MongoDB Atlas: retryWrites, write concern, and read preference that work

retryWrites, write concern, read preference

mongodb+srv://cluster0.abc123.mongodb.net/mydb?retryWrites=true&w=majority

(credentials go in the user-info part of the real string; shown without them here)

What each does

• retryWrites=true: the driver retries a write once after a transient network error. Required for multi-document transactions; without it, transactions fail outright. Modern drivers default it to true, but set it explicitly so a hand-built string is never wrong.
• w=majority: the write is acknowledged after a majority of replica set members have it. This is the durability you want for anything that matters. w=1 is faster and less safe; know which you chose.
• readPreference=primary (the default): reads go to the primary. Transactions require it. Change it to secondary or nearest only for analytics-style reads that tolerate staleness, and never inside a transaction.

Rules

• Keep retryWrites=true on every Atlas connection string. There is no good reason to turn it off against Atlas.
• Retryable writes cover single writes. Multi-document transactions need their own retry loop for TransientTransactionError and UnknownTransactionCommitResult; the driver does not do that for you.
• Do not set read preference to secondary to "reduce primary load" for operational writes. You will read stale data and break transactions.

Verify

Run a multi-document transaction against the cluster. If it succeeds, retryWrites and read preference are correct.

Transaction requirements (retryWrites, read preference primary) and write concern semantics verified against https://www.mongodb.com/docs/manual/core/transactions/ on 2026-09-26.