In this exampe we filter the ChangeMessages rather strictly (only delete messages are allowed to be returned) and we start at a specific Entrypoint (for reproduction you need to specify an own entrypoint for this )
==> MongoDB will start to query the oplog for delete -changemessages. This can take some time, depending on the count of messages from starting-point to the first delete-message.
==> Before the first returning data (MongoDB is still querying) the ChangeStream is closed.

Expected Behaviour:
The Next-Call rejects with a "MongoError - Closed" instantly (or in "near" future at least).

Actual Behaviour:
The Next-Call rejects with a "MongoError - Closed" after quite some time (most likly after the initial query for the changestream is finished).

Running the Script (with multiple Millions of Inserts between Entrypoint and Head of te Oplog):

closed!
default: 5:20.648 (m:ss.mmm)

import {MongoClient} from "mongodb";

async function main(): Promise<void> {
  const client = await MongoClient.connect('mongodb://localhost:27017');
  const changeStream = client.watch([
    {$match: {operationType: 'delete'}}
  ], {
    resumeAfter: {_data: "82604F45E4000010452B022C0100296E5A1004B3CAF2964D1040069D88BCB4BD88DF91465A5F6964005A10049439945EFADA4818B6B5F9766ED13A310004"},
  });

  
  const next = changeStream.next();
  await changeStream.close();
  console.time();
  console.log("closed!");
  try {
    await next;
  } catch (err) {
    // Closed error will be thrown here
  } finally {    
    console.timeEnd();
  }
  process.exit();
}
 main().catch(console.error);