# Troubleshoot managed CNI

This page explains common managed CNI problems with Cloud Service Mesh and
how to resolve them. If you need additional assistance, see
[Getting support](https://docs.cloud.google.com/service-mesh/docs/getting-support.md.txt).

## Unsupported managed CNI enabled configuration

Managed Cloud Service Mesh with the `TRAFFIC_DIRECTOR` control plane
implementation requires managed CNI and does not support disabling it. You may
see the `CNI_CONFIG_UNSUPPORTED` code in the feature state message if the
`mesh.cloud.google.com/managed-cni-enabled` label exists but does not have the
value `true` in the control plane revision (CPR) custom resource (CR) or if the
CNI entry in the asm-options configmap exists but does not have the value `on`.

To resolve this error message, you must remove any attempts to disable managed
CNI.

- Case 1: Remove the managed CNI Enabled label in the CPR CR in the cluster.

  ```yaml
  apiVersion: v1
  items:
  - apiVersion: mesh.cloud.google.com/v1beta1
    kind: ControlPlaneRevision
    metadata:
      annotations:
        mesh.cloud.google.com/proxy: '{"managed":"false"}'
      creationTimestamp: "2024-02-18T08:13:30Z"
      generation: 1
      labels:
        app.kubernetes.io/created-by: mesh.googleapis.com
        mesh.cloud.google.com/managed-cni-enabled: false # Remove the "mesh.cloud.google.com/managed-cni-enabled" label
      name: asm-managed
      namespace: istio-system
      resourceVersion: "13422558"
      uid: 3ad755ec-78ab-4d57-8fb9-c5e1a07740d5
  ```
- Case 2: Remove the CNI entry asm-options configmap `ASM_OPTS` data string.

  ```yaml
  apiVersion: v1
  data:
    ASM_OPTS: CNI=off # Remove CNI entry in the ASM_OPTS data.
    multicluster_mode: connected
  kind: ConfigMap
  metadata:
    creationTimestamp: "2024-02-18T08:13:30Z"
    name: asm-options
    namespace: istio-system
    resourceVersion: "1640225"
    uid: 576602da-e60b-4df7-9427-5be06e5bf014
  ```

## CNI Pod unschedulable

You may see this error if the managed CNI Daemonset cannot schedule Pods in any
one of the nodes in the cluster.

Note that in-cluster resources require at least `memory: 100Mi` on each node.
For more information see
[Cloud Service Mesh requirements](https://docs.cloud.google.com/service-mesh/docs/onboarding/provision-control-plane.md.txt#requirements).
If your cluster already has sufficient memory allocated, see
[Pod unschedulable](https://docs.cloud.google.com/kubernetes-engine/docs/troubleshooting.md.txt#PodUnschedulable)
for additional troubleshooting steps.

## Nodes stuck in NotReady status with missing containerID error

You may see this error when your cluster nodes are unable to become ready after
upgrading your Google Kubernetes Engine (GKE) version.

**Error message:**

`container runtime network not ready: NetworkReady=false reason:NetworkPluginNotReady message:Network plugin returns error: missing containerID`

**Cause:**

This issue stems from an incompatibility between the Cloud Service Mesh
Container Network Interface (CNI) version 1.20.x and the [Container Network
Interface (CNI) plugins in
GKE](https://docs.cloud.google.com/kubernetes-engine/docs/learn/learn-gke-net-architecture#how-pod-networking-works).
The CNI plugins require compatibility with the CNI 1.1.0 spec. Clusters
running GKE version `1.35.1-gke.1516000` or later automatically
enable support for this spec. Because Cloud Service Mesh CNI version
1.20.x doesn't support CNI spec version 1.1.0, node health checks fail
and the nodes remain in a `NotReady` status.

**Workaround:**

If your cluster is still running Cloud Service Mesh CNI version 1.20.x, don't
upgrade your GKE node pools to `1.35.1-gke.1516000` or later.

If you have already upgraded and are experiencing this issue, the recommended
workaround is to downgrade your GKE version to a version older
than `1.35.1-gke.1516000` until your cluster's managed CNI is upgraded to
version 1.23.x or later.