If we shard over

{ x : 1 }

with index

{ x : 1, y : 1 }

, and have chunks A:

{x:2}

->

{x:3}

and B:

{x:3}

->

{x:4}

, a document D :

{ x : 3, y : "abcde" }

is currently included in migrations and other operations for both A and B.

This is b/c when we get index key bounds from the prefix, we complete them with MinKey for min and MaxKey for max, so the index max bound of A is

{x:3,y:MaxKey}

and the index min bound of B is

{x:3,y:MinKey}

, and both ranges overlap D. Included test shows how this results in extra documents removed during migration.

Proposed solution - complete with just MinKey?

The problem is harder to trigger with our standard tests with _id indexes, not sure why.