Closed markusweigelt closed 2 years ago
Have you tried with bertsky/ocrd_controller from Dockerhub already? Not sure about the ocrd/all version used here.
But I also have not seen that kind of error before. Can you start the built container manually?
No but i just tried it with the same problem. Seams to be a problem with WSL and GPU. I do not have an NVIDIA GPU. Is this necessary for our development?
When i uncomment the reservations of deployment resources it seams to work.
reservations:
devices:
- capabilities: [gpu]
count: 1 # share GPU device across workers (for multi-GPU: use swarm)
Is this a necessary config or nice to have?
Ok the count seams to be the problem. When I uncomment that line than container starts.
Oh, I see! Thanks for digging. So, surprisingly, the semantics is "at least", not "up to".
I did not grasp that directly from the documentation.
So how do we get "up to" configured?
For the moment, we should use all
though (which is "up to").
all
is apparently also too much and didn't worked. So only without count has worked on my system.
all
is apparently also too much and didn't worked. So only without count has worked on my system.
Oh, I see. Thanks!
See https://github.com/bertsky/ocrd_controller/commit/e58e4e4be53a8bb107e3dc6c67eb7ed5fb70e1e9
Ah, sorry, did not see your PR earlier!
Tries to start container from base ocrd directory. Had to set --env-file explicit cause default .env is used from docker-compose without using parameter --project-directory.