transport: fix race accessing s.recvCompress

[menghanl]

issue #4626

Regression caused by #3313

• Fix on HEAD and v1.40.x will be sent separately.

RELEASE NOTES:

• transport: fix race in transport stream accessing s.recvCompress

[menghanl]

@knz This should fix #4626. Can you give it a try? Thanks!

And what test did you run to reproduce this?
I didn't find a way to reproduce locally. It seems to only happen with a race between ctx.Cancel and header received, and I couldn't reproduce it.

[knz]

@menghanl thanks for the fast turnaround on this.
I won't be able to personally test this today, but I'll ask a coworker.
Otherwise I'll come back to it probably this Friday.

[adityamaru]

Hey @menghanl! Thanks for the fix, I took it for a spin in cockroachdb/cockroach#68197 and it appears to have fixed the data race.

We have a test https://github.com/cockroachdb/cockroach/blob/1d0597990de15fc7722d288a4a3ebd50cf06b50a/pkg/gossip/client_test.go#L279 which was previously failing reliably on v1.39.0 with a data race.

[dfawley]

And what test did you run to reproduce this?
I didn't find a way to reproduce locally. It seems to only happen with a race
between ctx.Cancel and header received, and I couldn't reproduce it.

Fix looks good.

We really should try to find a way to reproduce this to add a regression test.