Canary Release

You might say

Send 5% of users to the new release first; expand only while error rate stays healthy, and return to the stable release on anomaly.

A canary release exposes a new version to a small amount of real traffic before health signals decide whether to expand or stop.Checkout v2 can serve 5% of requests while stable v1 keeps the other 95%; if v2 crosses its error threshold, its traffic returns to v1. This limits exposure to a release version. It is not merely switching one feature on or off inside the same release.

Know first

Deployment Monitoring

Canary ReleaseCanary Deployment

How a small amount of production traffic reaches v2 first

PRODUCTION TRAFFICcheckout-api

v1 stable

User requests100%

STABLEv1.8.4100%

Checkout errors0.2%

CANDIDATEv2.0.00%

Checkout errorsNo traffic

v1 serves every request; v2 is ready but has no real users yet.

What to inspect before expanding

Limit possible impact to a small scope: An all-at-once checkout release exposes every request to unknown defects. Giving v2 5% of traffic while verified v1 keeps the other requests tests the new release under real conditions without immediately widening impact.

Let agreed signals decide the next step: Define an observation period, error or latency threshold, and return action before release. Expand only while signals are healthy; stop and restore the stable version above the threshold. A quiet small sample still cannot prove every path is correct.

You can say this to an AI Agent

Canary checkout v2 at 5% traffic and observe checkout error rate and latency for 15 minutes. Expand only while error rate stays below 2%; above the threshold, stop and return traffic to v1. Report actual traffic and metrics for both versions.

Learn next

1Monitoring→2Rollback→3Blue-Green Deployment→

Recommended tool

Web fine-tuning tool · Selector Pro

Select page elements for precise AI edits

Click a button, text, or card, note the change, then copy its exact location to AI.

Open Selector Pro website

Checkout v2 is serving 5% of traffic, but its error rate reaches 7%, above the 2% threshold. What should happen next?

A. Stop expansion, return v2 traffic to v1, and confirm error rate and checkout success recoverB. Expand to 100% anyway and wait for a larger sampleC. Keep 5% traffic indefinitely because the deployment job succeeded