commit 8488ae024860f0b4a4cd81c5e22c6e39de1e5588
author John Grossman <johngro@fuchsia.infra.roller.google.com> Wed Apr 09 15:42:24 2025 -0700
committer Copybara-Service <copybara-worker@google.com> Wed Apr 09 15:43:44 2025 -0700
tree a853e05dee666d3d2ac4eee685a43e3ae7137700
parent 3cffadf2f5f7eb5e4acbe80f4bc0867eae006134

[roll] Roll fuchsia [pager][test] Add a time limiter to race regression tests.

There are a few different pager tests that we have which are
attempting to exercise race situations which, in the past, have led to
failures.  In general, these races were hard to reproduce in the past,
so the tests attempt to set up a similar race, repeated 1000, or even
10000 times trying to catch a regression.

While not perfect, what this does for us is give us a lot of chances
during the automated test runs to catch any potential regression.
Even if one pass of 1000 attempts does not find anything, the
assumption is that if everyone is trying 1000 times, the regression
would eventually show up in CI/CQ as a flake.  So, the more iterations
we can get in the time that we have on each run, the better off we
are.

Unfortunately, there is another side to this coin.  If the test is
running in a naturally slow environment (say, TCG QEMU for RISCV), and
the test harness is also very overloaded, it is possible that we might
time the test out trying to get to our iteration count, just because
things happen to be very slow in this pathologically bad situation.

So, add a time limit to the system as well.  Basically, try to get X
iterations in within an absolute time limit of Y.  If we run out of
time, don't fail the test, but do print a warning saying that we
didn't get all of the trials in that we would have liked to.  Most of
the time, we expect these things to run to completion, but even if
they only get a fraction of their attempts into a test run, it is
better that nothing (and also better than causing false positive test
flake).

For now, the iteration count limits have been kept as they were, and
the timeout used has been set to 60 seconds.  CI/CQ test timeouts seem
to be on the order to 2 minutes right now, so hopefully this will
avoid flake, even in worst case scenarios.

Original-Fixed: 389991524
Original-Reviewed-on: https://fuchsia-review.googlesource.com/c/fuchsia/+/1249107
Original-Revision: 8c5c7f4ada929056c155f8403b6089c4da9cfd84
GitOrigin-RevId: 25dbe4af31bebc310e952465612e72e2538064cf
Change-Id: I8eb6ea40966a4ef8546934363b0446915c80dab8