As a backup for situation where resume token is not available on empty change stream (such is the case for DocumentDB) we are executing a ping command in order to get an operation time to start streaming from.

However there is a difference in command output for self managed and atlas instances

self managed:

rs0 [direct: primary] inventory> db.runCommand({ping: 1})
{
  ok: 1,
  '$clusterTime': {
    clusterTime: Timestamp({ t: 1689951962, i: 1 }),
    signature: {
      hash: Binary(Buffer.from("9c5e50c7ccc0636b26eddf1b3e01ac3256de3123", "hex"), 0),
      keyId: Long("7258283812986028038")
    }
  },
  operationTime: Timestamp({ t: 1689951962, i: 1 })
}

atlas:

{ ok: 1 }

A possible solution is switching to the hello command (we are fine with anything which doesn't generate an oplog entry)