Re: [Tracker] Killing run-away child processes

["Edward Duffy" <eduffy gmail com>]

On 11/15/06, Jamie McCracken <jamiemcc blueyonder co uk> wrote:
> Edward Duffy wrote:
> > Text filtering, metadata extraction, and thumbnail generation are all
> > handled out of process so they can't crash trackerd.  However these
> > processes are called with g_spawn_sync and there's no gaurentee they
> > will come back.
> >
> > Here's a patch that will set SIGALRM for these external processes, it
> > uses the child_setup callback function, so there's no need for every
> > possible child process to set its own alarm.
> >
> >
>
> so we get back a SIGALRM signal which we dont currently handle after 30
> secs?
that's the idea

> dont we need a way to kill these external processes if they timeout?
nope..that's what alarm is for.  Put "while (1);" in tracker-extract.c
if you want to try it out.

> (kinda tricky without the pids although we could execute "killall -9
> tracker-extract" though that wont do for the other types and is a bit
> hacky!)
that's why there's alarm!  trackerd doesn't get the alarm, the child
process does.

> Another way to do this is within tracker-extract with the extraction
> performed in another thread with the main thread being a timer that
> kills itself if it timeouts. We could also have a tracker-filter program
> to do likewise with text filters ( and tracker-thumb too) - it would be
> cleaner but do you fancy doing the work?
I could, but it'll be a while... I'm going out of town for the next
week and a half
> --
> Mr Jamie McCracken
> http://jamiemcc.livejournal.com/