Make test target failing with race conditions

[elsesiy]

As mentioned in #306 running make test fails with some data races.as
I attached the log output and here's what I did:

1. Clone repo
2. make precommit
3. make test

I also had to modify the shebang as proposed in #336.

Env info:
go version go1.13.4 darwin/amd64

make_test_output.txt

[iredelmeier]

@elsesiy do you want to pick this up as part of your work for #306 or do you want someone else to? At a glance, it looks like the problem is that the testExporter (and maybe some other metrics test types) is missing a lock.

[jmacd]

I'll take this one.

[jmacd]

I'm confused because when I run go test -race directly on this test, it fails for me. But when I run make precommit it does not fail for me locally w/ the same code. Is this flakiness?

[jmacd]

It looks like the test should wrap a mutex around its records field. I hadn't written the mutex because in the real SDK export is called in a single threaded context, but it's not in this test.

[jmacd]

Also, I've determined why this hasn't failed very often, since it fails reliably for me on my machine. (1) I wasn't running the make test rule, I thought make precommit would do that for me. (2) the CircleCI machine has very limited concurrency, as opposed to my machine.

[elsesiy]

@jmacd I noticed that too but after digging through the targets I found that race detection is only done on make test. Thanks for the update!

[jmacd]

We added race detection to the CI tests. Thanks for finding this. :)