----- Original Message -----
> On 02/10/2015 11:05 AM, Nathan Scott wrote:
> well the id<->name is 1:1 isn't it? So shouldn't be a lot of extra work??
Yes, but that assertion doesn't follow. :) Its very cheap to find all the
hash names and to see if the set has changed, its more work to find all of
the other metric values there.
> And you could have a containers.id metric whose indom is the name and value
> is the id (to export the mapping). An ascii string name is a lot easier to
> use (and remember) than a whopping 64 character hash id :)
Yes, I definitely thought about wearing the extra cost. Maybe we should,
maybe we will end up having to wear this cost anyway in order to solve...
> >> BTW pminfo -f containers.state.running shows none are running, but I
> >> know
> >> for sure that one of them is.
> >
> > Hmm, thats a good clue. Can you send through the container.pid values?,
> > keen
> > to see what pmdaroot discovered there.
>
> containers.pid should pid is 0 for all containers.
>
> I'll fire up gdb and see where the ENOSUP is coming from (might
> not be until later today ...)
If all your containers are showing up as inactive, and you know some are
running, this is likely to be a big part of the problem. Can you tar up
your /var/lib/docker directory & send it to me (off-list) please?
cheers.
--
Nathan
|