fix(api): refactor protocol api integration tests to prevent thread leakage

[jbleon95]

Overview

This fixes an issue found where after a recent mergeback commit into edge, API unit tests were hanging and when escaped, would show as a bunch of OSError: [Errno 24] Too many open files.

Through investigation this was found to be caused by too many integration tests that individually called simulate.get_protocol_api. There is a known issue with this in that the threads this creates (Hardware API and for 2.14 and above contexts, Protocol Engine) are not cleaned up properly. This was not causing any issues in our test suite until recently when the magic number of tests that could be run without too many threads opening was hit.

In order to fix this, the simulating context was refactored to be a pytest fixture. In this new fixture, we yield a ProtocolContext and then afterwards, we close the threads. For engine based contexts, we take advantage of the glboal ExitStack() simulate uses to create the PE context to close and unwind the context stack they are created in, allowing it to be properly garbage collected. For older contexts we just call the clean_up function on the hardware API instance.

Test Plan and Hands on Testing

This is a test only fix, so if CI passes this covers this PR.

Changelog

• refactor protocol_api_integration tests to prevent thread leakage

Review requests

Can someone else who has this failing test this on their machine to ensure this fix works for not just me.

Risk assessment

Low.

[sfoster1]

Like it or not, this is what peak performance looks like

[SyntaxColoring]

Makes sense, thanks!

[SyntaxColoring]

I was not having the original issue on my machine, but running make -C api test does still work for me with this PR.

[vegano1]

Tested this on my machine, LGTM.

[sanni-t]

Thanks!