Istio Subset- Level traffic splitting cannot be used for Rolling Update

[gyz2009]

Summary

Update process does not Scale down the old RS, Its renewal process is more like blue-green.
It waits for all new ReplicaSet pods to started and scale down the old ReplicaSet Pods to 0.

strategy:
  canary:
    # maxSurge: "25%"
    # maxUnavailable: "25%"
    scaleDownDelaySeconds: 30
    trafficRouting:
      istio:
        virtualService:
          name: helloworld-vsvc # required
          routes:
            - primary # optional if there is a single route in VirtualService, required otherwise
        destinationRule:
          name: helloworld-destrule # required
          canarySubsetName: canary # required
          stableSubsetName: stable # required
    steps:
      - setWeight: 20
      - pause: {duration: 20s}
      - setWeight: 40
      - pause: {duration: 20s}
      - setWeight: 60
      - pause: {duration: 20s}
      - setWeight: 80
      - pause: {duration: 20s}

Diagnostics

What version of Argo Rollouts are you running?
1.0.4

---

Message from the maintainers:

Impacted by this bug? Give it a 👍. We prioritize the issues with the most 👍.

[huikang]

Hi, @gyz2009 , could you paste the rollout status in the middle of an update, e.g., after step 5 (setWeight: 60)?

[huikang]

Never mind; I can reproduce this bug. Will look into this. Thanks.

NAME                                                         KIND         STATUS        AGE   INFO
⟳ rollouts-demo-istio                                        Rollout      ॥ Paused      2m4s  
├──# revision:2                                                                               
│  ├──⧉ rollouts-demo-istio-6b5cbf7465                       ReplicaSet   ✔ Healthy     107s  canary
│  │  ├──□ rollouts-demo-istio-6b5cbf7465-5djdj              Pod          ✔ Running     107s  ready:1/1
│  │  └──□ rollouts-demo-istio-6b5cbf7465-9blgw              Pod          ✔ Running     99s   ready:1/1
│  └──α rollouts-demo-istio-6b5cbf7465-2-1                   AnalysisRun  ✔ Successful  106s  ✔ 1
│     └──⊞ d5c11d15-7bfa-4555-93c1-72c13bc858c2.sleep-job.1  Job          ✔ Successful  106s  
└──# revision:1                                                                               
   └──⧉ rollouts-demo-istio-6668595956                       ReplicaSet   ✔ Healthy     2m4s  stable
      ├──□ rollouts-demo-istio-6668595956-jd7s7              Pod          ✔ Running     2m4s  ready:1/1
      ├──□ rollouts-demo-istio-6668595956-mdsq7              Pod          ✔ Running     2m4s  ready:1/1
      ├──□ rollouts-demo-istio-6668595956-sb8ql              Pod          ✔ Running     2m4s  ready:1/1
      ├──□ rollouts-demo-istio-6668595956-w4fqc              Pod          ✔ Running     2m4s  ready:1/1
      └──□ rollouts-demo-istio-6668595956-wgrxw              Pod          ✔ Running     2m4s  ready:1/1

[huikang]

@jessesuen , checking the following

argo-rollouts/utils/replicaset/canary.go

Lines 110 to 112 in cf19b5b

	if rollout.Spec.Strategy.Canary.TrafficRouting != nil {
	return desiredNewRSReplicaCount, rolloutSpecReplica
	}

which seems cause the oldRs or ohterRS not to be scaled down. Is this the desired behavior for traffic routing?

[harikrongali]

@huikang https://github.com/argoproj/argo-rollouts/pull/341/files#r363676969

[huikang]

@harikrongali , thanks so much for pointing me to the change. So this is the expected behavior.

@gyz2009 , I think we can close this one per design.

[gyz2009]

@huikang Ok, I have some RS Replicas 100+, due to insufficient resources, update will be in pending state.

[harikrongali]

@gyz2009 #1029 solve what you are seeing and should be available in 1.1 release

[jessesuen]

Duplicate of #1029

[gyz2009]

@harikrongali I tested 1.1.0-RC1 and there was no change

strategy:
  canary: 
    dynamicStableScale: true
    # maxSurge: "25%"
    # maxUnavailable: "25%"
    # scaleDownDelaySeconds: 30
    trafficRouting:
      istio:
        virtualService:
          name: helloworld-vsvc
          routes:
            - primary
        destinationRule:
          name: helloworld 
          canarySubsetName: canary # required
          stableSubsetName: stable # required
    steps:
      - setCanaryScale:
          matchTrafficWeight: true
      - setCanaryScale:
          weight: 30
      - pause: {duration: 60s}
      - setCanaryScale:
          weight: 60
      - setCanaryScale:
          weight: 100