Fix race condition in writePB().

[juliusv]

The RLock already needs to be acquired when reading r.dimHashesByName.

This fixes #61

@beorn7 @stapelberg

[grobie]

What about adding the referenced example to the test suite to ensure the advertised behavior of concurrent/late registration of metrics work? This will also allow to check for race conditions with the test suite.

[juliusv]

@grobie I could add a test case, but don't know a way to test the race directly. I'd have to add a separate Makefile step to run that specific test with the race detector enabled. Is that worth it? I think if we have races again, it would be much more likely a different race that that particular test wouldn't catch anyways, and it's not possible to run the whole test suite with -race enabled because it uses too much RAM.

[beorn7]

👍

[beorn7]

@grobie @juliusv In principle, think it makes sense to add a test for it, and then mention in the doc string that it can be run with -race to test for races. We have a similar case in prometheus/prometheus/storage/local. The test would be able to detect some other races, too, especially if it's a bigger fuzz test of some kind.

We can do that later, though.