Lambda runs your code in response to events without you managing servers. You supply a function, AWS supplies the compute, and you pay for the time it runs rather than for an instance sitting idle.\n\nThe model is genuinely different from a server, and most Lambda problems come from carrying server assumptions into it.\n\n\u003cAwsDiagram type="lambda-flow" /\u003e\n\n## The execution model\n\nAn execution environment is a small, isolated sandbox that runs one invocation at a time. Ten simultaneous requests mean ten environments. There is no in-process concurrency to share.\n\nWhat matters operationally is that AWS reuses environments. After an invocation finishes, the environment is kept warm for a while and reused for the next request. That has two consequences:\n\nCode outside the handler runs once per environment, not once per invocation. This is the single most useful optimisation available:\n\n```python\nimport boto3\n\n# Runs once per environment. Reused across invocations\ns3 = boto3.client("s3")\n\ndef handler(event, context):\n # Runs every invocation\n return s3.get_object(Bucket="acme", Key=event["key\
Design VPC subnets properly: what makes a subnet public, CIDR sizing and the five reserved IPs, one subnet per AZ, required tags and running out of addresses.
How VPC route tables route traffic: the local route, longest prefix match, the main route table trap, one table per AZ and debugging no connectivity.
See what an AWS NAT gateway does, why it is often the largest networking cost on your bill, and how VPC endpoints and design changes reduce it.