Comment # 1
on bug 1110
from Nathan Scott
(In reply to comment #0)
> Each and every pcp client fetch to a container-aware metric involves IPC and
> context-switching to-from pmdaroot.
Yes, as each and every client fetch involves IPC and context-switching to/from
pmcd - this in itself is not an issue.
> for a single query from pmval (or pminfo), even if there is no lifespan
> change to the set of containers.
The first two (directory) stats are there to (attempt to) ascertain lifespan
changes - if the code continues forward from there (as evidenced here and by
Marks reports earlier), indeed there woud appear to be plenty of scope for
optimisation.
> Worker pmdas should not need to communicate with pmdaroot after a container
> name is resolved at connection time.
The PMDAs need to ascertain whether the container is running at all, somehow,
and whether it has restarted and is running with a different PID since last
fetch. But, agreed, these things can and will certainly need to be optimised.
cheers.