For canary deployment analysis within your Spinnaker pipeline, how should you configure the comparison?

In canary deployment analysis, the proper configuration for comparison is to compare the canary with the existing deployment of the current production version. This approach ensures that you can accurately assess how the new changes in the canary version perform in relation to what users are currently experiencing. By evaluating the canary against the production version being served to users, you can directly identify potential issues or improvements that the new version brings compared to the stable version.

This method is essential as it provides an immediate baseline, which reflects the actual performance metrics that users are experiencing in real-time. By comparing the canary deployment to this current production version, you can determine whether the new changes introduce regressions in performance, availability, or other critical metrics before fully rolling out the new version.

The other options do not provide a clear baseline for accurate comparison. For example, comparing to a previous production version may not reflect recent performance realities, and relying on a sliding window average can obscure individual version performance anomalies which can be crucial for decision-making in a canary release scenario.