Re: How to debug why a service doesn't want to start.

* Charles M. Gerungan <[email protected]>:
> >>For completeness' sake. This can't be seen from the trace, right?
> >
> >Not from the trace you provided (because it didn't follow the
> >execution of your runfile, because you haven't specified an
> >option to follow forks).
> 
> From my the message where I post the trace:
> 
> # truss -faep 28244
> 
> The manpage says that -f does that (Trace decendants of the
> original traced process created by fork(2), vfork(2), etc.).

Yes, you are right. I wasn't fully awake yet, it seems ;)

> That 
> it not what you meant? Or I have attached to the wrong process?
> (See the grepped output of ps in that older message.)

No, you did everything correctly.

I've looked at your trace again, and you can see some parts of
the problem when you look at it

33572: execve("/usr/bin/basename", [ "basename" ], [
"PATH=/command:/
usr/local/bin:/usr/local/sbin:/bin:/sbin:/usr/bin:/usr/sbin:/usr/
X11R6/bin" ])
[ lots of stuff ]
33572: write(2,0x8048af0,78)                     = 78 (0x4e)
33572: exit(0x1)
[ some more stuff ]
33570: break(0x8069000)                          = 0 (0x0)
33570: exit(0x1)


Sadly, truss does not show what write() actually wrote (linux
strace usually does that). But the 2 tells you that the message
was written to stderr. Now, i we look at your runscript again:

#!/bin/sh -e

USER=`basename $PWD` <-- We are here!
SVPATH=/home/$USER/service

exec 2>&1
exec chpst -u$USER runsvdir $SVPATH

So, basename tried to write something to stderr, which apparently
got discarded somewhere. You also see that the shell running the
scripts (33570) exited shortly after calling basename (33572).

This doesn't tell you exactly what and where the problem is, but
it tells you WHERE something is broken, and what is happening.

This is quite a generic way to approach unix problems when you
don't have enough output. The other way is to approach the
problem using logic, like Laurent did. But this only works if you
are already quite knowledgable on said topic.

Hope this helps,

Lukas