Configure a Flink JDBC sink for exactly-once delivery to a relational database using XA transactions

Steps

1. Build the sink with JdbcSink.exactlyOnceSink(...), supplying the SQL statement, a JdbcStatementBuilder, JdbcExecutionOptions, JdbcExactlyOnceOptions, and a driver-specific XADataSource supplier.
2. Set JdbcExecutionOptions.maxRetries to 0, since exactly-once guarantees only hold when retries are disabled; a nonzero retry count can produce duplicates.
3. For PostgreSQL, raise the target database's max_prepared_transactions setting above zero, since XA relies on prepared transactions that PostgreSQL disables by default.
4. For MySQL 8+, grant the Flink database user the XA_RECOVER_ADMIN privilege so Flink can recover in-doubt XA transactions after a failure.
5. Account for databases that only allow a single XA transaction per connection (e.g. PostgreSQL, MySQL); Flink opens a separate connection per XA transaction in that case, so size the connection pool accordingly.