Bug 1240408 - 3% win7 e10s svgx talos regression on inbound (v.46) on Jan 12 from push 8e0a8cdc62ad

Testing :

Baseline: https://treeherder.mozilla.org/#/jobs?repo=try&revision=35be01a98f23

Regression: https://treeherder.mozilla.org/#/jobs?repo=try&revision=3c39a1898cfd

Guh, lots of failures attempting to get profiles. Getting this error:

20:13:23     INFO -  no such option: --download-cache

20:13:23  WARNING - Return code: 2

Apparently when attempting to install requests.

I guess something in our infrastructure changed, and doing old try builds is going to break. Luckily, I found the fix in bug 1240993 - just had to apply https://hg.mozilla.org/try/rev/1c396281c01b

Profiles :

Baseline: https://treeherder.mozilla.org/#/jobs?repo=try&revision=1b23860b127e

Regression: https://treeherder.mozilla.org/#/jobs?repo=try&revision=cd7156cdf85a

BWARRRRR - I forgot to hg add TalosParentProfiler.js. :/ Okay, for reals this time.

Profile builds:

Baseline: https://treeherder.mozilla.org/#/jobs?repo=try&revision=2e8252f137de

Regression: https://treeherder.mozilla.org/#/jobs?repo=try&revision=3c601c831741

Profile links:

Baseline: https://cleopatra.io/#report=7918d1d0b7de549a888fcc4cd39a00045f1bf466

Regression: https://cleopatra.io/#report=7c9ac843bed630812becdc2d3e313418804e4fd6

Need a faster sampling rate, I think...

Profile builds:

Baseline: https://treeherder.mozilla.org/#/jobs?repo=try&revision=fbf532a2c8f0

Regression: https://treeherder.mozilla.org/#/jobs?repo=try&revision=731c301ad5f2

Now with SnowWhite Collection logging:

Profile builds:

Baseline (deaf7c6ca55b ) : https://treeherder.mozilla.org/#/jobs?repo=try&revision=c918a0c82fd1

Regression (8e0a8cdc62ad): https://treeherder.mozilla.org/#/jobs?repo=try&revision=9957c53e7e8d

Now without pausing profiling in the parent.

Profile builds:

Baseline (deaf7c6ca55b): https://treeherder.mozilla.org/#/jobs?repo=try&revision=28e7948d5ab4

Regression (8e0a8cdc62ad): https://treeherder.mozilla.org/#/jobs?repo=try&revision=b7ddf5ed2b78

Profile builds with more instrumentation in NewTabUtils.onLinkChanged:

Baseline (deaf7c6ca55b): https://treeherder.mozilla.org/#/jobs?repo=try&revision=06896c166e73

Regression (8e0a8cdc62ad): https://treeherder.mozilla.org/#/jobs?repo=try&revision=b7ddf5ed2b78

Experiment to remove some setTimeouts:

Baseline: https://treeherder.mozilla.org/#/jobs?repo=try&revision=782137197fac

Regression: https://treeherder.mozilla.org/#/jobs?repo=try&revision=38ee2dca5aec

Comparison: https://treeherder.mozilla.org/perf.html#/compare?originalProject=try&originalRevision=782137197fac&newProject=try&newRevision=38ee2dca5aec

Instrumenting setTimeouts:

Baseline: https://treeherder.mozilla.org/#/jobs?repo=try&revision=cde97ebc080f

Regression: https://treeherder.mozilla.org/#/jobs?repo=try&revision=88e8f35ba27d

Hum. Regression gone. How about if I bump the sampling rate back up? But keep the instrumentation.

Baseline: https://treeherder.mozilla.org/#/jobs?repo=try&revision=ba039985e841

Regression: https://treeherder.mozilla.org/#/jobs?repo=try&revision=3cc2257a7adc

Baseline run cycle:

14:35:51     INFO -  PROCESS | 3028 | [#4] composite-scale-rotate-opacity.svg  Cycles:25  Average:22.00  Median:22.00  stddev:1.38 (6.3%)  stddev-sans-first:0.93

14:35:51     INFO -  PROCESS | 3028 | Values: 27.0  22.0  21.0  22.0  24.0  22.0  22.0  22.0  21.0  21.0  21.0  21.0  23.0  24.0  21.0  22.0  21.0  21.0  22.0  21.0  23.0  21.0  22.0  22.0  21.0

Regression run cycle:

13:02:51     INFO -  PROCESS | 3000 | [#4] composite-scale-rotate-opacity.svg  Cycles:25  Average:23.00  Median:23.00  stddev:1.35 (5.9%)  stddev-sans-first:1.23

13:02:51     INFO -  PROCESS | 3000 | Values: 26.0  24.0  24.0  25.0  24.0  24.0  23.0  21.0  21.0  22.0  24.0  23.0  21.0  23.0  21.0  23.0  21.0  22.0  23.0  22.0  24.0  23.0  24.0  24.0  23.0

Bah, fuck it. WONTFIX’d.