Lambda 30-minute timeout: what changes for long jobs
AWS doubled the on-demand Lambda timeout to 30 minutes on November 19, 2026. What changed, who should care, what it costs and how to try it safely this week.
On November 19, 2026, AWS announced a 30-minute timeout for on-demand Lambda Event Functions, doubling the previous 15-minute limit. It applies to synchronous, asynchronous and event source mapping invocations, in all commercial AWS Regions. If you run long jobs on Lambda, such as document processing, data exports or an agent that calls a model many times, you may be able to drop a workaround. If your functions finish in seconds, you can ignore it.
What was announced, and what is different?
According to the announcement, three things changed or were confirmed:
- On-demand Event Functions: the maximum timeout is now 30 minutes, up from 15. The timeout documentation says you can set it in 1-second steps up to 1,800 seconds. The default is still 3 seconds.
- Lambda Managed Instances: synchronous invocations also support up to 30 minutes. Asynchronous and event source mapping invocations on Managed Instances keep the 90-minute timeout AWS launched in September 2026.
- Compatibility: AWS says the longer timeout works with Lambda durable functions and with Step Functions Standard workflows.
Nothing else is announced as changed. The timeout is a ceiling, not a default: every function keeps the value you set.
Source: AWS What's New, November 19, 2026, for on-demand Event Functions.
Who should care, and who can ignore it?
You should care if you:
- Run jobs that sit close to 15 minutes today, and split them into chunks or move them to containers only because of that limit.
- Run agent tasks with many model and tool calls in one invocation.
- Process large files or batches from a queue, where one slow item can time out the whole batch.
You can ignore it if your functions already finish well inside the old limit, or if your work is better suited to a container or a workflow engine. A 30-minute function is still a function: it has no built-in progress saving. For long work with waits and retries, AWS's durable functions checkpoint progress, and that remains a separate design choice.
What does the caller do while your function runs?
This is where most surprises live. When another service invokes your function synchronously, AWS documents that it may stop waiting before the function finishes, and the function keeps running and keeps billing until it returns or times out. As of November 2026, the documented waits include:
| Caller | How long it waits |
|---|---|
| API Gateway REST APIs | 29 seconds by default |
| API Gateway HTTP APIs | 30 seconds maximum |
| Amazon Cognito user pool triggers | 5 seconds |
| Step Functions | Until the function returns |
| Lambda function URLs | Until the function returns |
So a 30-minute timeout does not make a 30-minute HTTP request through API Gateway possible. If a user clicks a button and waits, the answer is still to start the job asynchronously and return an identifier. The longer limit helps the worker behind that pattern.
For your own SDK or CLI clients, the synchronous invocation guide says synchronous invocations can run up to 30 minutes and the client may be disconnected while it waits. It advises a client read timeout higher than the function timeout, and TCP keep-alive turned on.
- Your userWaits for an answer
- CallerAPI Gateway stops at 29 to 30 seconds
- Lambda functionKeeps running and billing up to its timeout
Based on the caller limits in the Lambda documentation, as of November 2026.
What does it cost?
AWS states that "standard AWS Lambda pricing applies". The Lambda pricing page lists, for US East (N. Virginia) as of November 2026, $0.0000166667 per GB-second on x86 and $0.20 per million requests.
A quick, illustrative calculation (not a benchmark): a function with 1 GB of memory that runs the full 1,800 seconds uses 1,800 GB-seconds. At the x86 rate, that is about $0.03 per invocation. That is small for one run, but it matters for a bug. A job that hangs on a slow dependency can burn twice as much before it is stopped if you raise its timeout from 15 to 30 minutes. A retry storm multiplies it. This is the real cost risk: not the price, but the time a stuck invocation is allowed to run.
How would a team try it safely this week?
- Pick one function that already fails or is split because of the 15-minute limit. Do not raise timeouts across the account.
- Raise queue visibility first. If the function reads from SQS, AWS says to set the queue's visibility timeout to at least six times the function timeout, and that Lambda does not check this when you change the function timeout. If the function timeout goes above the visibility timeout, messages can become visible again while still being processed, so they may run twice.
- Then raise the timeout on a staging copy of the function, with
aws lambda update-function-configuration --function-name <name> --timeout 1800, or theTimeoutproperty in SAM or CloudFormation. The timeout is a setting of the function itself, so test it on a separate copy, not on production. - Make the work idempotent. A longer run means a longer window for a retry or duplicate to overlap with the first attempt.
- Add a deadline inside your code. Have the code track its own elapsed time and stop cleanly, saving progress, a minute or two before the timeout.
- Alarm on duration. Alert when p99 duration passes, say, 80 percent of the timeout (our rule of thumb), so you see drift before a failure.
What would we check first before adopting it?
We would start with four questions:
- Is the job really a Lambda job? If it needs progress saving, long waits or more than 30 minutes, a durable function, Step Functions or a container may fit better.
- Who is waiting? Trace each caller and its own timeout, using the table above.
- What happens on a retry? Check how your asynchronous invocations retry and where failures go, so one stuck job does not repeat silently.
- What is the worst case? Multiply memory, the full timeout and the concurrency you allow, and decide if a budget alarm is needed.
If you want a second pair of senior eyes on a long-running workload, our services page describes how we work.
This article is general information, not financial or architecture advice. Check current AWS documentation and pricing for your Region before you decide.