cloudzero-cloudwatch-metrics

CloudZero Container Metrics

A helm chart for collecting performance metrics down to the pod level. CloudZero uses this data to calculate container cost to the pod, label, namespace, or Cluster level. See our documentation here.

Chart Installation

Add the CloudZero repository to Helm:

helm repo add cloudzero https://cloudzero.github.io/cloudzero-k8s-charts

Prerequisite

The agent must have permission to create and write to a CloudWatch LogGroup and LogStream. Adding the AWS Managed policy arn:aws:iam::aws:policy/CloudWatchAgentServerPolicy to the cluster nodes allows this. Review how you manage your nodes to see how best to attach this managed policy to the node role before installation.

Considerations Before Installing

The agent consumes cpu and memory in relation to your cluster density. The density is defined by counting resources like number of nodes, pods, endpoints, and replicasets. The chart values defined for CPU and Memory limits/requests are suitable for a cluster density of 300 nodes, 5000 pods, and 70,000 ReplicaSets.

Install the latest version

helm upgrade --install cloudzero-cloudwatch-metrics           \
             cloudzero/cloudzero-cloudwatch-metrics           \
             --namespace cloudzero-metrics --create-namespace \
             --set clusterName=<Your Cluster>

Note: these are helm3 commands that creates a namespace for this deployment; you can also use an existing namespace.

Sampling Frequency

By default, the agent queries and pushes metrics every 120 seconds. This is a granular enough sampling rate to get usable metrics while keeping costs low. The potential downside is that very short lived pods will not be sampled if they are launched and terminated between sampling times. If sampling short-lived pods is desirable, the agent can be configured to record samples at shorter intervals by updating the metricsCollectionInterval value to one of the allowed values. Please be aware that an increased sampling rate will cause an increase in cost, as well as increased load on the kube API.

Configuration

Parameter	Description	Default	Required
image.repository	Image to deploy	cloudzero/cloudwatch-agent	✔
image.tag	Image tag to deploy	cz-cloudwatchagent-0853e59.1
image.pullPolicy	Pull policy for the image	IfNotPresent	✔
clusterName	Name of your cluster	cluster_name	✔
serviceAccount.create	Whether a new service account should be created	true
serviceAccount.name	Service account to be used
hostNetwork	Allow to use the network namespace and network resources of the node	true
nodeSelector	Node labels for pod assignment	{}
tolerations	Optional deployment tolerations	{}
annotations	Optional pod annotations	{}
metricsCollectionInterval	Metrics collection interval in seconds. It can be set to 5, 10, 15, 30, 45, 60 or 120	120