On Fri, 24 Feb 2017 13:17:18 +0100 Ingeborg Hellemo ingeborg.hellemo@uit.no wrote:
But interesting things are happening. A lot of jobs have overhead-values in the 0:00:00.X range like the example above, but then suddenly a bulk of jobs have overhead values up to 10 seconds!
Well, I couldn't resist playing around a bit with this. Here's a patch to log some timings specific to threaded work, if you're interested:
https://gist.github.com/lunkwill42/6c14ea6829270b93a39829a9ee0c9c2b
The code path is only enabled when DEBUG logging is enabled for `nav.ipdevpoll.db.timings`. The logged overhead value here will represent the time spent waiting for an available thread worker to run the SQL code.
There is a margin of error here, where ipdevpoll will spend its time running other code, but I can see the overhead increase quickly, but temporarily, once ipdevpoll is getting very busy - while increasing the threadpool size will visibly work against these overheads.
Looking forward to 4.7
Ditto :)
Have a great weekend, Ingeborg!