How Do I Handle a Canary Upgrade Failure?¶

There are many reasons for a canary upgrade failure. In case of a canary upgrade failure, you can use the following solutions to handle it.

Failed to check custom resource definitions (CRDs) before the upgrade.
Solution: New Istio version does not support some CRDs, including ClusterRbacConfigs, ServiceRoles, ServiceRoleBindings, and Policies. If there are resources to be discarded in the current version, delete them before the upgrade.
Failed to check Istio gateway labels before the upgrade.
Solution: Configure Istio gateway labels (specified by matchLabels) in {app: istio-ingressgateway, istio: ingressgateway} format.
Failed to check add-ons before the upgrade.
Solution: ASM 1.8 and later versions do not support the tracing, kiali, grafana, and prometheus add-ons. Uninstall the add-ons before the upgrade. You can install open-source add-ons or use APM.
Failed to check the cluster status before the upgrade.
Solution: If the cluster is unavailable before the upgrade, do not perform the upgrade.
Failed to query resources before the upgrade.
Solution: Prepare the required resources for the canary upgrade.
Failed to check the cluster version before the upgrade.
Solution: Use the cluster version listed in the following table.
Service Mesh Version
Supported Cluster Version
1.15
1.21, 1.23, 1.25, and 1.27
1.18
1.25,1.27, and 1.28
Failed to check the component affinity before the upgrade.
Solutions: If you upgrade Istio from a non-canary version to a canary version, ensure that there are at least twice as many nodes labeled with istio:master as there are istiod instances, and at least twice as many schedulable nodes as there are istio-ingressgateway or istio-egressgateway instances (depending on which one is larger). If such conditions are not met, add nodes to meet the scheduling requirements or set the anti-affinity of istiod, istio-ingressgateway, and istio-egressgateway to Preferred.
- Method 1: Add nodes labeled with istio:master on the CCE console.
- Solution 2: Edit the YAML file to modify the anti-affinity policy on the CCE console.
```
preferredDuringSchedulingIgnoredDuringExecution:
  - weight: 1
    podAffinityTerm:
      labelSelector:
        matchExpressions:
            - key: app
              operator: In
              values:
                - istiod (For the anti-affinity policy of istio-ingressgateway, replace the value of values with istio-ingressgateway. For the anti-affinity policy of istio-egressgateway, replace the value of values with istio-egressgateway.)
       namespaces:
         - istio-system
       topologyKey: kubernetes.io/hostname
```
Alternatively, change the anti-affinity from Required to Preferred on the CCE console.
Failed to check the automatic namespace injection before the upgrade.
Solution: If there are pods in the namespace when you migrate service mesh data from the Dedicated edition to the Basic edition, enable automatic injection for the namespace.

last updated: 2024-11-13 13:47 UTC - commit: 02ebb7bc547caa1c49ad4bb55e9e92979aa48241

Service Mesh Version	Supported Cluster Version
1.15	1.21, 1.23, 1.25, and 1.27
1.18	1.25,1.27, and 1.28