Progressive delivery lets you ship to 5% of users first and roll back in 30 seconds if something breaks — here is how to implement canary deployments with Argo Rollouts and Flagger on Kubernetes.
Add a minRequestCount condition to your AnalysisTemplate so the metric evaluation is skipped if traffic volume is below a threshold, preventing false rollbacks from sparse data. Use a longer analysis interval during off-peak windows and pair it with a maximumEligibleWeight cap to avoid promoting a canary to high traffic percentages during low-signal periods.
Nginx canary weight annotations are eventually consistent — the ingress controller reconciles them asynchronously. During a rapid Flagger rollback, there is a window where the annotation is updated but Nginx has not yet reloaded. Mitigate by increasing Flagger's rollback timeout and setting nginx.ingress.kubernetes.io/proxy-next-upstream to handle upstream errors gracefully during the transition window.
Discussion0