yasinzaehringer-paradime/temporal-sdk-issue

Reproduce memory leak in temporal Python SDK 1.1.0

1. Create venv using poetry
2. Copy env.tpl to .env and populate the fields with your temporal cloud credentials (URL + certs)
3. Start worker via python main.py worker
4. Trigger multiple runs via time poetry run python main.py trigger --timeout=0.1 --sub-activities=0 --sub-workflows=100 &
5. Every run adds ~20MB according to the MacOS activity monitor
6. According to tracemalloc (which prints every 30s( we have a surprising amount of dataclasses hanging around

Remarks:

• The workflow spans 100 sub-workflows which all wait for 0.1s - no activity is run (did not see any weird behaviour when there were only sub-activies)
• gc.collect() has no effect, these dataclasses really do exist somewhere
• Going higher than 100 sub-workflows leads to Potential deadlock detected, workflow didn't yield within 2 second(s) - which is problematic in it's own right