service: Add support for ETW logging

[bkeryan]

• This contribution adheres to CONTRIBUTING.md.

What does this Pull Request accomplish?

Add support for ETW logging using Microsoft's traceloggingdynamic package.

Notes:

• PerfView is able to correctly calculate elapsed time for these log messages.
• I wasn't able to get NI's internal ETW viewer to identify these as paired events. I tried specifying event ids, but sometimes the viewer mismatched client/server calls together.
• This code currently doesn't push/pop the current thread's activity id, so the related activity id is always 0.

Why should this Pull Request be merged?

As a developer, I want better visibility into MeasurementLink measurement execution time

What testing has been done?

Manually tested with PerfView and NI's internal ETW viewer.

[bkeryan]

Example of PerfView output:

[github-actions]

Test Results

     12 files  ±0       12 suites  ±0   1m 54s ⏱️ -5s
   182 tests ±0     157 ✔️ ±0    25 💤 ±0  0 ❌ ±0 
2 172 runs  ±0  1 872 ✔️ ±0  300 💤 ±0  0 ❌ ±0 

Results for commit 922d359. ± Comparison against base commit f54b16e.

♻️ This comment has been updated with latest results.

[jasonmreding]

I wasn't able to get NI's internal ETW viewer to identify these as paired events. I tried specifying event ids, but sometimes the viewer mismatched client/server calls together.

Can multiple start events be emitted before the corresponding stop is emitted? If so, there should be a way to annotate the event as recursive, and you can see if that makes a difference. You can look at the TransactionStart event on the PF event source for an example of this. I'm not sure if a similar concept exists or not in the Python library you're using.

The pairing logic in the viewer has always been a little finicky. I thought there were three different options for trying to pair up event start/stop data. Do none of them work? I guess if the lack of recursive option is causing the events to get dropped, it would probably throw off all of the pairing algorithms.

You could also generate a unique ID that is included with the output of the start/stop event and then use that unique ID to pair start/stop events. That requires hard coding knowledge of that pairing for that event source in ETW Viewer so it's kind of a pain. We could probably generalize the pattern if we find it useful. That might also incur some boxing overhead when logging unless you also provide a matching override of WriteEvent in the event source.

[bkeryan]

Run poetry run pytest ./tests -v --cov=ni_measurementlink_service --junitxml=test_results/nims-windows-latest-py3.8.xml
ImportError while loading conftest 'D:\a\measurementlink-python\measurementlink-python\tests\conftest.py'.
tests\conftest.py:9: in <module>
    from ni_measurementlink_service._internal.discovery_client import _get_registration_json_file_path
ni_measurementlink_service\__init__.py:12: in <module>
    from ni_measurementlink_service.measurement.service import (
ni_measurementlink_service\measurement\service.py:30: in <module>
    from ni_measurementlink_service._internal.discovery_client import DiscoveryClient
ni_measurementlink_service\_internal\discovery_client.py:19: in <module>
    from ni_measurementlink_service._loggers import ClientLogger
ni_measurementlink_service\_loggers.py:29: in <module>
    from ni_measurementlink_service import _tracelogging
ni_measurementlink_service\_tracelogging.py:11: in <module>
    _event_provider = traceloggingdynamic.Provider(b"NI-MeasurementLink-Python")
.venv\lib\site-packages\traceloggingdynamic.py:1265: in __init__
    self.id = id if id != None else providerid_from_name_bytes_utf[8](https://github.com/ni/measurementlink-python/actions/runs/5966151523/job/16185221550#step:7:9)(name_utf8)
.venv\lib\site-packages\traceloggingdynamic.py:113: in providerid_from_name_bytes_utf8
    return providerid_from_name(str(providername_utf8, encoding='utf-8'))
.venv\lib\site-packages\traceloggingdynamic.py:[9](https://github.com/ni/measurementlink-python/actions/runs/5966151523/job/16185221550#step:7:10)3: in providerid_from_name
    sha1 = hashlib.sha1(usedforsecurity = False)
E   TypeError: 'usedforsecurity' is an invalid keyword argument for openssl_sha1()
Error: Process completed with exit code 1.

I filed an issue about this: microsoft/tracelogging#58