fix(tests): flaky test in BucketFallbackMapper tests

[volmedo]

Fixes #103

The original implementation launched an HTTP server for the test concurrently in a different goroutine. This poses a race condition between starting the server and running the tests. Sometimes, the test was run before the server was ready and the test failed.

Finding the cause has been tricky because, in case of an error in the HEAD request, BucketFallbackMapper swallows the original error and returns ErrKeyNotFound, masking the real cause of the issue. This error handling has been improved to avoid this.

To fix the test, the regular HTTP server being used has been replaced with an httptest.Server. It simplifies the test code as handling its shutdown is easier, but the reason it fixes the issue is that httptest.Server prepares the network listener synchronously, and then starts the server concurrently. That means that, even in the case the test is run before the server is ready to server requests, the listener will be able to buffer them and prevent the test from failing.

[codecov]

Codecov Report

Attention: Patch coverage is 25.00000% with 3 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
pkg/aws/bucketfallbackmapper.go	25.00%	2 Missing and 1 partial ⚠️

Files with missing lines	Coverage Δ
pkg/aws/bucketfallbackmapper.go	76.92% <25.00%> (-4.71%)	⬇️

... and 1 file with indirect coverage changes

[alanshaw]

LGTM - and less code! I wonder why there was a new server for every test case before? I assume it's fine that there's just one?

[volmedo]

yeah, I guess each test case having its own server aligns with the guideline of isolating the tests as much as possible, but the handler is always the same, so I think it's ok in this case.