| [02:28] <sadmac> Keybuk: ping | |
| === Md_ is now known as Md | |
| [12:26] <Keybuk> sadmac: pong | |
| [16:57] <sadmac> Keybuk: I was wanting to ask about your current thoughts for the profiles system | |
| [17:44] <Keybuk> sadmac: sure | |
| [17:51] <sadmac> Keybuk: What does your design for it look lik? | |
| [17:51] <Keybuk> I don't have one | |
| [17:51] <sadmac> s/lik/\0e/ | |
| [17:51] <Keybuk> AlexExtreme has been doing one, iirc | |
| [17:51] <Keybuk> the idea being to define profiles in a file | |
| [17:51] <Keybuk> that look something like | |
| [17:51] <Keybuk> disable apache | |
| [17:51] <Keybuk> disable squid | |
| [17:51] <Keybuk> or | |
| [17:51] <Keybuk> disable * | |
| [17:51] <Keybuk> enable foo | |
| [17:51] <Keybuk> enable bar | |
| [17:52] <sadmac> hmm | |
| [17:52] <sadmac> I had another idea for it | |
| [17:53] <sadmac> my idea was to add a stanza to the job definitions, one of either "disable unless" or "disable if" | |
| [17:53] <sadmac> followed by a series of flag names. | |
| [17:53] <sadmac> we could then load a list of set flags off the filesystem at boot time. | |
| [17:59] <sadmac> that does get kind of messy if you have flags with large effects, like single user mode. | |
| [18:08] <Keybuk> yeah, that's more in line with the "flags" idea | |
| [18:09] <Keybuk> flags and profiles are two competing ideas for the same functionality | |
| [18:09] <Keybuk> profiles implies that profile is a first-class object which includes or excludes jobs | |
| [18:09] <Keybuk> and maybe that you only have one profile active at one time | |
| [18:09] <Keybuk> flags is that jobs disable/enable themselves in their own configuration depending on which flags are set | |
| [18:09] <Keybuk> with the implication that multiple flags can be set at one time | |
| [18:10] <Keybuk> the profiles idea is better for a sysadmin, since they don't need to edit the job definitions | |
| [18:10] <Keybuk> flags is vaguely better for a distro | |
| [18:22] <sadmac> Keybuk: with enough flags built into the jobs they become logically equivalent | |
| [18:23] <Keybuk> that's somewhat true | |
| [18:23] <Keybuk> one of the other inits has a feature whereby kernel arguments turn on/off jobs | |
| [18:24] <sadmac> thats easy to do. Doesn't even have to be done in code. | |
| [18:25] <sadmac> an early job definition that takes a peak at /proc could do it. | |
| [19:59] <Torne> does upstart work if it's not PID 1? | |
| [19:59] <Torne> as the child of another init? | |
| [20:00] <Keybuk> no | |
| [20:01] <Torne> aw. how come? :) | |
| [20:03] <Keybuk> you can run it as another pid by disabling a couple of its features | |
| [20:03] <Keybuk> (supervision of forking daemons, mostly) | |
| [20:04] <Torne> nothing i run forks anyway :) | |
| [20:04] <Torne> is that compile time? | |
| [20:04] <Keybuk> why do you want to run it under another init? | |
| [20:05] <Torne> i'm switching distribution | |
| [20:05] <Keybuk> it's "edit the source and recompile" time | |
| [20:05] <Keybuk> basically comment out the getpid() check | |
| [20:06] <Torne> ah, right. i'll experiment in a bit then | |
| [20:06] <Torne> thanks. was just wondering if there was an easier way | |
| [20:07] <Keybuk> not really | |
| [20:07] <Keybuk> it's not something I want to support | |
| [20:07] <Keybuk> since people will complain when things don't work because they're not init | |
| [20:08] <Torne> hm. if it's just a matter of turnign a few thigns off and disabling the check, surely that's a reasonable compile time option | |
| [20:08] <Torne> and anyone who builds one with that on will presumably know what they're getting | |
| [20:09] <Keybuk> adding an option means supporting it ;) | |
| [20:09] <Keybuk> "blah doesn't work when not pid 1" ... "so run as pid 1" ... "but there's an option to not run as pid 1" | |
| [20:10] <Torne> i think you're overestimating the chance of that happening, but hey :) | |
| [20:12] <Keybuk> perhaps | |
| [20:12] <Torne> thanks anyway. i'll fiddle with it. | |