A Lambda API talks to RDS PostgreSQL. During a traffic spike the database starts rejecting connections with "too many clients". Why, and how do you fix it?

The database runs out of connections, because every Lambda environment opens its own connections. An execution environment is one running copy of your function. It handles one request at a time and keeps its own connection or pool.

During a spike, hundreds of requests run at the same time. Lambda then creates hundreds of environments, and each one opens connections. You quickly go over PostgreSQL's max_connections limit. Idle environments also keep their connections open until Lambda removes them.

The main fix is RDS Proxy. It sits between Lambda and the database and keeps a pool of database connections. It shares (multiplexes) many client connections over that small pool. It also:

• handles failover more smoothly
• can use IAM authentication
• can use Secrets Manager credentials

Also:

• Create the database client outside the handler, with a pool size of 1, so each environment reuses one connection.
• Set reserved concurrency on the function. This caps how many environments can exist.
• Consider whether another design fits the workload better: a long-running service on ECS with a normal pool, or DynamoDB.