AWS AppConfig AllAtOnce is for dev and staging; use gradual strategies with alarms in prod

TL;DR: Never deploy configuration changes to production with AllAtOnce; pick a gradual strategy with bake time between steps. Wire a CloudWatch alarm on your error rate into the deployment so AppConfig stops the rollout and reverts to the previous version automatically. Treat a config deploy like a code deploy: gradual, monitored, and reversible, not an instant flip.

Fix

1. Never deploy configuration changes to production with AllAtOnce; pick a gradual strategy with bake time between steps. Expected: this specific failure stops.

2. Production rollouts should use Linear (e.g. Expected: this specific failure stops.

Details

Wire a CloudWatch alarm on your error rate into the deployment so AppConfig stops the rollout and reverts to the previous version automatically. Treat a config deploy like a code deploy: gradual, monitored, and reversible, not an instant flip. AppConfig's AllAtOnce strategy deploys to 100% of targets immediately and is meant for dev and staging only.

When to use

You hit exactly this: AWS AppConfig AllAtOnce is for dev and staging; use gradual strategies with alarms in prod in AWS.

When not to use

A different error, or the same symptom in a different tool. This page only covers the failure above.

Compatibility

AWS.

Edge cases

• Context from the original thread: Architecture notes (dpacgdm/system-design, feature flags and progressive delivery): documents the deployment-strategy gotcha that trips agents pushing config to production.
• 10% more targets each minute), Exponential, or a custom Canary strategy with bake periods, so a CloudWatch alarm on error rate can trigger an automatic rollback before the blast radius grows.

Issue: AWS AppConfig AllAtOnce is for dev and staging; use gradual strategies with alarms in prod Asked: Architecture notes (dpacgdm/system-design, feature flags and progressive delivery): documents the deployment-strategy gotcha that trips agents pushing config to production. AppConfig's AllAtOnce strategy deploys to 100% of targets immediately and is meant for dev and staging only. Production rollouts should use Linear (e.g. 10% more targets each minute), Exponential, or a custom Canary strategy with bake periods, so a CloudWatch alarm on error rate can trigger an automatic rollback before the blast radius grows. Verified resolution: Never deploy configuration changes to production with AllAtOnce; pick a gradual strategy with bake time between steps. Wire a CloudWatch alarm on your error rate into the deployment so AppConfig stops the rollout and reverts to the previous version automatically. Treat a config deploy like a code deploy: gradual, monitored, and reversible, not an instant flip. Source: https://github.com/dpacgdm/system-design/blob/HEAD/Week-07-Specialized-Components/Feature%20Flags%20and%20Progressive%20Delivery.md