I don't understand why these should be flaky on MSAN bots.  They (iirfilter and biquad-automation) just do a bunch of numerical computations and compare with expected results.

See  issue 700139  for a related report.

Aye -- since that report, I think the bot was restarted (see https://bugs.chromium.org/p/chromium/issues/detail?id=700394#c7).

All three of those tests had timeouts before March 10, but the rate of timeouts increased recently (see the flakiness dashboard link).

Do you think it could just be that the tests are relatively slow and should be marked slow? (As is, they timeout after 66 seconds - even when they don't time out, they take around ~50ish seconds).

The tests themselves haven't changed, but maybe they've gotten slower suddenly after Mar 10.  Don't know what would have caused that either.

They might be too slow for MSAN.  Let me look to see if I can make the tests generate fewer samples (and run faster) without losing coverage.

The iirfilter.html test can be sped up by a factor of 4 or so without any apparent loss in coverage.  The bots must be really slow.  My local msan release build runs iirfilter.html in about 1.2 sec (as reported by run-webkit-tests).

Not sure about making biquad-automation faster.  audioparam-method-chaining needs a bit of refactoring to make it run faster. (Too many hard-wired constants.)

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/9efae816aebf5f8547d75b667a39f73527346ba5

commit 9efae816aebf5f8547d75b667a39f73527346ba5
Author: rtoy <rtoy@chromium.org>
Date: Tue Mar 14 21:02:11 2017

Reduce test runtimes for MSAN bots

Reduce the run time for some tests by reducing the render duration,
sample rate or both. The thresholds for biquad-automation had to be
adjusted slightly for the new duration (because the automations now
end sooner.)

On a Z620 machine running linux, the debug runtimes before:

webaudio/AudioParam/audioparam-method-chaining.html
 took: 1.343

webaudio/BiquadFilter/biquad-automation.html
 took: 1.970

webaudio/IIRFilter/iirfilter.html
 took: 2.847

After:

webaudio/AudioParam/audioparam-method-chaining.html
 took: 0.745

webaudio/BiquadFilter/biquad-automation.html
 took: 1.658

webaudio/IIRFilter/iirfilter.html
 took: 1.277

BUG= 701418 
TEST=AudioParam/audioparam-method-chaining.html,
BiquadFilter/biquad-automation.html, IIRFilter/iirfilter.html

Review-Url: https://codereview.chromium.org/2749073002
Cr-Commit-Position: refs/heads/master@{#456826}

[modify] https://crrev.com/9efae816aebf5f8547d75b667a39f73527346ba5/third_party/WebKit/LayoutTests/webaudio/AudioParam/audioparam-method-chaining.html
[modify] https://crrev.com/9efae816aebf5f8547d75b667a39f73527346ba5/third_party/WebKit/LayoutTests/webaudio/BiquadFilter/biquad-automation.html
[modify] https://crrev.com/9efae816aebf5f8547d75b667a39f73527346ba5/third_party/WebKit/LayoutTests/webaudio/IIRFilter/iirfilter.html

I checked the flakiness dashboard just now, and it seems iirfilter and audioparam-method-chaining are passing now.  The runtimes are significantly reduced.  However the biquad-automation test is still timing out.

I'll have to do some more work on reducing that runtime.

One more test that started timing out :-/
webaudio/Oscillator/osc-negative-freq.html

Again, it was a bit slow before (50-60s, occasionally timing out then passing on retry), but now (since something has made the MSAN bot slower) it also started timing out... :-(

https://test-results.appspot.com/dashboards/flakiness_dashboard.html#tests=webaudio%2FOscillator%2Fosc-negative-freq.html&testType=webkit_tests

Stop making the MSAN bots slower! :-)

I'll look to what we can do for osc-negative-freq.html.

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/c9da924bade650770cb6f3a7a0d54c41df840334

commit c9da924bade650770cb6f3a7a0d54c41df840334
Author: rtoy <rtoy@chromium.org>
Date: Wed Mar 15 21:45:36 2017

Reduce test runtimes for MSAN bots

Make biquad-automation run faster by making the duration shorter.
Slightly adjusted thresholds to account for the shorter duration.

On a Z620 machine running linux, the debug runtimes was 1.970 sec.
Now it is 0.753 sec.

BUG= 701418 
TEST=biquad-automation.html

Review-Url: https://codereview.chromium.org/2749383002
Cr-Commit-Position: refs/heads/master@{#457218}

[modify] https://crrev.com/c9da924bade650770cb6f3a7a0d54c41df840334/third_party/WebKit/LayoutTests/webaudio/BiquadFilter/biquad-automation.html

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/db917052bfb936546b2602cdb657c310778c6451

commit db917052bfb936546b2602cdb657c310778c6451
Author: rtoy <rtoy@chromium.org>
Date: Thu Mar 16 23:47:01 2017

Reduce runtime of osc-negative-freq test

Reduce the sample rate and duration of the osc-negative-freq test to
make it run faster for MSAN tests.  Tightened up the thresholds for
the new test.

On a Z620 machine running linux, the debug runtime was 1.414 sec. It
is now 0.860, roughtly 60% as long as before.

BUG= 701418 
TEST=osc-negative-freq.html

Review-Url: https://codereview.chromium.org/2751213002
Cr-Commit-Position: refs/heads/master@{#457608}

[modify] https://crrev.com/db917052bfb936546b2602cdb657c310778c6451/third_party/WebKit/LayoutTests/webaudio/Oscillator/osc-negative-freq.html

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/fdaf992478720559b046d7f953c99c9728a694c4

commit fdaf992478720559b046d7f953c99c9728a694c4
Author: rtoy <rtoy@chromium.org>
Date: Fri Mar 17 19:43:59 2017

Reduce runtime of osc-negative-freq test

Reduce the sample rate and duration of the osc-negative-freq test to
make it run faster for MSAN tests.  Tightened up the thresholds for
the new test.

On a Z620 machine running linux, the debug runtime was 1.414 sec. It
is now 0.860, roughtly 60% as long as before.

BUG= 701418 
TEST=osc-negative-freq.html

Review-Url: https://codereview.chromium.org/2751213002
Cr-Original-Commit-Position: refs/heads/master@{#457608}
Committed: https://chromium.googlesource.com/chromium/src/+/db917052bfb936546b2602cdb657c310778c6451
Review-Url: https://codereview.chromium.org/2751213002
Cr-Commit-Position: refs/heads/master@{#457858}

[modify] https://crrev.com/fdaf992478720559b046d7f953c99c9728a694c4/third_party/WebKit/LayoutTests/webaudio/Oscillator/osc-negative-freq.html

Looks like the last CL landed and all of the slow (flaky) tests aren't anymore.  I am surprised that osc-negative-freq went from 60+ sec to just 8 now.

In any case, I think these are all fixed, so I'm closing this.  

If you see more, please file a new bug.

Thanks rtoy@!

Oh, and  issue 703226  might be relevant.  Don't know how MSAN builds affect V8, but debug builds make some webaudio tests take 3 times longer, and V8 Turbofan was enabled by default around the same time as the MSAN bots started showing flaky webaudio tests.

 Issue 703513  has been merged into this issue.