Lambda

AWS Lambda runs a function in response to an event. There is no server to provision, no capacity to plan, and no charge when nothing is happening: you supply code and a configuration, and AWS creates and destroys execution environments to match demand.

Functions are best kept small and single-purpose. Anything that needs a long warm-up, holds state between calls, or runs at sustained high volume at a predictable rate is usually better on containers or AWS Batch.

Networking

By default a function runs outside your VPC with internet access. Attach it to a VPC — subnets and security groups — to reach private resources such as an RDS instance or an internal load balancer. A VPC-attached function reaches the internet only through a NAT gateway or a VPC endpoint, which is the usual cause of a function that suddenly cannot call a public API.

Pricing

Charged per request and per GB-second of duration, with the free tier covering one million requests and 400,000 GB-seconds per month. Provisioned concurrency is charged separately and is not covered by the free tier.

Because duration is billed in GB-seconds, more memory is not automatically more expensive: a function that runs twice as fast at double the memory costs the same and returns sooner. Tune memory by measurement, not by assumption.

Configuration

Memory and CPU. 128 MB to 10,240 MB, in 1 MB increments. CPU is allocated in proportion: one vCPU at 1,769 MB, up to roughly six vCPUs at the maximum. A function that is single-threaded stops getting faster past one vCPU.

Execution time. 900 seconds (15 minutes) by default. Functions using AWS Lambda Managed Instances allow up to 5,400 seconds (90 minutes) for asynchronous invocations and for event source mappings, except Amazon MQ and Amazon DocumentDB. For anything longer, or anything that needs a host, use ECS, AWS Batch or EC2.

Runtimes. Managed runtimes for Node.js, Python, Java, .NET and Ruby, each pinned to a major language version and each with a published deprecation date. Go and Rust run on the OS-only runtime. All supported runtimes run on both x86-64 and arm64 (Graviton), and arm64 is usually cheaper. Check the supported runtimes table before pinning a version — a deprecated runtime eventually blocks function creation and then function updates.

Packaging. A .zip archive, optionally with layers, or a container image up to 10 GB.

Triggers

Scheduled events through Amazon EventBridge
HTTP requests through API Gateway, an Application Load Balancer, or a Lambda function URL
Object events in Amazon S3
Messages from Amazon SQS, Amazon SNS or Amazon Kinesis
Streams from Amazon DynamoDB
Many other AWS services, directly or through EventBridge

Monitoring

CloudWatch Logs for output, CloudWatch metrics for invocations, duration, errors, throttles and concurrency, and AWS X-Ray for traces across the services a function calls.

Storage

Each execution environment has a /tmp directory sized between 512 MB and 10,240 MB. It survives between invocations that reuse the same environment, which makes it useful as a cache, and disappears when the environment is reclaimed — so never treat it as durable. For anything larger or shared, use Amazon S3 or mount an Amazon EFS file system (which requires a VPC configuration).

Quotas and limits

Compute and storage

1,000 concurrent executions by default per Region, increasable on request
128 MB – 10,240 MB memory, in 1 MB increments
512 MB – 10,240 MB of /tmp storage
4 KB total for all environment variables
Concurrency scaling rate: 1,000 additional execution environments every 10 seconds, per function

Execution

900 seconds maximum, or 5,400 seconds for Managed Instances functions invoked asynchronously or through an event source mapping (except Amazon MQ and DocumentDB)
625 Mbps network bandwidth per execution environment by default

Deployment and payloads

Zipped .zip package ≤ 50 MB when uploaded through the API or console; use S3 for larger
Unzipped package, including layers, ≤ 250 MB
Container image ≤ 10 GB uncompressed
5 layers per function
Request and response payload ≤ 6 MB each, synchronous
Streamed response ≤ 200 MB, synchronous — uncapped bandwidth for the first 6 MB, then 2 MB/s
Asynchronous invocation payload ≤ 1 MB

New accounts start with reduced concurrency and memory quotas that AWS raises automatically as usage grows. Confirm the numbers that apply to your account in the Service Quotas console rather than assuming the defaults.