On Wed, 03 Dec 2014 14:44:30 +0100 Ingeborg Hellemo ingeborg.hellemo@uit.no wrote:
The transition from NAV 3.15 to 4.X has made Geomap incredibly slow. If I change zoom-level it takes several (four) minutes before new dots and lines show up.
I guess this has to do with graphite's webserver not being able to deliver data fast enough (5000 lines of "POST /render/ HTTP/1.1" 200 XXX "-" "Python-urllib/2.7" in httpd-access.log) . The httpd process runs at 100% WCPU
Your observation is correct. This is one of the sad facts of the transition from RRD to Graphite.
The design of Geomap is geared towards lazy-loading the traffic stats of each interface individually - because, I think, the subset of nodes from the total NAV topology that are actually going to be in the Geomap display port is decided very late in the process.
This worked reasonably well when local RRD files were involved. It was basically a short trip from Python->librrdtool->disk subsystem and back.
This design translates very poorly to Graphite, since it is service oriented. Each interface load query is turned into a single HTTP request, with all its associated overhead and lag.
Is there something I can tune to make Geomap usable again?
You could tune the Geomap codebase ;-)
We have previously discussed these issues with NTNU and are painfully aware of the problem, but I think our general sentiment has been "we don't really want to touch the Geomap code with a ten foot pole, and we have other, more important, features and bugs to look at".
I can take another stab at the code, just to get my bearings, but I can't promise anything ATM. Maybe you could have Ole Martin take a fresh-eyed look at it?