Comment # 19
on bug 1067
from Nathan Scott
(In reply to comment #18)
> [...]
> like a separate PR opened for others like 'network.interface.duplex', where
> restricting the instance domain still makes the pmda fetch
> /sys/class/net/*/duplex for all interfaces?
The CPU-time-shaving progress here (which we should expect to be small
in terms of the overall sample time) was mainly convenient fallout from
refactoring the code to deal with net/mnt namespace switching correctly.
If there's evidence of any significant overhead there, sure, go ahead
and open a new bug if you like (or better, just hack on it, if you are
certain there is still something worth fixing there - I don't believe
anyone else intends to continue hacking on this at this stage, unless
there's a genuine observed problem reported).
The real gains from the recent changes were the switch from ioctl to sysfs
by default, resolving the root-cause problem Carlos & Michele identified,
which was not related to the original diagnosis.
Resulting from that switch, I hope to see Carlos' Vector-induced pmwebd
timeouts resolved now - it would be good to have that confirmed, Carlos?