Looks like this issue is flaky. 😟

I'm going to leave this open and stop commenting.

A human should fix and close this.

When run at the same commit (f2e5277), this test passed in one build (Build Status, Sponge) and failed in another build (Build Status, Sponge).

Error:

    objects_test.go:148: downloadUsingRequesterPays: Object("foo.txt").NewReader: storage: object doesn't exist

Based on the test I don't really see how this flake should be possible as a standalone, given that the object is also queried in the previous lines. Maybe some subtlety with requester pays; @BrennaEpp can you take this?

I'll take a look

This is weird...

• There are no metadata updates to cause errors with delayed propagation, and as @tritone mentions, this object is used before (and after) in the test without failure.
• It's using the same project that owns the bucket for billing, so there really shouldn't be any permission issues.
• The upload is done using preconditions so it should be idempotent (and in any case it should error if it fails to upload).
• I don't see any operations that would cause side effects causing flakiness, but it's hard to tell since there is so much going on and different buckets being interspersed in the test.

I think what I'm going to do here is separate out this part of the test which will make it obvious that nothing else is going on if it fails again.

Bumping down priority as the issue is not related to requesterpays and requires instead some changes to testutil instead (these are in progress).