On a sharded cluster, listDatabases can miss the database being moved through a movePrimary when the movePrimary operation runs concurrently with listDatabases.

Root cause

When listDatabases is executed on a sharded cluster, the mongoS queries all shards sequentially and merges results causing the following race condition scenario:

• If listDatabases queries the recipient shard before movePrimary completes, the recipient doesn't have the database yet.
• If listDatabases queries the donor shard after movePrimary completes, the donor will not report the database.
  As a result, the database being moved may be missing from the listDatabases response.

Annotations

• This issue can also occur with concurrent moveChunk or moveCollection operations in the following scenarios:
• When moveChunk moves an entire sharded collection and this collection is the only one for its database.
• When moveCollection moves a collection that is the only one for its database.

• The race condition with movePrimary can only happen when all the collections of the database being moved are untracked. This is because tracked collections are not moved along with a movePrimary operation.