Long running task with importlib crashes workers about uvicorn-gunicorn-fastapi-docker

[tiangolo]

Thanks for reporting back and closing the issue!

[SansDK]

I rewrote the module which I previously imported with importlib, so that I can import it at the beginning of the file without it executing all code.
Now, when I call an endpoint, I call a method in the imported module ('run()') which executes all the (long running) code.

The result is still the same. I also already tried making everything async / await with no luck.
This is the log with just 1 worker when calling the endpoint which crashes the worker:

[2019-06-04 12:05:11 +0000] [8] [INFO] ('172.17.0.1', 53413) - "GET /info HTTP/1.1" 200
Start training
Called long-running method 'run()' in module
[2019-06-04 12:05:20 +0000] [17] [INFO] Booting worker with pid: 17
email-validator not installed, email fields will be treated as str.
To install, run: pip install email-validator
[2019-06-04 12:05:27 +0000] [17] [INFO] Started server process [17]
[2019-06-04 12:05:27 +0000] [17] [INFO] Waiting for application startup.
[2019-06-04 12:05:27 +0000] [17] [DEBUG] ('172.17.0.1', 53429) - Connected
Start training
Called long-running method 'run()' in module
[2019-06-04 12:05:33 +0000] [21] [INFO] Booting worker with pid: 21
email-validator not installed, email fields will be treated as str.
To install, run: pip install email-validator
[2019-06-04 12:05:39 +0000] [21] [INFO] Started server process [21]
[2019-06-04 12:05:39 +0000] [21] [INFO] Waiting for application startup.

[SansDK]

Seems that it was a (out-of-)memory problem which caused the workers to be restarted. It was really difficult to debug because it wasn't indicated anywhere, but in hindsight it sounds very logical and I should've paid attention to that before.

For anyone who is running into similar problems; check your memory usage and allocation.