Skip to content

The Kubernetes operator of Lumigo provides a one-click solution to monitoring Kubernetes clusters with Lumigo

License

Notifications You must be signed in to change notification settings

lumigo-io/lumigo-kubernetes-operator

Repository files navigation

Lumigo Kubernetes Operator

The Lumigo Logo

Artifact Hub

The Kubernetes operator of Lumigo provides a one-click solution to monitoring Kubernetes clusters with Lumigo.

Setup

Installation

Install the Lumigo Kubernetes operator in your Kubernets cluster with helm:

helm repo add lumigo https://lumigo-io.github.io/lumigo-kubernetes-operator
helm install lumigo lumigo/lumigo-operator --namespace lumigo-system --create-namespace --set cluster.name=<cluster_name>

Note: You have the option to alter the namespace from lumigo-system to a name of your choosing, but its important to be aware that doing so might cause slight discrepancies throughout the steps below.

(The cluster.name is optional, but highly advised, see the Naming your cluster section.)

You can verify that the Lumigo Kubernetes operator is up and running with:

$ kubectl get pods -n lumigo-system
NAME                                                         READY   STATUS    RESTARTS   AGE
lumigo-kubernetes-operator-7fc8f67bcc-ffh5k   2/2     Running   0          56s

Note: While installing the Lumigo Kubernetes operator via kustomize is generally expected to work (except the uninstallation of instrumentation on removal), it is not actually supported1.

EKS on Fargate

On EKS, the pods of the Lumigo Kubernetes operator itself need to be running on nodes running on Amazon EC2 virtual machines. Your monitored applications, however, can run on the Fargate profile without any issues. Installing the Lumigo Kubernetes operator on an EKS cluster without EC2-backed nodegroups, results in the operator pods staying in Pending state:

$ kubectl describe pod -n lumigo-system lumigo-kubernetes-operator-5999997fb7-cvg5h

Namespace:    	lumigo-system
Priority:     	0
Service Account:  lumigo-kubernetes-operator
Node:         	<none>
Labels:       	app.kubernetes.io/instance=lumigo
              	app.kubernetes.io/name=lumigo-operator
              	control-plane=controller-manager
              	lumigo.auto-trace=false
              	lumigo.cert-digest=dJTiBDRVJUSUZJQ
              	pod-template-hash=5999997fb7
Annotations:  	kubectl.kubernetes.io/default-container: manager
              	kubernetes.io/psp: eks.privileged
Status:       	Pending

(The reason for this limitation is very long story, but it is necessary for Lumigo to figure out which EKS cluster is the operator sending data from.) If you are installing the Lumigo Kubernetes operator on an EKS cluster with only the Fargate profile, add a managed nodegroup.

Naming your cluster

Kubernetes clusters does not have a built-in nothing of their identity1, but when running multiple Kubernetes clusters, you almost certainly have names from them. The Lumigo Kubernetes operator will automatically add to your telemetry the k8s.cluster.uid OpenTelemetry resource attribute, set to the value of the UID of the kube-system namespace, but UIDs are not meant for humans to remember and recognize easily. The Lumigo Kubernetes operator allows you to set a human-readable name using the cluster.name Helm setting, which enables you to filter all your tracing data based on the cluster in Lumigo's Explore view.

1 Not even Amazon EKS clusters, as their ARN is not available anywhere inside the cluster itself.

Upgrading

You can check which version of the Lumigo Kubernetes operator you have deployed in your cluster as follows:

$ helm ls -A
NAME  	NAMESPACE    	REVISION	UPDATED                              	STATUS  	CHART             	APP VERSION
lumigo	lumigo-system	2       	2023-07-10 09:20:04.233825 +0200 CEST	deployed	lumigo-operator-13	13

The Lumigo Kubernetes operator is reported as APP VERSION.

To upgrade to a newer version of the Lumigo Kubernetes operator, run:

helm repo update
helm upgrade lumigo lumigo/lumigo-operator --namespace lumigo-system

Enabling automatic tracing

Supported resource types

The Lumigo Kubernetes operator automatically adds distributed tracing to pods created via:

The distributed tracing is provided by the Lumigo OpenTelemetry distribution for JS, the Lumigo OpenTelemetry distribution for Java and the Lumigo OpenTelemetry distribution for Python.

The Lumigo Kubernetes operator will automatically trace all Java, Node.js and Python processes found in the containers of pods created in the namespaces that Lumigo traces. To activate automatic tracing for resources in a namespace, create in that namespace a Kubernetes secret containing your Lumigo token, and reference it from a Lumigo (operator.lumigo.io/v1alpha1.Lumigo) custom resource. Save the following into the lumigo.yml:

apiVersion: v1
kind: Secret
metadata:
  name: lumigo-credentials
stringData:
  # Kubectl won't allow you to deploy this dangling anchor.
  # Get the actual value from Lumigo following this documentation: https://docs.lumigo.io/docs/lumigo-tokens
  token: *lumigo-token #  <--- Change this! Example: t_123456789012345678901
---
apiVersion: operator.lumigo.io/v1alpha1
kind: Lumigo
metadata:
  labels:
    app.kubernetes.io/name: lumigo
    app.kubernetes.io/instance: lumigo
    app.kubernetes.io/part-of: lumigo-operator
  name: lumigo
spec:
  lumigoToken:
    secretRef:
      name: lumigo-credentials # This must match the name of the secret; the secret must be in the same namespace as this Lumigo custom resource
      key: token # This must match the key in the Kubernetes secret (don't touch)

After creating lumigo.yml, deploy it in the desired namespace:

kubectl apply -f lumigo.yml -n <YOUR_NAMESPACE>

ℹ️ Important note

Apply the secret and the custom resource to the namespace you wish to start tracing, not to lumigo-system.

Each Lumigo resource keeps in its state a list of resources it currently instruments:

$ kubectl describe lumigo -n my-namespace
Name:         lumigo
Namespace:    my-namespace
API Version:  operator.lumigo.io/v1alpha1
Kind:         Lumigo
Metadata:
  ... # Data removed for readability
Spec:
  ... # Data removed for readability
Status:
  Conditions:
  ... # Data removed for readability
  Instrumented Resources:
    API Version:       apps/v1
    Kind:              StatefulSet
    Name:              my-statefulset
    Namespace:         my-namespace
    Resource Version:  320123
    UID:               93d6d809-ac2a-43a9-bc07-f0d4e314efcc

Disabling automatic tracing

The tracing feature can be entirely turned-off in case it's not desired, by setting the spec.tracing.enabled field to false in the Lumigo resource:

apiVersion: operator.lumigo.io/v1alpha1
kind: Lumigo
metadata:
  labels:
    app.kubernetes.io/name: lumigo
    app.kubernetes.io/instance: lumigo
    app.kubernetes.io/part-of: lumigo-operator
  name: lumigo
spec:
  lumigoToken: ...
  tracing:
    enabled: false
  logging:
    enabled: true # usually set to when `tracing.enabled` is `false`, otherwise injecting Lumigo into the workload will not be useful

this is usually done when only logs from the workloads should be sent to Lumigo, regardless of the tracing content in which the logs was generated.

  • Note that this does not affect the injection of the Lumigo distro into pods - only the fact that the distro will not send traces to Lumigo. For more fine-grained control over the injection in general, see the Opting out for specific resources section.

Logging support

The Lumigo Kubernetes operator can automatically forward logs emitted by traced pods to Lumigo's log-management solution, supporting several logging providers (currently logging for Python apps, Winston and Bunyan for Node.js apps). Enabling log forwarding is done by adding the spec.logging.enabled field to the Lumigo resource:

apiVersion: operator.lumigo.io/v1alpha1
kind: Lumigo
metadata:
  labels:
    app.kubernetes.io/name: lumigo
    app.kubernetes.io/instance: lumigo
    app.kubernetes.io/part-of: lumigo-operator
  name: lumigo
spec:
  lumigoToken: ... # same token used for tracing
  logging:
    enabled: true # enables log forwarding for pods with tracing injected

Opting out for specific resources

To prevent the Lumigo Kubernetes operator from injecting tracing to pods managed by some resource in a namespace that contains a Lumigo resource, add the lumigo.auto-trace label set to false:

apiVersion: apps/v1
kind: Deployment
metadata:
  labels:
    app: hello-node
    lumigo.auto-trace: "false"  # <-- No injection will take place
  name: hello-node
  namespace: my-namespace
spec:
  selector:
    matchLabels:
      app: hello-node
  template:
    metadata:
      labels:
        app: hello-node
    spec:
      containers:
      - command:
        - /agnhost
        - netexec
        - --http-port=8080
        image: registry.k8s.io/e2e-test-images/agnhost:2.39
        name: agnhost

In the logs of the Lumigo Kubernetes operator, you will see a message like the following:

1.67534267851615e+09    DEBUG   controller-runtime.webhook.webhooks   wrote response   {"webhook": "/v1alpha1/inject", "code": 200, "reason": "the resource has the 'lumigo.auto-trace' label set to 'false'; resource will not be mutated", "UID": "6d341941-c47b-4245-8814-1913cee6719f", "allowed": true}

Settings

Inject existing resources

By default, when detecting a new Lumigo resource in a namespace, the Lumigo controller will instrument existing resources of the supported types. The injection will cause new pods to be created for daemonsets, deployments, replicasets, statefulsets and jobs; cronjobs will spawn injected pods at the next iteration. To turn off the automatic injection of existing resources, create the Lumigo resource as follows

apiVersion: operator.lumigo.io/v1alpha1
kind: Lumigo
metadata:
  labels:
    app.kubernetes.io/name: lumigo
    app.kubernetes.io/instance: lumigo
    app.kubernetes.io/part-of: lumigo-operator
  name: lumigo
spec:
  lumigoToken: ...
  tracing:
    injection:
      injectLumigoIntoExistingResourcesOnCreation: false # Default: true

Remove injection from existing resources

By default, when detecting the deletion of the Lumigo resource in a namespace, the Lumigo controller will remove instrumentation from existing resources of the supported types. The injection will cause new pods to be created for daemonsets, deployments, replicasets, statefulsets and jobs; cronjobs will spawn non-injected pods at the next iteration. To turn off the automatic removal of injection from existing resources, create the Lumigo resource as follows

apiVersion: operator.lumigo.io/v1alpha1
kind: Lumigo
metadata:
  labels:
    app.kubernetes.io/name: lumigo
    app.kubernetes.io/instance: lumigo
    app.kubernetes.io/part-of: lumigo-operator
  name: lumigo
spec:
  lumigoToken: ...
  tracing:
    injection:
      removeLumigoFromResourcesOnDeletion: false # Default: true

Note: The removal of injection from existing resources does not occur on uninstallation of the Lumigo Kubernetes operator, as the role-based access control is has likely already been deleted.

Collection of Kubernetes objects

The Lumigo Kubernetes operator will automatically collect Kubernetes object versions in the namespaces with a Lumigo resource in active state, and send them to Lumigo for issue detection (e.g., when you pods crash). The collected object types are: corev1.Events, corev1.Pods, appsv1.Deployments, apps/v1.DaemonSet, apps/v1.ReplicaSet, apps/v1.StatefulSet, batch/v1.CronJob, and batch/v1.Job. Besides events, the object versions, e.g., pods, replicasets and deployments, are needed to be able to correlate events across the owner-reference chain, e.g., the pod belongs to that replicaset, which belongs to that deployment.

To disable the automated collection of Kubernetes events and object versions, you can configure your Lumigo resources as follows:

apiVersion: operator.lumigo.io/v1alpha1
kind: Lumigo
metadata:
  labels:
    app.kubernetes.io/name: lumigo
    app.kubernetes.io/instance: lumigo
    app.kubernetes.io/part-of: lumigo-operator
  name: lumigo
spec:
  lumigoToken: ...
  infrastructure:
    kubeEvents:
      enabled: false # Default: true

When a Lumigo resource is deleted from a namespace, the collection of Kubernetes events and object versions is automatically halted.

Modify manager log level

By default, the manager will log all INFO level and above logs.

The current log level can be viewed by running:

kubectl -n lumigo-system get deploy lumigo-lumigo-operator-controller-manager -o=json | jq '.spec.template.spec.containers[0].args'

With the default settings, there will be no log level explicitly set and the above command will return:

[
  "--health-probe-bind-address=:8081",
  "--metrics-bind-address=127.0.0.1:8080",
  "--leader-elect"
]

To set the log level to only show ERROR level logs, run:

kubectl -n lumigo-system patch deploy lumigo-lumigo-operator-controller-manager --type=json -p='[{"op": "add", "path": "/spec/template/spec/containers/0/args/-", "value": "--zap-log-level=error"}]'

If a log level is already set, instead of using the add operation we use replace and modify the path from /args/- to the index of containing the log level setting, such as /args/3:

kubectl -n lumigo-system patch deploy lumigo-lumigo-operator-controller-manager --type=json -p='[{"op": "replace", "path": "/spec/template/spec/containers/0/args/3", "value": "--zap-log-level=info"}]'

NOTE: The container argument array is zero indexed, so the first argument is at index 0.

Uninstall

The removal of the Lumigo Kubernetes operator is performed by:

helm delete lumigo --namespace lumigo-system

In namespaces with the Lumigo resource having spec.tracing.injection.enabled and spec.tracing.injection.removeLumigoFromResourcesOnDeletion both set to true, supported resources that have been injected by the Lumigo Kubernetes operator will be updated to remove the injection, with the following caveat:

Note: The removal of injection from existing resources does not apply to batchv1.Job resources, as their corev1.PodSpec is immutable after the batchv1.Job resource has been created.

TLS certificates

The Lumigo Kubernetes operator injector webhook uses a self-signed certificate that is automatically generate during the installation of the Helm chart. The generated certificate has a 365 days expiration, and a new certificate will be generated every time you upgrade Lumigo Kubernetes operator's helm chart.

Events

The Lumigo Kubernetes operator will add events to the resources it instruments with the following reasons and in the following cases:

Reason Created on resource types Under which conditions
LumigoAddedInstrumentation apps/v1.Deployment, apps/v1.DaemonSet, apps/v1.ReplicaSet, apps/v1.StatefulSet, batch/v1.CronJob  If a Lumigo resources exists in the namespace, and the resource is instrumented with Lumigo as a result
LumigoCannotAddInstrumentation apps/v1.Deployment, apps/v1.DaemonSet, apps/v1.ReplicaSet, apps/v1.StatefulSet, batch/v1.CronJob  If a Lumigo resources exists in the namespace, and the resource should be instrumented by Lumigo as a result, but an error occurs
LumigoUpdatedInstrumentation apps/v1.Deployment, apps/v1.DaemonSet, apps/v1.ReplicaSet, apps/v1.StatefulSet, batch/v1.CronJob  If a Lumigo resources exists in the namespace, and the resource has the Lumigo instrumented updated as a result
LumigoCannotUpdateInstrumentation apps/v1.Deployment, apps/v1.DaemonSet, apps/v1.ReplicaSet, apps/v1.StatefulSet, batch/v1.CronJob  If a Lumigo resources exists in the namespace, and the resource should have the Lumigo instrumented updated as a result, but an error occurs
LumigoRemovedInstrumentation apps/v1.Deployment, apps/v1.DaemonSet, apps/v1.ReplicaSet, apps/v1.StatefulSet, batch/v1.CronJob  If a Lumigo resources is deleted from the namespace, and the resource has the Lumigo instrumented removed as a result
LumigoCannotRemoveInstrumentation apps/v1.Deployment, apps/v1.DaemonSet, apps/v1.ReplicaSet, apps/v1.StatefulSet, batch/v1.CronJob  If a Lumigo resources is deleted from the namespace, and the resource should have the Lumigo instrumented removed as a result, but an error occurs

Footnotes

  1. The user experience of having to install Cert Manager is unnecessarily complex, and Kustomize layers, while they may be fine for one's own applications, are simply unsound for a batteries-included, rapidly-evolving product like the Lumigo Kubernetes operator. Specifically, please expect your Kustomize layers to stop working with any release of the Lumigo Kubernetes operator. 2 3