render-test-flakiness:clear worker storage

[arindam1993]

Clears the mapId keyed caches on the worker so memory is released when prewarm() is used.

This should help keep runaway memory growth when running render tests on CI in check.

Closes #10888

[mourner]

Makes a lot of sense! How's the flakiness looking with multiple runs now?

[arindam1993]

@mourner Still not good :(
I think theres also an issue with IO/file system thrashing because the node server is getting bombarded with actual.png's to write while also having to serve assets

The most common cause of error seems like its coming from tile loading, I'm going to test by skipping the writing of actual.pngs on CI

[arindam1993]

That (not writing actual.png and thrashing IO) seems to have worked, I've had 6 continuous successful runs. I also updated it to move away from the pinned version of chrome, so this closes #10888.

Had one flake happen, but its not like the usual flaking we see with exceptions being thown, instead its this one variable anchor test with a difference in the symbol placement that I think is known to be a bit flaky

.

[karimnaaji]

That (not writing actual.png and thrashing IO) seems to have worked, I've had 6 continuous successful runs. I also updated it to move away from the pinned version of chrome, so this closes #10888.

Had one flake happen, but its not like the usual flaking we see with exceptions being thown, instead its this one variable anchor test with a difference in the symbol placement that I think is known to be a bit flaky

Nice! Should we put this one in the ignore list and have a reference issue for it?

[arindam1993]

I think its fine for now, we can follow up later if these 1-off flakes end up being too annoying.