https://bugzilla.wikimedia.org/show_bug.cgi?id=52829
Faidon Liambotis <fai...@wikimedia.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|RESOLVED |REOPENED
CC| |fai...@wikimedia.org,
| |o...@wikimedia.org
Resolution|FIXED |---
--- Comment #8 from Faidon Liambotis <fai...@wikimedia.org> ---
So, because this checked against the list we already have,
$wgSquidServersNoPurge, which is currently 225 entries of single IPs, this ran
a CIDR match up to 225 times, which in turn increased the appserver load by
about 20-30% -- the CPU cost could be differ a lot depending on hour of the day
(-> service region -> datacenter -> list order :)).
This was deployed yesterday and we had to revert today, which brought
appservers down from 80% to 50% usage and API appservers from 60% to 30%.
Reedy optimized this check with https://gerrit.wikimedia.org/r/#/c/95163/ which
Antoine reviewed and merged, but hasn't deployed yet. We could also reduce our
load in another way, by aggregating our list to CIDR and cutting it down to
10-15 entries at most (which was the original intention anyway).
Since the isInRange seems to be expensive, though, I'd feel more comfortable if
someone took a closer look and optimized the call (e.g. by making it be *just*
CIDR with a simple bitwise operation, not ranges in general) and/or made a
separate array for CIDR ranges as it sounds pretty silly to do such expensive
checks on what is known to have been a list of single IP addresses until now.
--
You are receiving this mail because:
You are on the CC list for the bug.
_______________________________________________
Wikibugs-l mailing list
Wikibugs-l@lists.wikimedia.org
https://lists.wikimedia.org/mailman/listinfo/wikibugs-l
