[Agent-Python] AsyncIO + ThreadPool Span Tracing

[gautam-ergo]

Describe the bug:

Current setup: A flask endpoint that summons a sync function running asyncio code block and a threadpool executor that dovetails on the results of the asyncio code block

Looking for a way to monitor the function execution via the instrumentation of code running in those several threads.

• Both @elasticapm.capture_span() and async_capture_span were not able to trace the function execution as desired.

• Implemented the workaround suggested in #499 but did not work in my case.

  Any advice/directions on this would be of great help. TIA.

Environment (please complete the following information)

• OS: System Version: macOS 10.14.6 (18G6020) Kernel Version: Darwin 18.7.0
• Python version: 3.8.0
• Framework and version: Flask 1.1.4
• APM Server version: 7.6.2
• Agent version: 6.4.0

Additional context Attaching a blueprint of the function block that gets triggered on hitting the flask request endpoint.

Essentially a bulk indexer job that posts to ES /_bulk endpoint post various data pre-processing.

On the APM front, only the request call/response duration is being captured by the agent:

Any advice/directions on this would be of great help. TIA.

[beniwohli]

Hi @gautam-ergo! I experimented with patching ThreadPoolExecutor to make the transaction/span available in the executing thread a while ago in #705. I updated that branch just now to current master. If you have a test environment available, it would be great if you could give it a try by installing it like so:

pip install https://github.com/beniwohli/apm-agent-python/archive/refs/heads/thread-executor-support.zip\#egg\=elastic-apm\=\=100

[gautam-ergo]

Thanks and appreciate the prompt response @beniwohli , will try this out and post the results; just one clarification, do I instrument the code with begin_transaction and end_transaction in conjunction with capture_span or is the flask's app level configuration just about enough to capture the info

[beniwohli]

The stock flask integration should be enough :+1:

[gautam-ergo]

Hey @beniwohli tried testing with the updated branch, APM is not able to pick up the trace info as expected, apm output is the same as the one I had posted earlier...not sure if i missed anything with the stock integration, any merit in trying the approach outlined in this repo's asyncio-tests