Kubernetes / Docker currently caches images being run on the cluster, but there is a separate cache on each node. With the completion of our new GlusterFS prototype, we could, in-theory, mount a cluster-wide Docker image cache that would prevent the need to pull the same image to multiple nodes. This apparently does not work for technical reasons.

Another performance increase along the same vein would be to run our own Docker Registry to provide another level of local caching for these images.

This ticket is complete when we have a cluster-wide cache of local Docker images provided by one or both of GlusterFS or a private Docker Registry.