Filing this for Pyry so I can stop thinking about it for now :P
Each tracker updates each device row once every 15 seconds, even if it hasn't
really changed, so mb_asof can be updated.
This ends up being another pain point of scaling the tracker count, as each are
doing their own thing.
A quick patch would be to only update if mb_asof from the most recent read is
over 15 seconds old.
An even better one might involve a lock from the tracker. Monitor worker grabs
exclusive lock for updating 'device' rows, then updates any over 15 seconds
old. That stops trackers from selecting from device at the same time then
co-updating all of the rows.
Original issue reported on code.google.com by dorma...@rydia.net on 28 Feb 2012 at 2:48
Original issue reported on code.google.com by
dorma...@rydia.net
on 28 Feb 2012 at 2:48