I'm trying to cover all of our pipeline requirements with ArgoCD Rollout and stuck on the next one:
we want Experiment service canary and baseline configurations to be identical to canary and stable service configurations as much as possible. To be sure in releasing process a bit more.
We use Istio as a traffic management system and Istio is a big part of every service.
For now, ArgoCD Rollout only controls labels and subsets in VS and DR. It's ok because you can manually provide required settings to stable/canary in VS or DR and Argo doesn't modify them.
But it's ok until you wanna have experiments:
In that case, Argo just adds labels, subsets and nothing more for experiment replica sets:
So, when you run Kayenta analysis - you don't cover network configuration at all and you can get unexpected results from the deployment process.
It would be really great to cover network configuration as well (actually, we can't proceed without it)
As a total newbie with Argo Rollouts, I see only two options:
Add switcher to copy or not VS and DR settings (except controlled by Argo) for baseline (from stable) and canary (from canary) experiment replica sets
Add a switcher to preserve or not (current behavior is creating new hashes) rollouts-pod-hash-template for baseline and canary
TBD
Use Cases
Relevant canary analysis results during rollout to the new version with Experiment step and Istio as a Traffic management
Probably, it could be reasonable for other Traffic systems as well
Thanks in advance for any ideas/feedback!
Message from the maintainers:
Impacted by this bug? Give it a 👍. We prioritize the issues with the most 👍.
Summary
I'm trying to cover all of our pipeline requirements with ArgoCD Rollout and stuck on the next one: we want Experiment service canary and baseline configurations to be identical to canary and stable service configurations as much as possible. To be sure in releasing process a bit more.
We use Istio as a traffic management system and Istio is a big part of every service. For now, ArgoCD Rollout only controls labels and subsets in VS and DR. It's ok because you can manually provide required settings to stable/canary in VS or DR and Argo doesn't modify them.
But it's ok until you wanna have experiments: In that case, Argo just adds labels, subsets and nothing more for experiment replica sets:
So, when you run Kayenta analysis - you don't cover network configuration at all and you can get unexpected results from the deployment process. It would be really great to cover network configuration as well (actually, we can't proceed without it)
As a total newbie with Argo Rollouts, I see only two options:
Use Cases
Relevant canary analysis results during rollout to the new version with Experiment step and Istio as a Traffic management Probably, it could be reasonable for other Traffic systems as well
Thanks in advance for any ideas/feedback!
Message from the maintainers:
Impacted by this bug? Give it a 👍. We prioritize the issues with the most 👍.