Hi -
A long-anticipated speedup of the pmwebd graphite code was implemented
today. It hung not too low but not too high. Please see
git://sourceware.org/git/pcpfans.git branch fche/pmwebd
commit 12a8e1e250426a572ee5a1c3d479923f3966db8b
Author: Frank Ch. Eigler <fche@xxxxxxxxxx>
Date: Tue Aug 25 21:44:25 2015 -0400
pmwebd speedup: graphite mode for multiple metric/instance per archive
Reworked the inner loop of the archive fetching jobscheduler.
Formerly, each metric/instance time series was scanned individually,
(with some multithreaded parallelism). The new code clumps together
all metrics/instances requested against the same archive, for a single
scanning loop pass. (Multithreaded parallelism is thus exploited
across distinct archives only, which is still a win.)
This dramatically accelerates queries like the grafana default
dashboards that ask about network interfaces and memory utilization.
For example, the 3986 metrics found here for a long-span 30-day query
of network traffic for a bunch of hosts went from 50 seconds to 9.
QA is not affected, as it is a pure performance improvement.
|