Retry uploading build events more often

[avdv]

This change retries event upload 256 times on Windows (default is 4).

This remedies an issue (buildbuddy-io/buildbuddy#4467) where the build would fail eventually because the connections to the remote get closed intermittently.

Since we were seeing this mainly for PRs started from forks (where the BuildBuddy API secret is not set), deliberately opened from my fork.

---

Still no dice. Two jobs fail with a timeout error:

ERROR: The Build Event Protocol upload timed out. com.google.common.util.concurrent.TimeoutFuture$TimeoutFutureException: Timed out: NonCancellationPropagatingFuture@57251f74[status=PENDING, info=[delegate=[SettableFuture@12279bf4[status=PENDING]]]]

[aherrmann]

Given

Still no dice. Two jobs fail with a timeout error:

I guess this is something to still experiment a bit further on before a review is needed?

[avdv]

I guess this is something to still experiment a bit further on before a review is needed?

Oh yes, I was too optimistic.

[avdv]

I guess third time's a charm, so the one failing job succeeded on the third try...

At least these changes seem to improve the situation. @aherrmann could you have a look please?

[aherrmann]

Thank you!

[aherrmann]

Just to be sure, does this mean this step is skipped if the condition for collect_logs (i.e. failure()) is false?

[avdv]

Yes:

But I still want to test if this step runs when the job failed, that may not be the case... I'll check that on a different branch and adjust here if it doesn't.