# Versions, Aliases and Safe Rollouts — Serverless & Edge

Source: https://www.skillbyai.com/en/serverless-edge/se-safe-deploys

> Use immutable versions, aliases and weighted traffic to ship canary releases and roll back in seconds.

## Move a pointer, not the code

Deploying a function replaces code that thousands of requests a minute may be hitting, so platforms support safe rollouts. On Lambda, publishing creates an immutable, numbered **version**. An **alias** such as `live` is a named pointer to a version, and triggers call the alias. Rolling forward or back is then just moving the pointer. An alias can also split traffic by weight between two versions, which enables a **canary**: send 10% of traffic to the new version, watch error metrics, then shift the rest, or roll back automatically if an alarm fires. AWS SAM can drive this through CodeDeploy using `AutoPublishAlias` and a `DeploymentPreference`. Edge platforms offer comparable features, such as gradual deployments and instant rollback to an earlier deployment. Two rules keep rollbacks safe: keep database schema changes backward compatible for at least one release, and send every deploy through the pipeline, never through console edits.

## A canary rollout in SAM

Each deploy publishes a new version, sends 10% of traffic to it for five minutes, then shifts the rest, unless the referenced alarm fires first and triggers a rollback.

```yaml
  OrderFunction:
    Type: AWS::Serverless::Function
    Properties:
      Runtime: nodejs22.x
      Handler: index.handler
      CodeUri: src/
      AutoPublishAlias: live
      DeploymentPreference:
        Type: Canary10Percent5Minutes
        Alarms:
          - !Ref OrderErrorsAlarm
```

## Trying a new bus route on one depot first

A city transport department testing a new timetable runs it from one depot for a week before rolling it out citywide. If complaints spike, only a few routes revert. A canary deploy does the same with a slice of your traffic.

## Rollback breaks if the schema moved on

If version 2 renames a column and you roll back to version 1, version 1 crashes on the new schema. Use expand-and-contract migrations: add the new column, deploy code that writes both, migrate, and only drop the old column in a later release.

**Quiz:** In Lambda, what is an alias?

- [ ] A copy of the function deployed to another region
- [ ] An immutable snapshot of the function code
- [ ] An environment variable that stores the function name
- [x] A named pointer to a version that can also split traffic between two versions

*Answer:* A named pointer to a version that can also split traffic between two versions. Versions are the immutable snapshots; an alias points at one (or weights two), so triggers need not change during a rollout.
