| Keybuk | magcius: then connecting to the open ssh port (opened by init) would result in sshd starting, and being passed your connection (actually the listening socket) | 16:34 |
| Keybuk | the first connection would take longer due to the overhead of starting ssh | 16:35 |
| magcius | Keybuk: yeah, gotcha | 16:35 |
| mezcalero | magcius: from the clients' perspective little changes... the port is open with or without it | 16:35 |
| magcius | Keybuk: I forgot that systemd held control of sockets | 16:35 |
| mezcalero | but anyway, for stuff like dbus or syslog we don't do on-demand loading | 16:35 |
| Keybuk | but then sshd would be running (until a quit time of its own choosing) so subsequent connections while running would be normal | 16:35 |
| mezcalero | we start all that in parallel | 16:35 |
| magcius | does systemd hold all sockets or just the standard <1024 | 16:35 |
| mezcalero | Keybuk: actually for sshd we chose to do it like macos, and do it in inetd-like per-connection mode | 16:35 |
| mezcalero | Keybuk: but thats the only daemon where we do that | 16:36 |
| Keybuk | mezcalero: makes sense, since then you can isolate each login | 16:36 |
| magcius | about networkmanager, here's another thing I hate | 16:36 |
| mezcalero | Keybuk: yepp, it's really nice actually because you get a seperate cgroup named after the connection parameters for each instance | 16:36 |
| magcius | if you have no wireless, you can't connect to localhost | 16:36 |
| mezcalero | Keybuk: o.e. the cgroup carries the clients ip addresses and such | 16:36 |
| magcius | so you have to go in and uncheck "Enable Wireless" | 16:36 |
| Keybuk | magcius: yes you can | 16:36 |
| Keybuk | if you can't connect to localhost, you have a bigger problem | 16:37 |
| magcius | Keybuk: last time I tried, it returned E_AGAIN | 16:37 |
| magcius | Keybuk: I believe it was either gethostbyaddr or getaddrinfo that failed | 16:37 |
| Keybuk | sounds like your box is misconfigured | 16:37 |
| magcius | it was a fedora box :P | 16:42 |
| Keybuk | magcius: I don't really know much about Fedora, sorry | 16:42 |
| Keybuk | while recently there's been some movement towards making things like Fedora and Ubuntu more similar | 16:43 |
| Keybuk | recent events mean it's more likely they're going to become very different again, possibly even incompatible | 16:44 |
| mezcalero | Keybuk: hey, you are welcome to adopt systemd in ubuntu too, to work against that! ;-) | 16:56 |
| Keybuk | mezcalero: not going to happen ;-) | 16:56 |
| mezcalero | Keybuk: tss, just admit it, deep in your heart you want it too! | 17:06 |
| Keybuk | mezcalero: it'd certainly make my life much easier | 17:07 |
| Keybuk | but then we'd have to rename our distro to Ubuntu Lennarx | 17:07 |
| mezcalero | i am a humble man, no need for that ;-) | 17:11 |
| mezcalero | Keybuk: btw, have you seen how i love ubuntu and now am a facebook fan of ubuntu? if that's not proof that i don't hate ubunut i don't know what could be proof enough ;-) | 17:11 |
| * Keybuk has one of the "I am Fedora" t-shirts | 17:12 | |
| Keybuk | can we make out now? | 17:12 |
| mezcalero | nope, sorry, not into dudes. Also, I have more Suse shirts than Ubuntu shirts... | 17:13 |
| mezcalero | but otoh i have more ubuntu pens than any other distro pens | 17:13 |
| sadmac | mezcalero: ...so who does that leave you making out with? | 17:14 |
| mezcalero | but in regards to usb keys mandriva i love the most | 17:14 |
| mezcalero | sadmac: Suse is a german girl's name... | 17:14 |
| sadmac | mezcalero: also wasn't it Novell that inserted that cute Linux girl into the I'm a Mac/I'm a PC commercials? | 17:16 |
| Keybuk | oh, usb keys I'm totally for ubuntu | 17:16 |
| Keybuk | we have boxes full of them at the office | 17:16 |
| Keybuk | I never need to pay for portable storage again | 17:16 |
Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!