What Languages Does AWS Lambda Support?

AWS Lambda ships managed runtimes for five language families: Node.js, Python, Java, .NET (which covers C# and PowerShell), and Ruby. Go and Rust are supported as well, but through the OS-only runtime. That means you compile your code to a binary and Lambda runs that binary on Amazon Linux. Any other language, such as PHP, Swift, or C++, runs through a custom runtime that you supply, or inside a container image of up to 10 GB.

A runtime, in Lambda's vocabulary, is the language-specific layer that receives each event, hands it to your function, and returns the result. A managed runtime is one that AWS builds, patches, and keeps current for you. This page lists the options, explains cold starts, and gives a rule for choosing a language. If you came here for the languages Amazon writes its own services in, that question is answered in Which language is Amazon written in?.

The full list, and how each language runs

Version numbers are left out on purpose. AWS adds a managed runtime when a language version reaches long-term support, and retires it when the language community stops patching that version. The exact list therefore changes every few months.

What a cold start is

Lambda does not keep your code running between requests. When a request arrives and no ready copy of your function exists, Lambda creates a new execution environment, which is a small isolated virtual machine. It then downloads your code, starts the language runtime, and runs any setup code outside your handler. The time that takes is the cold start, and it is added to the response time of that one request.

After the first request, Lambda keeps the environment warm for a while and reuses it. The next requests skip the whole sequence. A function that receives steady traffic sees cold starts rarely. A function that receives one request every 20 minutes may see a cold start on every request.

The size of the cold start depends on the language. Python and Node.js start in a fraction of a second, because the interpreter is small and starts fast. Java and .NET load a virtual machine and compile code before the first line runs. For a function with many dependencies, that commonly takes one to several seconds. Go and Rust produce a single compiled binary, so they start about as fast as Python with no interpreter at all.

The two ways to remove cold starts

SnapStart. Lambda runs your function's setup once at deployment and takes a snapshot of the initialized memory. On each cold start it restores from that snapshot instead of starting from nothing. AWS offers SnapStart for Java, Python, and .NET, and it brings a Java cold start from seconds to well under a second. Your setup code must be safe to run once and reuse, so it must not generate a secret that later copies would share.

Provisioned concurrency. You tell Lambda to keep a fixed number of environments, say 10, initialized and waiting at all times. Requests up to that number never see a cold start, and it works for every runtime including custom ones. The cost is that you pay for those 10 environments whether or not they receive traffic.