ISSUE DESCRIPTION AND IMPACT
As a workaround for SERVER-23192, MongoDB 3.2.10 introduced an option where a node never stops monitoring a replica set once it has started, no matter how long it appears to be down for. Using this option means you can encounter problems if you remove a shard then add back a shard with the same replica set name.

This parameter is set to false by default, and can be set by executing following command:

db.adminCommand( {setParameter: 1, 'timeOutMonitoringReplicaSets': true} )

DIAGNOSIS AND AFFECTED VERSIONS
This option is included MongoDB 3.2.10 and subsequent releases of MongoDB 3.2. Please note that it is not included in MongoDB 3.4.

REMEDIATION AND WORKAROUNDS
If the operator wishes to re-add the shard using different hosts at a later date, the operator has two choices:

1. Restart all the affected nodes.
2. Toggle the timeOutMonitoringReplicaSets server parameter introduced in SERVER-25516 from false to true on each affected node. Once the the shard is discovered, switch timeOutMonitoringReplicaSets back to false, usually this process takes about two minutes.