----- Original Message -----
> ...
> After checking some more this morning, it appears running in
> unprivileged mode does not affect any of the tracing for the metrics,
> however it breaks pmstore for the gfs2.control.glock_lock_time metric as
> it is now unable to write to the trace-point enable file (needs root).
>
> This is a bit of an oversight on my behalf, only checked the metrics not
> pmstore :S.
Ah, OK - that sounds like a hole in the testing too.
> It looks like the pmda will have to run privileged for now,
> I shall roll back that last commit on my end.
OK. Also had a quick look over the other changes, and was a bit concerned
about the way output filtering has been removed in other areas - the loopN
change in particular, deals with non-determinism in the order of evaluation
of /dev/loop[01] - can that be either resolved differently if you want to
see the [01] part explicitly, or put back as it was perhaps?
There was also a bug introduced in the code that dealt with checking the
index into some arrays ... can you double check that? This was in the fetch
refresh code IIRC (don't have it handy atm) - there were checks on an array
index before accessing the array, and the new code added direct array lookup
before the check - which looked like a sigsegv candidate if the array index
passed in was invalid.
cheers.
--
Nathan
|