Amazon Web Services announced on November 17, 2026 that AWS Lambda durable functions now support pausing a running execution and resuming it later from its last checkpoint. The same update lets teams delete the stored data of a finished execution before its retention period runs out.
Durable functions are Lambda's model for long, multi-step workflows that can run for up to a year with automatic checkpoints and retries. Until now, an operator facing a broken dependency had few choices beyond letting retries burn out or stopping the run for good. The new controls give developers a middle option and a cleaner way to handle sensitive payloads.
What AWS Added
The launch adds three operator actions that work on one execution at a time, identified by its execution ARN. None of them requires a code change in the function itself.
- Pause: stops Lambda from invoking the function for that execution, while keeping all progress saved so far.
- Resume: picks the execution back up from its most recent checkpoint, on the same function version it started with.
- Delete: permanently removes a finished execution's history, checkpoints and input and output payloads.
All three are available from the Lambda console, the AWS CLI and the AWS SDKs. AWS says they work in every Region where durable functions are offered.
How Pausing Works
According to the Lambda Developer Guide, the result of a pause depends on whether your code is running at that moment. If the execution is idle, waiting on a timer, a callback or a retry delay, its status moves straight from RUNNING to PAUSED.
If an invocation is in flight, the status first becomes PAUSING. The code carries on until its next checkpoint is written, for example when a step finishes. Lambda then ends the invocation and marks the run as PAUSED.
Pausing does not freeze everything around the execution. Wait timers still count down, external systems can still complete callbacks, and child executions keep running. Lambda records those events and hands them to the function in its first invocation after the resume.
Out of the box, the overall timeout is also frozen while a run is paused, and Lambda adds the paused time back when it resumes. Teams can override that with a CLI option so the timeout keeps ticking. Either way, the guide says no execution can live beyond 366 days from its start.
Deleting Execution Data
Deletion is allowed only for finished runs, meaning ones that succeeded, failed, timed out or were stopped. A running or paused run has to be stopped first. Once deleted, the execution's details and result cannot be recovered, and the same execution name can be used again for a fresh run.
The docs point out what deletion leaves behind. CloudWatch logs and metrics, CloudTrail records, child executions and any data the function wrote to other services all remain under their own retention rules. Deleting early also does not cut storage charges, because AWS bills retained data for the full retention period.
What It Means for Developers
The pause feature is aimed at incidents. When a payment API is down or an AI model provider starts rate limiting, you can hold affected workflows instead of letting them exhaust retries or write half-finished data. It also helps during risky deployments of the systems a workflow calls.
Keep in mind that a pause covers existing executions only. New invocations still start new runs, so you may also need to disable event source mappings or stop the callers. A paused run still counts against the running-executions quota, though it incurs no invocation charges while it waits.
The delete action is mostly a compliance tool. If a workflow input contained personal data, you can now purge it on request instead of waiting for retention to expire. For lower storage bills, AWS recommends setting a shorter retention period, which applies to executions started after the change.
Teams already building serverless APIs on AWS may also want to look at the recent API Gateway JWT authorizer support for REST APIs, which removes another piece of Lambda glue code.
How to Get Started
The new API calls are PauseDurableExecution, ResumeDurableExecution and DeleteDurableExecution. Each needs a matching IAM permission on the execution ARN, and functions that use a customer managed KMS key also need decrypt rights for pause and resume.
- Find the runs to act on with ListDurableExecutionsByFunction, which can filter by status and start time.
- To pause a run while an invocation is active, update to a supported Durable Execution SDK: JavaScript 2.5.0, Python 2.1.0, Java 2.3.0, Go 1.0.0 or .NET 2.2.0 and later.
- Check the outcome with GetDurableExecution, since the status changes after the API call returns.
- Grant the delete permission only to people or roles allowed to remove data for good, and audit those calls with CloudTrail.
Conclusion
Pause, resume and delete give Lambda durable functions the kind of operator controls that long-running workflow engines usually offer. The features need no code changes, only newer SDKs and the right IAM permissions. Teams running multi-day or AI-driven workflows on Lambda should add these actions to their incident runbooks.
Sources
- AWS What's New: AWS Lambda durable functions now supports pausing and deleting executions
- AWS Lambda Developer Guide: Pausing and resuming durable executions
- AWS Lambda Developer Guide: Deleting durable executions
See my articles more often when you search Google.
