https://findit-for-me.appspot.com/waterfall/flake?key=ag9zfmZpbmRpdC1mb3ItbWVyvQELEhdNYXN0ZXJGbGFrZUFuYWx5c2lzUm9vdCKGAWNocm9taXVtLm1lbW9yeS9MaW51eCBDaHJvbWl1bSBPUyBBU2FuIExTYW4gVGVzdHMgKDEpLzIzODY0L2FzaF91bml0dGVzdHMvVTJOeVpXVnVUR0Y1YjNWMFQySnpaWEoyWlhKVVpYTjBMa1JwYzNCc1lYbE9iM1JwWm1sallYUnBiMjV6DAsSE01hc3RlckZsYWtlQW5hbHlzaXMYAQw

In this case, the first data point (build-level Swarming task) has 1800 rerun to confirm it is a stable one, the second data point (commit-level try-job) has 100 rerun with 99% pass rate, and the third data point (commit-level try-job) has 100 rerun with 64% pass rate.

The regression range between the first two data points are irrelevant to the test: https://chromium.googlesource.com/chromium/src/+log/f1b84062d58fd80832fccfb828b2d5c25b6b7388..ee07711fafdfd5d785eed3b07797714ae55d8ca6?pretty=fuller
two commits are for layout tests and the other is for perf test framework.

The failure message is:
16918:16924:0927/094440.499581:4860965953:FATAL:resource_provider.cc(1354)] Check failed: sync_token.HasData() || gl->GetGraphicsResetStatusKHR() != GL_NO_ERROR. 


Two possibilities:
1. This test is hard to repro in the first two data points. However it is pretty easy to repro in all following data points.
2. Our algorithm needs some improvement.