Write guarantees and retries

Choose when a write answers with the write concern. The driver retries on its own the ones that fail because a node went down.

Requirements

• A connection with a user that has readWrite on the database.

Choose the write concern

db.orders.insertOne({ total: 10 }, { writeConcern: { w: "majority", wtimeout: 5000 } });

Retries

The drivers' retryable writes are on. If the leader of a range goes down, the driver retries the operation in flight. A retried write is not applied twice.

Check

With w: "majority", the write that answered is still there even if the leader that took it goes down.

Common problems

• A w: 1 write can be lost if the leader goes down before a majority has it. Reads do not see it until a majority has it.

See also

• Read guarantees
• Consistency and isolation