The NoCursorTimeout bit in the OP_QUERY header prevents the server from closing a cursor that's idle for more than 10 minutes. Clients can set this bit if they may spend more than 10 minutes processing a batch of results.

The "aggregate" and "parallelCollectionScan" commands should allow the client to turn off cursor timeouts, too. The current syntax for specifying a cursor returned from aggregate is:

{
  aggregate: "collection",
  pipeline: [ ... ],
  cursor: { batchSize: 1000 }
} 

I propose adding:

{
    cursor: { batchSize: 1000, noTimeout: true }
}

Perhaps "timeout: false" is clearer, though less consistent with the query API.

"parallelCollectionScan" could include the option at the top level of the command spec:

{
  parallelCollectionScan: "collection",
  numCursors: 4,
  noTimeout: true
}

Prompted by this question:

http://stackoverflow.com/questions/24997404/cursor-not-found-when-using-parallel-scan-of-pymongo