APM agent not sending event requests to the APM server with gunicorn wsgi.

I’m running a flask application and everything works fine if I’m running the application in dev environment i.e. directly from command line or my IDE aka Pycharm.

But when I run the same app on production via gunicorn, it only sends agents requests and no events requests so in APM in kibana, no service shows up.

Reproduction steps:

1. Create a basic flask app and call client.begin_transaction and client.end_transaction before and after each request so as to log time of each request.
2. Run the app via gunicorn.
3. Make a request to the app via browser or postman.

Expected behavior:

The apm server logs show /intake/v2/events requests as well as /config/v1/agents request.

Environment/

• OS: macOS
• Python version: 3.5
• Framework and version: Flask 1.0.2
• Elasticsearch version: 7.4.1
• Kibana version: 7.4.1
• APM Server version: 7.4.1
• Agent version: 5.2.2

Additional context

I’m running Elasticsearch on 0.0.0.0:9200, Kibana on 0.0.0.0:5601, and APM on 0.0.0.0.8200.

A curl request to all these shows the desired result.

[basepi]

Thanks for the update. Forking models for concurrency have long given the agent issues, as we lose the background threads on fork. Rather than trying to come up with a solution for each implementation, we’re going to focus on #636 so that we will automatically detect forks and re-create those background threads. This should fix the majority of issues like this one.

[wodCZ]

I can confirm that disabling preloading in gunicorn fixes the issue - transactions are now properly saved.

We’re using this docker CMD with no configuration file:

CMD ["gunicorn", "--bind", ":8000", "--worker-tmp-dir", "/dev/shm/", "--workers", "2", "--threads", "4", "--worker-class", "eventlet", "--log-file", "-", "project.wsgi"]

I also had to switch the transaction type to request in the APM UI: