Currently after committing a new collection to the config server, we are refreshing all the shards with chunks as a best effort so they can warm up their cache. Right now it is done sequentially and waiting for the response for about 30 seconds, however, given the nature of this call, a better approach would be to fire and forget the requests, like this. This would improve the performance of the collection creation, specially in large clusters with a lot of chunks where the optimized path was taken.