Imagine you run the platform team in a company with a few hundred Lambda functions. A security patch lands for your internal SDK, or the company CA certificate changes. With layers, you publish a new layer version and then chase every team to bump the version and redeploy. With 20 functions that is annoying. With 400 it turns into a project with its own spreadsheet.
Since November 18, 2026, AWS Lambda offers a different model: customer-managed runtimes. I took a closer look and built a complete example with the Serverless Framework and AWS CDK in my article Lambda customer-managed runtimes: one runtime for all. Here is the short version.
The idea: separate the runtime from the function
A Lambda runtime is everything in the execution environment before your code arrives: OS libraries, the language (for example Node.js 24) and the runtime interface client that passes events to your handler. For nodejs24.x, AWS maintains all of that. With customer-managed runtimes, a team inside your company takes over that job.
The model has three building blocks:
- A runtime with a name and an ARN, for example
radwerk-node24 - Immutable runtime versions, each pointing to exactly one image in Amazon ECR
-
Runtime aliases such as
dev,testandprodthat point to a version
Function teams never reference an image directly. They set the runtime to custom, pick an alias and keep deploying ordinary .zip files. The publisher builds images, creates versions and decides when an alias moves on.
Why this is different from layers and container images
The key difference is who triggers an update. With a layer or a container image, every function needs a redeploy. With a customer-managed runtime, moving the alias is enough: functions pick up the new version at their next cold start. That also means old and new versions run side by side for a while, and there is no percentage-based traffic shifting. You control the pace through the order of your aliases and Regions.
Consumers still get a say via the runtime update mode: Auto (resolve the alias at every cold start), Function update (only on deploys), or Manual (pin a version, useful for a short freeze or rollback).
To check what a function is actually running on, two commands help:
# Which version does the prod alias point to?
aws lambda-core get-runtime-alias \
--runtime-name radwerk-node24 --name prod
# Which runtime version did a function resolve?
aws lambda get-function-configuration \
--function-name rental-dev-rentBike \
--query RuntimeVersionConfig
On top of that, the INIT_START log line contains the runtime version, so a Logs Insights query shows how far a rollout has progressed.
Where it gets serious
For me, the most important point of the whole feature is security. Whoever publishes a runtime can place code into the execution environment of every function that uses it, and that code runs with each function's execution role. A compromised publisher account becomes a lever on a large number of functions at once. I would treat the runtime pipeline like production infrastructure: a separate account, four-eyes approvals, image scans and immutable ECR tags. Organizations that are not ready can block the feature with an SCP.
A few other things worth knowing up front:
- If a publisher deletes an alias or version that is still in use, functions go
Inactiveand cold starts fail. Check theRuntimeInvocationsmetric before deleting anything. - If a team bundles the shared module itself, its own copy wins and runtime updates silently stop taking effect.
- SnapStart, Lambda@Edge and container image functions are not supported.
- Runtimes, versions and aliases cost nothing extra; you pay for ECR storage.
My take: this is one of the most useful Lambda features of the year, but mainly for organizations with many teams and a real platform team. It also shifts responsibility, so it only works with clear agreements like changelogs per version and fixed rollout windows for prod.
What the full article covers
- Building a Node.js 24 runtime image with a shared logger, plus the ECR policy Lambda needs
- Creating runtime, versions and aliases with
aws lambda-coreand sharing them across your AWS Organization - Wiring a function to the runtime with the Serverless Framework and with AWS CDK
- Scripts for staged rollouts, promotion and rollback, and how to monitor them
- Common mistakes, quotas and limits in detail
Read the complete guide here: Lambda customer-managed runtimes: one runtime for all.
๐ This is the condensed version. The full article on sascha-ms.de goes deeper โ with all code samples, diagrams and a short knowledge check.
๐ Next step: Curious what your own Lambda functions cost? Try my Lambda cost calculator โ it runs right in your browser.
๐ฌ Stay updated: New articles on serverless, AI and leading tech teams via RSS, browser push or the weekly newsletter (in German).
Originally published on sascha-ms.de. More articles on serverless, AI in everyday development and leading tech teams: sascha-ms.de/en
Top comments (0)