CreateCollectionCoordinator first chooses the shards that will own chunks and later [commits the metadata to the configsvr|https://github.com/mongodb/mongo/blob/ac8dae158c37de49c03c4139649f775adf888c8d/src/mongo/db/s/create_collection_coordinator.cpp#L1597-L1605. By the time we commit, one of the shards where we place chunks might have been removed. If that happens, some config.chunks documents will point to a shardId that is no longer part of the cluster. This results in:

• The ShardingDDLCoordinator never completing because [this routing table refresh|https://github.com/mongodb/mongo/blob/ac8dae158c37de49c03c4139649f775adf888c8d/src/mongo/db/s/create_collection_coordinator.cpp#L1928-L1929 fails repeatedly.
• Routers see the new collection exists (has been committed), but are unable to write to it.