Docker Hub rate limits
We are planning to run workloads where our GPU instances will be sparsely used during the night. Since the GPUs will be turned off when idle, whenever a new request comes in, the instance will need to pull our image from DockerHub.
Given DockerHub’s rate limits, we’re concerned that repeated image pulls might result in throttling. Could you share how this situation is generally handled on RunPod? For example, are there caching mechanisms, best practices, or alternative registries we should consider to avoid hitting DockerHub’s limits?
1 Reply
For pulls in the same datacenter, at times the user will pull their image from our cache instead of the Docker Hub. These can feel a little faster, but also reduces the load your API token creates against Docker Hub to help with this very problem.