This page describes how to auto-instrument Kubernetes workloads running in a cluster where the Cluster Agent is deployed. See Install the Cluster Agent.

For instrumentation options, see Container Installation Options

AppDynamics recommends using auto-instrumentation to simplify operations.

With auto-instrumentation, you can dynamically add an App Server Agent to workloads for these application types:

  • Java
  • .NET Core on Linux
  • Node.js

We support these Kubernetes workloads: DeploymentsDeploymentConfigs and StatefulSets.

Auto-Instrumentation Overview

To enable auto-instrumentation, you:

Then, you can apply the changes using kubectl or upgrade the Cluster Agent Helm Chart.

When the Cluster Agent detects a supported workload, and the workload matches the configured auto-instrumentation rules, the Cluster Agent modifies the workload's spec using the Kubernetes API. The Cluster Agent attaches an init container with the AppDynamics .NET Core, Node.js, or Java Agent image to the workload. When the application restarts, the required agent is copied into the application container. As a result, the application container references the AppDynamics agent (Node.js Agent, .NET Core on Linux Agent, or Java Agent) in an auto-instrumented application. 

Requirements

AppDynamics Requirements

  • Cluster Agent >= 20.5. See Cluster Agent Requirements and Supported Environments.

  • Installed the latest Cluster Agent and Operator versions in the cluster. The cluster-agent-operator.yaml sets up the permissions required by the Cluster Agent to perform auto-instrumentation. See Install the Cluster Agent.
  • At least one application is deployed to the cluster that was not previously instrumented with the required AppDynamics agent.
  • A Controller with sufficient agent licenses based on the number of applications that will be auto-instrumented.
  • Ensure that you have sufficient cluster capacity to process pod restarts. See Minimize the Impact of Pod Restarts.
  • Ensure that you have specified the app, the tier, and the node name tuple to be unique across all the Kubernetes instances. If the name is not unique, the nodes may not report properly.
    This is applicable to those agents on which app, tier, and node uniqueness is required such as Java and Node.js Agents.

Language-Specific Requirements

  • Node.js >= 8.6

  • Java 
    • Java applications must support including the -javaagent argument in the Java command using an environment variable. By default, the Cluster Agent uses JAVA_TOOL_OPTIONS; however, you can change this using the defaultEnv property.
    • Based on Java Agent resource requirements, it may be necessary to adjust the configured memory requests or limits for the application pods (see Install the Java Agent).
  • .NET Core on Linux and Node.js
    • Ensure that the application base image Operating System (OS) matches the App Server Agent base image OS (Linux versus Alpine). For example, if your .NET Core on Linux application uses an Ubuntu base image, then you must set the imageInfo.image tag to the Linux version. In this example, the image tag is 20.11.0-linux
      If the application used an Alpine Linux base image OS, then the tag would be 20.11.0-alpine. See the AppDynamics Docker Hub page.

      apiVersion: appdynamics.com/v1alpha1
      kind: Clusteragent
      metadata:
        name: k8s-cluster-agent
        namespace: appdynamics
      spec:
        # content removed for brevity
        instrumentationRules:
          - namespaceRegex: dev
            language: dotnetcore
            appName: MyDotNetAppOnUbuntu
            imageInfo:
              image: "docker.io/appdynamics/dotnet-core-agent:20.11.0-linux"
              agentMountPath: /opt/appdynamics
      YML

      If the base image Operating Systems do not match, then the App Server Agent may not start. See Validate Auto-Instrumentation. The Node.js Agent also has specific Node.js runtime requirements that may prevent the agent from starting. See Node.js Supported Environments.

    • For .NET Core and Node.js applications that communicate with an on-premises Controller, see Use Auto-Instrumentation with an On-Premises Controller.
    • If transaction analytics data is required, then you must configure the Cluster Agent analyticsHost and analyticsPort properties.

Enable Auto-Instrumentation for the Cluster Agent 

To set up the Cluster Agent feature:

  1. First, remove any deleted pods from the Controller Tiers & Nodes Dashboard. Then, re-create the cluster-agent-secret that was created in Install the Cluster Agent to include api-user. Set the api-user value to a local user from the Controller with the Administrator role:

    kubectl -n appdynamics delete secret cluster-agent-secret
    kubectl -n appdynamics create secret generic cluster-agent-secret --from-literal=controller-key=<access-key> --from-literal=api-user="<username>@<customer>:<password>"
    CODE

    The Cluster Agent uses the api-user to mark the associated node in the Controller as historical upon pod deletion.
     

  2. Add auto-instrumentation configuration to the cluster-agent.yaml or the Helm values.yaml file. The configuration determines which DeploymentsDeploymentConfigs and StatefulSets workloads to target for auto-instrumentation and which agent types and versions to use. See Auto-Instrumentation Configuration.

  3. After you save the configuration, apply or upgrade the Cluster Agent deployment. The related pods and containers restart based on the deployment rollout strategy associated with the applications.

    kubectl apply -f cluster-agent.yaml
    BASH
    helm upgrade -f ./ca1-values.yaml "<my-cluster-agent-helm-release>" appdynamics-charts/cluster-agent --namespace appdynamics
    BASH


    To validate and troubleshoot auto-instrumentation, see Validate the Cluster Agent Installation.

  • If a workload does not match the properties defined in instrumentationRules, then auto-instrumentation is not enabled. 
  • If an auto-instrumentation property is not defined as a default, or in instrumentationRules, then the Cluster Agent uses the corresponding default value specified in Auto-Instrumentation Configuration. If there are no corresponding default values, then auto-instrumentation is not enabled.

Configuration Examples

Example 1 targets Java applications in the namespaces that match the ecom.* pattern. Each matching application will be instrumented with a 20.20.1 Java Agent and will report to the Ecommerce application in the AppDynamics Controller. By default, the tier name is the name of the Kubernetes workload, but you can override it by setting the tierName property.

Example 1: cluster-agent-auto-1.yaml

apiVersion: appdynamics.com/v1alpha1
kind: Clusteragent
metadata:
  name: k8s-cluster-agent
  namespace: appdynamics
spec:
  appName: "<app-name>"
  controllerUrl: "<protocol>://<appdynamics-controller-host>:8080"
  account: "<account-name>"
  image: "docker.io/appdynamics/cluster-agent:20.12.1"
  serviceAccountName: appdynamics-cluster-agent
  nsToMonitorRegex: ecom.*
  #
  # auto-instrumentation config
  #
  instrumentationMethod: Env
  nsToInstrumentRegex: ecom.*
  defaultAppName: Ecommerce
  instrumentationRules:
    - language: java
      imageInfo:
        image: docker.io/appdynamics/java-agent:20.20.1
        agentMountPath: /opt/appdynamics
YML


Example 2 targets namespaces that contain Java and .NET Core on Linux applications, and incorporates these advanced configurations:

  • Uses multiple instrumentationRules to target Java applications versus .NET Core on Linux applications.
  • Uses the labelMatch strategy to determine the agent type and associated agent image based on the value of the framework label in the workload spec auto-instrumented-dotnet-app.yaml and auto-instrumented-java-app.yaml below.
  • Rather than assigning a Controller application name in the YAML file, the configuration uses appNameStrategy: label to assign an application name based on a label from the workload spec.
  • For the Java applications, it uses instrumentContainer: select and containerMatchString.*service to instruct the Cluster Agent to auto-instrument only the application service container only, and ignore any other defined containers defined.

Example 2: cluster-agent-auto-2.yaml

apiVersion: appdynamics.com/v1alpha1
kind: Clusteragent
metadata:
  name: k8s-cluster-agent
  namespace: appdynamics
spec:
  appName: "<app-name>"
  controllerUrl: "<protocol>://<appdynamics-controller-host>:8080"
  account: "<account-name>"
  image: "docker.io/appdynamics/cluster-agent:20.12.1"
  serviceAccountName: appdynamics-cluster-agent
  nsToMonitorRegex: ecom.*
  #
  # auto-instrumentation config
  #
  instrumentationMethod: Env
  nsToInstrumentRegex: stage
  appNameStrategy: label
  instrumentationRules:
    - namespaceRegex: stage
      language: dotnetcore
      labelMatch:
        - framework: dotnetcore
      appNameLabel: appName
      imageInfo:
        image: "docker.io/appdynamics/dotnet-core-agent:20.11.0-linux"
        agentMountPath: /opt/appdynamics
    - namespaceRegex: stage
      language: java
      labelMatch:
        - framework: java
      appNameLabel: appName
      instrumentContainer: select
      containerMatchString: .*service
      imageInfo:
        image: "docker.io/appdynamics/java-agent:21.3.0"
        agentMountPath: /opt/appdynamics
YML

Examples 3 and 4 show Deployment specs for .NET and Java services that define the appName and framework labels, based on the auto-instrumentation configuration from the cluster-agent-auto-2.yaml:

.NET

apiVersion: apps/v1
kind: Deployment
metadata:
  name: dotnet-profile-service
  labels:
    appName: backend-services
    framework: dotnetcore
spec:
  containers:
    - image: myrepo/profile-service:v2
      name: profile-service
  # ...
YML

Java

apiVersion: apps/v1
kind: Deployment
metadata:
  name: java-account-service
  labels:
    appName: backend-services
    framework: java
spec:
  containers:
    - image: myrepo/account-service:v2
      name: account-service
    - image: myrepo/proxy-util:v1
      name: proxy-util
  # ...
YML

The value of containerMatchString in cluster-agent-auto-2.yaml indicates that only the account-service container will be auto-instrumented in auto-instrumented-java-app.yaml.

For additional configuration examples, see Auto-Instrumentation Configuration Examples.

AppDynamics Application Name Strategies

The Controller's Application Dashboard provides three application name strategies. Select a strategy by assigning the appNameStrategy property to one of these values:

Manual Strategy

By default, the appNameStrategy is manual, which uses the defaultAppName or appName parameter to set the application name.

  • If defaultAppName is provided, then use it (unless overwritten in an instrumentation rule).
  • If appName is provided in an instrumentation rule, then use it.

For example in this spec, ECommerce is the default application name applied to the ecom and groceries namespace, and BookStore is the application name applied to the books namespace.

apiVersion: appdynamics.com/v1alpha1
kind: Clusteragent
metadata:
  name: k8s-cluster-agent
  namespace: appdynamics
spec:
  appName: "<cluster-name>"
  # ...
  # auto-instrumentation config
  instrumentationMethod: Env
  nsToInstrumentRegex: ecom|books|groceries
  appNameStrategy: manual
  defaultAppName: ECommerce
  instrumentationRules:
    - namespaceRegex: books
      appName: BookStore
YML

Label Strategy

This option uses the label parameter as the application name strategy. To use the label option, specify a value in the appNameLabel parameter. The appNameLabel value refers to a label specified in the workload spec.  

  • If spec.appNameLabel is specified, then spec level value is used.
  • If appNameLabel is specified in an instrumentation rule, then that value is used unless a different appNameLabel is specified in the spec.
  • If the appNameLabel mentioned in the instrumentation rule is not found in the deployment spec, then the spec level appNameLabel value is used.

In the following example spec, appNameLabel: app is used in the instrumentation rule, but the deployment spec does not have the label app. The spec has a label called appname which has a value eCommerce.
Therefore, Controller displays the data that is reporting to eCommerce application.

In the following spec, the workload spec label appname is used to set the application name in the ecom and groceries namespaces, and the label app is used in the books namespace.

apiVersion: appdynamics.com/v1alpha1
kind: Clusteragent
metadata:
  name: k8s-cluster-agent
  namespace: appdynamics
spec:
  appName: "<cluster-name>"
  # ...
  # auto-instrumentation config
  instrumentationMethod: Env
  nsToInstrumentRegex: ecom|books|groceries
  appNameStrategy: label
  appNameLabel: appname
  instrumentationRules:
    - namespaceRegex: books
      appNameLabel: app
YML

For an application deployed to the ecom or groceries namespaces that sets the label appname (shown in this Deployment spec snippet), it reports to the eCommerce application in the Controller's Application Dashboard.

apiVersion: apps/v1
kind: Deployment
metadata:
  name: ecom-app
  labels:
    appname: eCommerce
spec:
...
YML

Namespace Strategy

This option uses the Kubernetes namespace parameter as the application name strategy and allows you to use the namespace name where an application is deployed as the application name in the Controller's Application Dashboard.

In this spec, each application in ecombooks, and groceries namespace uses the application name based on the namespace that it is deployed to.

apiVersion: appdynamics.com/v1alpha1
kind: Clusteragent
metadata:
  name: k8s-cluster-agent
  namespace: appdynamics
spec:
  appName: "<cluster-name>"
  # ...
  # auto-instrumentation config
  instrumentationMethod: Env
  nsToInstrumentRegex: ecom|books|groceries
  appNameStrategy: namespace
YML

Minimize the Impact of Pod Restarts

When auto-instrumentation is enabled, the related pods restart based on the deployment rollout strategy associated with the workload. Pod restarts often create CPU and memory usage spikes that may adversely impact performance or exhaust available capacity. To accommodate pods restarts, you may need to increase memory and CPU quotas associated with the impacted namespaces. To reduce the impact of restarting a large number of pods, the Cluster Agent (by default) allows only two concurrent auto-instrumentation tasks. The subsequent workloads (resourcesToInstrument) are auto-instrumented after the rollout of an instrumented workload. However, you can configure the parameter, numberOfTaskWorkers, to specify the number of concurrent auto-instrumentation tasks based on your cluster's requirements.