This thread identified two improvements to scanning CME we would like to address:

• Implement a new command for doing a stateful scan of multiple cluster nodes.

Suggested command:
CSCAN <cluster cursor> <node cursor> [MATCH pattern] [COUNT count] [TYPE type]

The command does an incremental scan of cluster, using moved requests to indicate when the client has finished parsing the requests on the given node. If the command is sent to the wrong node, it will automatically be redirected to the correct one. The scan will guarantee that all data at the start and end of the scan will still hit at least once.

The command will allow use cases that allow external third parties to slowly scan the entire cluster, handling cases where slots may have moved during the course of the migration.

• Once Replace cluster metadata with slot specific dictionaries #11695 is merged, we will implement an optimization in the server so that clients who include a hashtag in their command, will just scan in the provided range.