Aggregating 50M of records

i have a timeseries collection whose size is expected to have 50M records, right now i’m doing a poc with 3M records in which i aggregate just using group and it takes 20 sec, how can i make it fast? Note: i have a index on source field but mongo ends up doing collscan.

Below query:

    $group: {
      _id: "$source",
      sum: {
        $sum: 1