Hmmm, ok, so maybe Frank was right about opening a can of worms
before we're ready to eat ....
However, I see multi-archive support and
(multi-)archive-transitioning-to-live support as related things
worth thinking about together.
In reading everyone's responses, feedback and ideas (thanks!) I
think that the two camps on the "use pmlogger as a server" idea have
possibly arisen because of different notions of what we're trying to
accomplish here. For my initial, mistaken, idea that we are trying
to provide multi-volume support which transitions to live, I had it
in my mind that we were talking about an archive directory which was
being actively written to by a pmlogger from which we wanted to
extract metrics from a time window extending from some time in the
past to some time in the future.
Hopefully, in that context, it is easier to see why I thought that
having clients talk to the active pmlogger as the single source of
all of the metrics might make sense. The client (via libpcp) would
not have to manage the ever-growing final archive volume, the
possible switch to a new ever-growing final archive volume (e.g. if
the current one were reach the 2GB size limit), or the possibility
that logging stops. The pmlogger would continue to log new data as
normal but would also simply relay it back to the client once all
data from the past had been sent. This is how I envisioned the "tail
-f" support working.
It now appears to me (and once again, please correct me if I am
wrong!) that the goal is for clients to be able to examine metrics
from multiple archives, each of which may have multiple volumes
(already supported) and each of which may be arbitrarily distinct
(i.e. not necessarily created by the same pmlogger or even on the
same machine).
It also appears to me that the idea of transitioning to live metrics
means transitioning to any arbitrary source (or sources?) of live
metrics which could be one (or more?) of:
- following an active archive as it grows, similar to 'tail -f'
- switching to an arbitrary active pmcd
Until we all agree on what the ultimate goal is here (and perhaps
I'm the only one who is still not sure), it would seem unlikely that
we can agree on an architecture or design. So please, correct me
where I am wrong and fill in the blanks where I have missed them.
FWIW, I still think that the "pmlogger as a server" idea may still
have value as a conceptually simple way of providing 'tail -f'
support and, as Frank has pointed out, could also provide a way of
accessing archives (with tailing) on remote hosts regardless of how
we do it for locally-accessible archives.
Dave
|
|