Enable circleci to run tests on windows

[bogdandrutu]

No description provided.

[codecov-io]

Codecov Report

Merging #976 into master will increase coverage by 0.01%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master     #976      +/-   ##
==========================================
+ Coverage   85.65%   85.67%   +0.01%     
==========================================
  Files         188      188              
  Lines       13149    13149              
==========================================
+ Hits        11263    11265       +2     
+ Misses       1435     1434       -1     
+ Partials      451      450       -1     

Impacted Files	Coverage Δ
receiver/opencensusreceiver/octrace/opencensus.go	93.33% <0.00%> (+3.33%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 1076b51...b0f8a1f. Read the comment docs.

[james-bebbington]

Note that when running the go test command successfully on Windows, we still have a number of tests that don't run as per #854.

In addition to that, note that the testbed processes don't get terminated properly on Windows so you will encounter a stall until that's fixed (I wrote a fix/hack for that locally that I can upload when I get a chance).

[james-bebbington]

Thanks for looking into this btw!

[james-bebbington]

Note that when running the go test command successfully on Windows, we still have a number of tests that don't run as per #854.

In addition to that, note that the testbed processes don't get terminated properly on Windows so you will encounter a stall until that's fixed (I wrote a fix/hack for that locally that I can upload when I get a chance).

I created a PR to fix the unit tests that are failing on Windows: #983

---

There is still an issue with running the testbed on Windows though. SIGTERM is not supported on Windows, so the agent process will get terminated via SIGKILL (the code currently waits 10s to perform a SIGKILL if SIGTERM fails - it would be straightforward to clean this code up so that it calls SIGKILL straight away on Windows).

If the test succeeds, everything works okay. But in the case that a test fails due to CPU/Mem Usage, the agent is forcefully killed and the load generator receives an "Unavailable" gRPC response. It then goes into an exponential backoff retry strategy with a 15m max timeout, before checking if a signal has been sent to stop generating data. This is what I thought was a "stall" as mentioned earlier. I guess the SIGTERM must result in a nicer shutdown of the gRPC channel that prevents similar issues on other OSs (I didn't look into the details of how this works).

This could probably be fixed by just running the generateMetrics function in a separate go routine. I can come back to this later if we need a fix, but I imagine its not necessary to run the testbed tests on Windows anyway?

[bogdandrutu]

@james-bebbington for the moment I am not running testbed on windows but this should be good enough.

[bogdandrutu]

@dmitryax PTAL

[bogdandrutu]

Would be good if I can cache this operation since it takes a while. If anyone knows how, I would be happy to review a followup PR :)

[james-bebbington]

A few of the tests are abnormally slow on Windows as well, I'll try to take a look into that if I get a chance later.

Edit: This PR should shave off ~1 min: #1009