We've set up a MongoDB 3.2.5 cluster on Microsoft Azure with 3 shards (each a replica sets with 3 servers) and a 3-server config set. We access the cluster via a router.

Filling and sharding smaller collections (~400 million documents) worked with sh.shardCollection(). However, after filling a new collection with ~1.8 billion documents we tried to shard this new collection but keep getting the following error:

Note: At first we also had timeouts with count() and chunk balancing. After decreasing the TCP keepalive on all servers in the cluster below 240 seconds (the Microsoft Azure timeout) - based on the info at https://docs.mongodb.org/manual/faq/diagnostics/#does-tcp-keepalive-time-affect-mongodb-deployments - counting and balancing worked again. However, sharding our big collection still does not work.

Might be connected to Issue SERVER-22392