Instrument applications with EDOT SDKs on Kubernetes

Serverless Observability Stack EDOT Collector

The Elastic Distributions of OpenTelemetry (EDOT) SDKs cover multiple languages. Read on to learn how to instrument applications for APM in a Kubernetes environment for all supported languages.

In Kubernetes environments with the OpenTelemetry Operator, automatic or zero-code instrumentation simplifies the instrumentation process by injecting and configuring instrumentation libraries into the targeted Pods.

On the other hand, manual instrumentation with OpenTelemetry allows you to customize trace spans, metrics, and logging directly in your application’s code. This approach provides more granular control over what and how data is captured.

Before starting with application auto-instrumentation, ensure the following prerequisites are in place for proper setup:

• Install the OpenTelemetry operator and EDOT collectors following the [quickstart guide]/solutions/observability/get-started/opentelemetry/quickstart/index.md).
• Ensure a valid kind: Instrumentation object exists in the cluster.

Zero-code instrumentation is handled by the operator through Instrumentation objects, used to automatically inject the necessary SDKs and configuration into application workloads.

If you followed the [quickstart guide]/solutions/observability/get-started/opentelemetry/quickstart/index.md) to install the operator, there should be an Instrumentation object with name elastic-instrumentation in namespace opentelemetry-operator-system:

 kubectl get instrumentation -A NAMESPACE NAME AGE ENDPOINT SAMPLER SAMPLER ARG opentelemetry-operator-system elastic-instrumentation 5d20h http://opentelemetry-kube-stack-daemon-collector.opentelemetry-operator-system.svc.cluster.local:4318 parentbased_traceidratio 1.0 

The Instrumentation object stores important parameters:

• The exporter endpoint represents the destination for the traces, in this case the HTTP receiver configured in the EDOT DaemonSet Collector. That endpoint has to be reachable by the Pods being instrumented.

   exporter: endpoint: http://opentelemetry-kube-stack-daemon-collector.opentelemetry-operator-system.svc.cluster.local:4318 

• Language-specific images used by the operator to inject the appropriate library into each Pod.

   dotnet: image: docker.elastic.co/observability/elastic-otel-dotnet:1.1.0 java: image: docker.elastic.co/observability/elastic-otel-javaagent:1.6.0 nodejs: image: docker.elastic.co/observability/elastic-otel-node:1.5.0 python: image: docker.elastic.co/observability/elastic-otel-python:1.9.0 

To turn on auto-instrumentation, add the corresponding language annotation to the Pods template in your Deployment or relevant workload object, like StatefulSet, Job, CronJob, and so on.

For example, the template would contain the following under spec.template.metadata.annotations:

 apiVersion: apps/v1 kind: Deployment metadata: name: myapp spec: # ... template: metadata: annotations: instrumentation.opentelemetry.io/inject-<LANGUAGE>: "opentelemetry-operator-system/elastic-instrumentation" # ... spec: containers: - image: myapplication-image name: app # ... 

Where <LANGUAGE> is one of: go , java, nodejs, python, dotnet

You can also turn on auto-instrumentation by adding the annotation at namespace level. This approach automatically applies instrumentation to all Pods within the specified namespace.

 apiVersion: v1 kind: Namespace metadata: name: mynamespace annotations: instrumentation.opentelemetry.io/inject-<LANGUAGE>: "opentelemetry-operator-system/elastic-instrumentation" 

After adding annotations to Pods or Namespaces, restart the applications for the instrumentation injection to take effect:

 kubectl rollout restart deployment/my-deployment 

In case you have multiple Instrumentation objects with different settings or images, make sure you point your Pods to the desired Instrumentation objects in the annotations.

The possible values for the annotation are detailed in the Operator documentation. For reference purposes, the values are:

• "true": to inject Instrumentation instance with default name from the current namespace.
• "my-instrumentation": to inject Instrumentation instance with name "my-instrumentation" in the current namespace.
• "my-other-namespace/my-instrumentation": to inject Instrumentation instance with name "my-instrumentation" from another namespace "my-other-namespace".
• "false": do not inject.

For details on instrumenting specific languages, refer to:

• Instrumenting Java
• Instrumenting Python
• Instrumenting Node.js

The following example creates a namespace with an annotation to instrument all Pods of the namespace with java libraries.

 kubectl create namespace java-apps # Annotate app namespace kubectl annotate namespace java-apps instrumentation.opentelemetry.io/inject-java="opentelemetry-operator-system/elastic-instrumentation" # Run a java example application in the namespace kubectl run otel-test -n java-apps --env OTEL_INSTRUMENTATION_METHODS_INCLUDE="test.Testing[methodB]" --image docker.elastic.co/demos/apm/k8s-webhook-test 

After adding the annotation and restarting the Pods, run kubectl describe on your application Pod to verify the SDK has been properly attached.

Make sure that the init container, volume, and environment variables described in how auto-instrumentation works have been successfully injected into the Pod.

The OpenTelemetry Operator automates the process of instrumenting applications by injecting the necessary libraries and configuration into the application Pods.

The process may vary slightly depending on the language, but it generally involves the following steps:

• Creating a shared volume: The operator declares an emptyDir shared volume within the Pod, and mounts it the app container and a new init container. This volume serves as the medium for sharing the instrumentation library between the new init container and the application container.

• Adding an init container: The operator adds an init container into the Pod. This container is responsible for copying the OpenTelemetry instrumentation library to the shared volume.

• Configuring the main container: The operator injects environment variables into the main application container to configure OpenTelemetry settings (for example, OTEL_EXPORTER_OTLP_ENDPOINT or OTEL_TRACES_SAMPLER). Additionally, it links the instrumentation library to the application using mechanisms specific to the language runtime, such as:

  • Java: The library is linked through the javaagent option using the JAVA_TOOL_OPTIONS environment variable.
  • Node.js: The library is linked through the NODE_OPTIONS environment variable.
  • Python: The operator uses the PYTHONPATH environment variable to load the library sitecustomize module.

You can apply OTel-specific configuration to your applications at two different levels:

• At Pod/container level, by using OTel-related environment variables.
• At Instrumentation object level, for example configuring different settings per language.

Use cases include the following:

• Change the library to be injected.
• Change the exporter endpoint.
• Apply certain logging level settings (OTEL_LOG_LEVEL).

Consider the creation of different Instrumentation objects for different purposes, such as:

• Different configuration options for certain languages.
• Trying out different versions of the SDKs.

When troubleshooting auto-instrumentation issues in a Kubernetes environment with the OpenTelemetry Operator, follow these steps:

1. Check that the operator is running. For example:

    $ kubectl get pods -n opentelemetry-operator-system NAME READY STATUS RESTARTS AGE opentelemetry-kube-stack-opentelemetry-operator-7b8684cfbdbv4hj 2/2 Running 0 58s ... 

2. Check that the Instrumentation object has been deployed. For example:

    $ kubectl describe Instrumentation -n opentelemetry-operator-system Name: elastic-instrumentation Namespace: opentelemetry-operator-system ... Kind: Instrumentation Metadata: ... Spec: Dotnet: Image: docker.elastic.co/observability/elastic-otel-dotnet:edge Go: Image: ghcr.io/open-telemetry/opentelemetry-go-instrumentation/autoinstrumentation-go:v0.14.0-alpha Java: Image: docker.elastic.co/observability/elastic-otel-javaagent:1.0.0 Nodejs: Image: docker.elastic.co/observability/elastic-otel-node:edge Python: Image: docker.elastic.co/observability/elastic-otel-python:edge ... 

3. Check that your pod is running. The following example checks the banana namespace:

    $ kubectl get pods -n banana NAME READY STATUS RESTARTS AGE example-otel-app 1/1 Running 0 104s 

4. Check that the pod has the instrumentation initcontainer installed, or container for Go.The events must show that the Docker image was successfully pulled and containers started:

    $ kubectl describe pod/example-otel-app -n banana Name: example-otel-app Namespace: banana ... Annotations: instrumentation.opentelemetry.io/inject-java: opentelemetry-operator-system/elastic-instrumentation Init Containers: opentelemetry-auto-instrumentation-java: Container ID: docker://7ecdf3954263d591b994ed1c0519d16322479b1515b58c1fbbe51d3066210d99 Image: docker.elastic.co/observability/elastic-otel-javaagent:1.0.0 Image ID: docker-pullable://docker.elastic.co/observability/elastic-otel-javaagent@sha256:28d65d04a329c8d5545ed579d6c17f0d74800b7b1c5875e75e0efd29e210566a ... Containers: example-otel-app: ... Events: Type Reason Age From Message ---- ------ ---- ---- ------- Normal Scheduled 5m3s default-scheduler Successfully assigned banana/example-otel-app to docker-desktop Normal Pulled 5m3s kubelet Container image "docker.elastic.co/observability/elastic-otel-javaagent:1.0.0" already present on machine Normal Created 5m3s kubelet Created container opentelemetry-auto-instrumentation-java Normal Started 5m3s kubelet Started container opentelemetry-auto-instrumentation-java Normal Pulling 5m2s kubelet Pulling image "docker.elastic.co/demos/apm/k8s-webhook-test" Normal Pulled 5m1s kubelet Successfully pulled image "docker.elastic.co/demos/apm/k8s-webhook-test" in 1.139s (1.139s including waiting). Image size: 406961626 bytes. Normal Created 5m1s kubelet Created container example-otel-app Normal Started 5m1s kubelet Started container example-otel-app 

5. Check your pod log for agent output. For example:

    $ kubectl logs example-otel-app -n banana ... [otel.javaagent 2024-10-11 13:32:44:127 +0000] [main] INFO io.opentelemetry.javaagent.tooling.VersionLogger - opentelemetry-javaagent - version: 1.0.0 ... 

6. If there is no obvious agent log output, restart the pod with agent log level set to debug and look for agent debug output. Setting the agent to debug is different for the different language agents. Add or set the OTEL_LOG_LEVEL environment variable to debug. For example:

    env: - name: OTEL_LOG_LEVEL value: "debug" 

   For Java, you can set the OTEL_JAVAAGENT_DEBUG environment variable to true.

While the Elastic APM Attacher for Kubernetes only supports the Elastic APM application agents, the OpenTelemetry operator can support both the Elastic APM application agents and the EDOT agents. The OpenTelemetry operator has more features and is being actively developed.

Migrating from the Elastic APM Attacher for Kubernetes consists of the following steps:

1. Install the OpenTelemetry operator.
2. Define and install Instrumentation CRDs which correspond to the currently defined values in the existing Elastic APM Attacher for Kubernetes deployment.
3. Change the annotations in deployment definitions from the co.elastic.apm/attach: ... to the instrumentation.opentelemetry.io/inject-... equivalents.
4. Deploy the new deployment definitions.

For example, if you have Elastic APM Java agents version 1.53.0 and Elastic APM Nodejs agents version 4.11.2 defined for your values, and additionally have set the ELASTIC_APM_LOG_SENDING environment variable to true for the Java agent by default, then the corresponding instrumentation would look as in the following snippet. Make sure to replace both <ELASTIC_APM_SERVER_ENDPOINT> and <ELASTIC_API_KEY> with your credentials:

 # # Filename: elastic-apm-instrumentation.yaml # apiVersion: opentelemetry.io/v1alpha1 kind: Instrumentation metadata: name: elastic-apm-instrumentation namespace: opentelemetry-operator-system spec: java: image: docker.elastic.co/observability/apm-agent-java:1.53.0 env: - name: ELASTIC_APM_SERVER_URL value: "<ELASTIC_APM_SERVER_ENDPOINT>" - name: ELASTIC_APM_API_KEY value: "<ELASTIC_API_KEY>" - name: ELASTIC_APM_LOG_SENDING value: "true" nodejs: image: docker.elastic.co/observability/apm-agent-nodejs:4.11.2 env: - name: ELASTIC_APM_SERVER_URL value: "<ELASTIC_APM_SERVER_ENDPOINT>" - name: ELASTIC_APM_API_KEY value: "<ELASTIC_API_KEY>" 

To add the instrumentation to your cluster, run the following command:

 kubectl apply -f elastic-apm-instrumentation.yaml 

After you've defined the new instrumentations, migrate your pods by replacing the annotation that currently applies to them with the new annotation for the instrumentation.

For example if you had a Java application deployment definition which specified the Elastic APM Attacher for Kubernetes annotation of co.elastic.apm/attach: java, you can replace that annotation with the equivalent annotation for the OpenTelemetry operator using the new instrumentation names you have defined. For the previous example Instrumentation, the annotation would be instrumentation.opentelemetry.io/inject-java: "opentelemetry-operator-system/elastic-apm-instrumentation"

Subsequent deployments will use the new instrumentation. The pods themselves will be auto-instrumented in the same way as they were already being instrumented. Any ELASTIC_* environment variables and configuration options will continue to apply.