Open rgaiacs opened 2 months ago
Our current assumption is that something goes wrong during the pulling of images when they are not already on the local host. We observed this issue when we tried to launch multiple repositories that required pulling their images from Docker Hub first; 3 out of 5 did not successfully pull.
We talked about (1) drop the GESIS server contribution to the federation to 10 and (2) drop the server total quota to 10.
This might help in reducing the overall load during image pulling. If the issue is indeed related to remote image pulling, I suspect that these reductions could increase the likelihood of successful pulls. However, we should do this only temporarily to debug.
Another step we should take is to increase our local image cache or even set up a local image registry configured as a pass-through. This should also help reduce the number of necessary remote pulls from Docker Hub.
Any ideas beyond that are more than welcome ;)
Do you have access to the logs or monitoring on your registry server(s)?
Do you have access to the logs or monitoring on your registry server(s)?
We are using Docker Hub. No logs or monitoring.
Do you have a paid Docker account, and is anyone else using it? https://docs.docker.com/docker-hub/download-rate-limit/
User type | Rate limit |
---|---|
Anonymous users | 100 pulls per 6 hours per IP address |
Authenticated users | 200 pulls per 6 hour period |
Users with a paid Docker subscription | Up to 5000 pulls per day |
Do you have a paid Docker account, and is anyone else using it?
The images are stored under the gesiscss
organisation name, see https://hub.docker.com/u/gesiscss.
The gesiscss
organisation uses the free tier of Docker Hub.
But the organisation is own by an paid account.
Is the BinderHub registry token associated with the paid user account?
Is the BinderHub registry token associated with the paid user account?
Yes.
@arnim can you write down here you strategy to search for the reason that GESIS search is unstable as mentioned in https://github.com/jupyterhub/mybinder.org-deploy/issues/3056?
We talked about (1) drop the GESIS server contribution to the federation to 10 and (2) drop the server total quota to 10.
After that, you want to stress test the system with old images that it need to pull from Docker Hub.
@arnim Thanks for the help!