sync: update CI config files

[web3-bot]

Syncing to commit protocol/.github@7b4d1b6.

[marten-seemann]

Tests seem to fail with race detector.

[iand]

Tests seem to fail with race detector.

Ran tests on Linux with no race failures.

CI failures are of form pubsub_test.go:151: [ValueStore 1] vssolve failed: routing: not found. That test has a 1 second sleep in it so it's possible it's a flake. Will look at whether it can be made deterministic or increase that wait time (:cry:)

update: lowering the sleep below 1 second reproduces the not found error on my machine

[marten-seemann]

Is that something that could be solved by using require.Eventually with a sufficiently high timeout value?

[iand]

32 bit tests are timing out. I'll investigate cause.

[iand]

Swapping order of 32/64 bit tests shows that it's not related to the architecture. Whichever tests get run second seem to timeout after 10 minutes so maybe there is some external state that is not cleared properly when first architecture's tests finish.

[marten-seemann]

How's that possible? Do we write any files on disk?
All memory state should be cleared when go test finishes, right?

[iand]

Yeah all process state should be gone. But maybe the vm is holding sockets or something? I am not entirely sure.

[iand]

It's strange that it's always the same two matrix combinations that fail: Go Test / ubuntu (go 1.17.x) (push) and Go Test / ubuntu (go 1.16.x) (pull_request) whereas the similar combinations Go Test / ubuntu (go 1.16.x) (push) and Go Test / ubuntu (go 1.17.x) (pull_request) pass just fine.

[marten-seemann]

push builds the current branch, pull_request builds a (fake) merge commit. As this branch is rebased on top of master, I'd expect them to be equivalent.
Are we able to gain any insight from the stack trace that's printed on timeout?

[iand]

Haven't spotted anything in the stack trace that points to a specific cause.

[iand]

Adding a sleep in the github action appears to have made some improvement 😢

[iand]

Reverted all my experiments on this branch. I suggest we merge it and then fix the tests in the repo itself. It could be timing issue during teardown of first test suite or setup of following test. Since it only occurs on ubuntu it possibly suggests something related to the linux network stack. /cc @marten-seemann