cni-metrics-helper

The cni-metrics-helper is a tool that can be used to scrape ENI and IP information, aggregate it on a cluster level and publish the metrics to CloudWatch. The following IAM permission is required by the worker nodes to publish metrics:

"cloudwatch:PutMetricData"

By default, IPAM will publish prometheus metrics on :61678/metrics.

The following diagram shows how cni-metrics-helper works in a cluster:

As you can see in the diagram, the cni-metrics-helper connects to the API Server over https (tcp/443), and another connection is created from the API Server to the worker node over http (tcp/61678). If you deploy Amazon EKS with the recommended security groups from Restricting cluster traffic, then make sure that a security group is in place that allows the inbound connection from the API Server to the worker nodes over tcp/61678.

Adding the CNI metrics helper will publish the following metrics to CloudWatch:

Metric	Description	Statistic¹
addReqCount	The number of CNI ADD requests that require an IP address	Sum
assignIPAddresses	The number of IP addresses assigned to pods	Sum
awsAPIErr	The number of times AWS API returns an error	Sum
awsAPILatency	AWS API call latency in ms	Max
awsUtilErr	The number of errors not handled in awsutils library	Sum
delReqCount	The number of CNI DEL requests	Sum
eniAllocated	The number of ENIs allocated	Sum
eniMaxAvailable	The maximum number of ENIs that can be attached to this instance, accounting for unmanaged ENIs	Sum
ipamdActionInProgress	The number of ipamd actions in progress	Sum
ipamdErr	The number of errors encountered in ipamd	Sum
maxIPAddresses	The maximum number of IP addresses that can be allocated to the instance	Sum
podENIErr	The number of errors encountered while managing ENIs for pods	Sum
reconcileCount	The number of times ipamd reconciles on ENIs and IP/Prefix addresses	Sum
totalIPAddresses	The number of IPs allocated for pods	Sum
totalIPv4Prefixes	The total number of IPv4 prefixes	Sum
totalAssignedIPv4sPerCidr	The total number of IP addresses assigned per cidr	Sum
forceRemoveENI	The number of ENIs force removed while they had assigned pods	Sum
forceRemoveIPs	The number of IPs force removed while they had assigned pods	Sum
ec2ApiReqCount	The number of requests made to EC2 APIs by CNI	Sum
ec2ApiErrCount	The number of failed EC2 API requests	Sum

Using IRSA

As per AWS EKS Security Best Practice, if you are using IRSA for pods then following requirements must be satisfied to succesfully publish metrics to CloudWatch

The IAM Role for your SA (IRSA) must have following policy attached

{
    "Version": "2012-10-17",
    "Statement": [
        {
            "Effect": "Allow",
            "Action": [
                "cloudwatch:PutMetricData"
            ],
            "Resource": "*"
        }
    ]
}

Specify the IRSA name in the cni-metrics-helper deployment spec alongwith the AWS_CLUSTER_ID (as described below). The value that you specify here will show up under the dimension 'CLUSTER_ID' for your published metrics. Specifying a value for this field is mandatory only if you are blocking IMDS access.

`AWS_CLUSTER_ID`

Type: String

Default: ""

An identifier for your Cluster which will be used as the dimension for published metrics. Ideally it should be ClusterName or ClusterID.

kind: Deployment
apiVersion: apps/v1
metadata:
  name: cni-metrics-helper
  namespace: kube-system
  labels:
    k8s-app: cni-metrics-helper
spec:
  selector:
    matchLabels:
      k8s-app: cni-metrics-helper
  template:
    metadata:
      labels:
        k8s-app: cni-metrics-helper
    spec:
      containers:
      - env:
        - name: AWS_CLUSTER_ID
          value: ""  
        - name: USE_CLOUDWATCH
          value: "true"
        name: cni-metrics-helper
        image: <image>
      serviceAccountName: <IRSA name>

With IRSA, the above deployment spec will be auto-injected with AWS_REGION parameter, and it will be used to fetch region information when we publish metrics. Possible scenarios for above configuration:

If you are not using IRSA, then Region and CLUSTER_ID information will be fetched using IMDS (should have access)
If you are using IRSA but have not specified AWS_CLUSTER_ID, we will fetch the value for CLUSTER_ID if IMDS access is not blocked
If you have blocked IMDS access, then you must specify a value for AWS_CLUSTER_ID in the deployment spec
If you have not blocked IMDS access but have specified AWS_CLUSTER_ID value, then this value will be used.

Installing the cni-metrics-helper

To install the CNI metrics helper, follow the installation instructions from the target version release notes.

Creating a metrics dashboard

After you have deployed the CNI metrics helper, you can view the CNI metrics in the Amazon CloudWatch console.

To create a CNI metrics dashboard

Open the CloudWatch console at https://console.aws.amazon.com/cloudwatch/.
In the left navigation pane, choose Metrics and then select All metrics.
Choose the Graphed metrics tab.
Choose Add metrics using browse or query.
Make sure that under Metrics, you've selected the AWS Region for your cluster.
In the Search box, enter Kubernetes and then press Enter.
Select the metrics that you want to add to the dashboard.
At the upper right of the console, select Actions, and then Add to dashboard.
In the Select a dashboard section, choose Create new, enter a name for your dashboard, such as EKS-CNI-metrics, and then choose Create.
In the Widget type section, select Number.
In the Customize widget title section, enter a logical name for your dashboard title, such as EKS CNI metrics.
Choose Add to dashboard to finish. Now your CNI metrics are added to a dashboard that you can monitor. For more information about Amazon CloudWatch Logs metrics, see Using Amazon CloudWatch metrics in the Amazon CloudWatch User Guide.

EKS CNI metrics

Get cni-metrics-helper logs

kubectl get pod -n kube-system
NAME                                  READY     STATUS    RESTARTS   AGE
aws-node-248ns                        1/1       Running   0          6h
aws-node-257bn                        1/1       Running   0          2h
...
cni-metrics-helper-6dcff5ddf4-v5l6d   1/1       Running   0          7h
kube-dns-75fddcb66f-48tzn             3/3       Running   0          1d

kubectl logs cni-metrics-helper-6dcff5ddf4-v5l6d -n kube-system