Sorry about that. We didn't see the problem on the chromium.gpu.fyi waterfall because they use top-of-tree ANGLE, the fix landed in ANGLE at the same time the new conformance test did, and I missed adding the new failure expectation for WebGL 2.0 in https://chromium-review.googlesource.com/562857 . Also, the optional trybots test with top-of-tree ANGLE.

Since ANGLE just rolled forward in https://chromium-review.googlesource.com/c/565051/ , this problem will disappear shortly. Sorry again about the inconvenience and please alert us right away if we've broken your waterfall again in the future. I do wonder whether we should switch your bots to use top-of-tree ANGLE like the chromium.gpu.fyi waterfall.

We could. But how often does the revers happen? E.g. ToT angle breaking the bots while stable remains ok?

Good question. To be honest I can think of at least one recent case where an intermittent crash was introduced on top-of-tree ANGLE and Yuly found and reverted it. So if the situation in this bug doesn't happen very often (and sorry for my oversight) I think it's safest to stick with the DEPS version of ANGEL.

The crashes that I handled were on FYI bots which were not part of ANGLE CQ. I believe that v8 bots configurations are covered by ANGLE CQ, so ANGLE changes that may break them should fail to land.

I'd recommend to go ahead with ToT ANGLE, and make sure that v8 configurations are indeed covered by ANGLE CQ.

But, the scenario in this issue is pretty rare, and the existing config worked fine so far, so we may just keep it as is and fix the occasional bugs when they happen.

I think v8 configuration == gpu fyi configuration (except angle version). So, if those broke by an angle ToT change, our bots would have broken too.

I'll wait for a bit more data to make a decision. Happened only once so far. Now I'm aware of this case and tracking it down next time will be much easier.