Allowing to set scaleDownDelaySeconds for basic replica-weighted canary rollouts

[esulzickis]

We are using basic replica-weighted canary rollout. However, we would like to keep the old version pods running for a while before we terminate them completely. Our rollout strategy is as follows:

  strategy:
    canary:
      analysis: # Run background analysis to check containers restarts count
      ...
      maxSurge: 100%
      maxUnavailable: 25%
      steps:
        - setWeight: 100
        - pause:
            duration: 1m
        - analysis: # Run analysis on the new version of pods to check 5xx error count
           ...

If we catch 5xx errors with this configuration our stable pods will be already terminated. We would like to keep them running for at least 10 minutes more. There is a value called scaleDownDelaySeconds, but it is only possible for traffic routing #4153, which we do not want to use yet.

What is the reason behind allowing to set this value only when traffic routing is enabled? I went through the issues and found only this #478
Would it be possible to enable this setting on basic replica-weighted routing as well?

[kostis-codefresh]

What is the reason behind allowing to set this value only when traffic routing is enabled?

How exactly would this work without a traffic provider?

• You want to keep old pods running
• You don't want to use a traffic provider
• You want your old pods to not get traffic

With the basic canary, traffic is just controlled by pod counts. So it is Kubernetes that handles the routing and not Argo Rollouts.

On a related note, the whole point of using canary deployments is for gradual changes. If you just straight away to setWeight: 100 why not use blue/green?

[esulzickis]

For now we are settingWeight to 100% in order not to slow down the rollout time (it is closer to our setup). We are considering make changes gradually eventually. However, for now it is 100%