AppDynamics Operator simplifies the configuration and lifecycle management of the AppDynamics ClusterAgent and the AppDynamics Machine Agent on different Kubernetes distributions and OpenShift. The Operator encapsulates key operational knowledge on how to configure and upgrade the ClusterAgent and the Machine Agent. It knows, for example, which configuration changes are benign and do not require restart of the ClusterAgent, which minimizes unnecesary load on the cluster API server.
The Operator is implemented using OperatorSDK and uses Kubernetes API to maintain the desired state of the custom resources that represent the ClusterAgent and the Machine Agent. When the Operator is deployed, it creates custom resource definitions (CRDs) for 2 custom resources:
- clusteragent, which represents the ClusterAgent.
- infraviz, which represents the Machine Agent bundled with netviz and analytics.
This level of abstraction further simplifies the management of monitoring and imstrumentation and ensures granular security policy of the ClusterAgent and the Machine Agent.
Create namespace for the operator and the ClusterAgent
-
Create namespace for AppDynamics components
- Kubernetes
kubectl create namespace appdynamics
- OpenShift
oc new-project appdynamics --description="AppDynamics Infrastructure"
- Kubernetes
-
Create Secret
cluster-agent-secret
with the following key- The "controller-key" - the access key to the AppDynamics controller.
kubectl -n appdynamics create secret generic cluster-agent-secret \
--from-literal=controller-key="<controller-access-key>" \
- Update the image reference in the Operator deployment spec (deploy/cluster-agent-operator.yaml), if necessary.
The default is "docker.io/appdynamics/cluster-agent-operator:latest".
- Deploy the Operator
kubectl create -f deploy/cluster-agent-operator.yaml
- On OpenShift
oc create -f deploy/cluster-agent-operator-openshift.yaml
By default "docker.io/appdynamics/cluster-agent-operator:latest" is used.
AppDynamics images are also available from Red Hat Container Catalog.
To enable pulling, create a secret in the ClusterAgent namespace. In this example, namespace appdynamics is used and appdynamics-operator account is linked to the secret.
$ oc -n appdynamics create secret docker-registry redhat-connect
--docker-server=registry.connect.redhat.com
--docker-username=REDHAT_CONNECT_USERNAME
--docker-password=REDHAT_CONNECT_PASSWORD --docker-email=unused
$ oc -n appdynamics secrets link appdynamics-operator redhat-connect
--for=pull
The AppDynamics Cluster Agent is a new, lightweight agent written in Golang providing APM monitoring for Kubernetes. The Cluster Agent is designed to help you understand how Kubernetes infrastructure affects your applications and business performance. With the AppDynamics Cluster Agent, you can collect metadata, metrics, and events about a Kubernetes cluster. You can query for the components in the cluster, and report metrics for those components. The Cluster Agent works across cloud platforms such as Kubernetes on AWS EKS, AKS on Azure, PKS on Pivotal and many others.
The clusteragent
is the custom resource that the Operator works with to deploy an instance of the ClusterAgent. When a clusteragent spec is provided, the Operator will create a single replica deployment and the necessary additional resources (a configMap and a service) to support the ClusterAgent.
Here is an example of a minimalistic spec of the ClusterAgent custom resource:
apiVersion: appdynamics.com/v1alpha1
kind: Clusteragent
metadata:
name: k8s-cluster-agent
namespace: appdynamics
spec:
controllerUrl: "https://saas.appdynamics.com"
account: "customer1"
appName: "Cluster1"
Parameter | Description | Default |
---|---|---|
controllerUrl |
Url of the AppDynamics controller | Required |
account |
AppDynamics Account Name | Required |
image |
Cluster Agent image reference | During the Beta period access to the image must be requested from AppDynamics |
resources |
Definitions of resources and limits for the machine agent | See resource recommendations below |
appName |
Name of the cluster | Required |
metricsSyncInterval |
Sampling interval at which metrics should be collected periodically | Default 30 sec |
logLevel |
Logging level (INFO , DEBUG , WARN , TRACE ) |
Default INFO |
logFileSizeMb |
The maximum file size of the log in MB. | Default is 5 |
logFileBackups |
The maximum number of backups the log saves. When the maximum number of backups is reached, the 3 oldest log file after the initial log file is deleted. | Default is 3 |
Example resource limits:
resources:
limits:
cpu: 300m
memory: "200M"
requests:
cpu: 200m
memory: "100M"
Appdynamics operator can be used to enable server and network visibility with AppDynamics Machine agent.
The operator works with custom resource infraviz
to deploy the AppDynamics Machine Agent daemon set.
Here is an example of a minimalistic infraviz
spec with the required parameters:
apiVersion: appdynamics.com/v1alpha1
kind: InfraViz
metadata:
name: appd-infraviz
namespace: appdynamics
spec:
controllerUrl: "https://appd-controller.com"
image: "docker.io/appdynamics/machine-agent-analytics:latest"
account: "<your-account-name>"
globalAccount: "<your-global-account-name"
The controller URL must be in the following format:
<protocol>://<controller-domain>:<port>
Parameter | Description | Default |
---|---|---|
controllerUrl |
Url of the AppDynamics controller | Required |
account |
AppDynamics Account Name | Required |
globalAccount |
Global Account Name | Required |
eventServiceUrl |
Event Service Endpoint | Optional |
enableContainerHostId |
Flag that determines how container names are derived (pod name vs container id) | "true" |
enableServerViz |
Enable Server Visibility | "true" |
enableDockerViz |
Enable Docker Container Visibiltiy | "true" |
uniqieHostId |
Unique host ID in AppDynamics. | Optional |
metricsLimit |
Number of metrics that the Machine Agent is allowed to post to the controller | Optional |
logLevel |
Logging level (info or debug ) |
info |
stdoutLogging |
Determines if the logs are saved to a file or redirected to the console | "false" |
netVizPort |
When > 0, the network visibility agent will be deployed in a sidecar along with the machine agent. By default the network visibility agent works with port 3892 |
Not set by default |
proxyUrl |
Url of the proxy server (protocol://domain:port") | Optional |
proxyUser |
Proxy user credentials (user@password) | Optional |
propertyBag |
A string with any other machine agent parameters | Optional |
pks |
PKS deployment flag. Must be used on PKS environments | Optional |
enableMasters |
When set to true server visibility will be provided for Master nodes. By default only Worker nodes are monitored. On managed Kubernetes providers the flag has no effect, as the Master plane is not accessible | Optional |
image |
The Machine Agent image | "appdynamics/machine-agent-analytics:latest" |
nodeSelector |
A set of nodes to deploy the daemon set pods to | Optional |
tolerations |
A list of tolerations | Optional |
env |
List of environment variables | Optional |
args |
List of command arguments | Optional |
resources |
Definitions of resources and limits for the machine agent | See example below |
Example resource limits:
resources:
limits:
cpu: 600m
memory: "1G"
requests:
cpu: 300m
memory: "800M"