tree 06f6ce767da3861b68856e375f2bf73c37ac9cd8
parent ac86ad9ba0220dafa3f83ffd077e5b7e466a7c05
author Kevin Lubick <kjlubick@google.com> 1571930192 -0400
committer Skia Commit-Bot <skia-commit-bot@chromium.org> 1571942719 +0000

[gold] Performance enhancements for /digests

Wrote two benchmarks to show 1 easy win and 1 hard win
(pre-caching Test() saves ~20ms in GetDigestDetails and
probably less than that elsewhere; If I were to re-write
lots of things to have a slice of traces instead of a
map, it would save about 15ms of overhead on the range)

Implemented a few small improvements, the aformentioned
pre-caching and being more efficient about looking up
if a trace had a given digest.

I hardened the endpoint against garbage data, in that we
bail out sooner if there's an unknown test or digest
(the previous impl did not do the right thing in this case).

The rest is cleanup that I found while scouring through.

In the end, based on my extensive testing (of two random
data samples [1], [2]), the /json/details endpoint response
dropped from ~2 seconds to ~250 milliseconds (on some large
cases, up to 1/3 of that time is gzipping the response, I think).

With that worst-case scenario addressed, I feel confident in
upping the QPS on this endpoint.

[1] https://gold.skia.org/detail?test=blurredclippedcircle&digest=412961706b174f726893ab4a054a658a
[2] https://gold.skia.org/detail?test=hslcolorfilter&digest=daf52d9be7a334fbfbb38e56d51017cd

Bug: skia:9557,skia:9080
Change-Id: Ifaf14917d8fa3fa9b5a22c85b3ab9b2bbef914a4
Reviewed-on: https://skia-review.googlesource.com/c/buildbot/+/250320
Reviewed-by: Ben Wagner aka dogben <benjaminwagner@google.com>
Commit-Queue: Kevin Lubick <kjlubick@google.com>