Have you seen these error messages before?
10:22:23.481686Z Exceeded hard memory limit of 384 MiB with 400 MiB after servicing 3 requests total. Consider setting a larger instance class in app.yaml.
10:18:23.473577Z Exceeded hard memory limit of 384 MiB with 393 MiB after servicing 7 requests total. Consider setting a larger instance class in app.yaml.
I saw these with:
- Google's AppEngine and a Python Flask webservice.
- Wanting to stick with Free Tier (and dynamic instance F1).
I should not have used the default gunicorn settings of 4 workers, but rather follow their guidance and use 2!
Massive speedup seen, from always >300ms and frequently over 5 seconds to respond, it's now down to just a few milliseconds. Probably because the instance was I guess using all the memory. But when memory really was exceeded the app was killed and restarted resulting in even higher latency.