Suppose:
Exists very big table users with fields: id, shard_id (> 10 million records)

|id |shard_id|
| 1 |NULL    |
| 2 |NULL    |
|...|NULL    |

Need to be done:
Set a random value from 1 up to 10 for each NULL valued shard_id field.

Solution should:

• Allows to stop updating at any time, fix the code and run it again, without updating processed data
• Be resilient to exceptions in case of updating individual records
• Scalable. Allows to run it multithreaded
• Should assume and display a potential time of ending
• Work in real production system without stopping the main app

Optional/

• Write an API to get number of unprocessed records

curl http://localhost:3000/api/sync/users/count

Possible to use any library