opentelemetry

Instrumenting Python applications with EDOT SDKs on Kubernetes

This document focuses on instrumenting Python applications on Kubernetes, using the OpenTelemetry Operator, Elastic Distribution of OpenTelemetry (EDOT) Collectors, and the EDOT Python SDK.

• For general knowledge about the EDOT Python SDK, refer to the getting started guide.

• For Python auto-instrumentation specifics, refer to OpenTelemetry Operator Python auto-instrumentation.

• To manually instrument your Python application code (by customizing spans and metrics), refer to EDOT Python manual instrumentation.

• For general information about instrumenting applications on kubernetes, refer to instrumenting applications.

Supported environments and configuration

• EDOT Python container image supports glibc and musl based auto-instrumentation for Python 3.12.

• musl based containers instrumentation requires an extra annotation and operator v0.113.0+.

• To enable logs auto-instrumentation, refer to auto-instrument python logs.

• To disable specific instrumentation libraries, refer to excluding auto-instrumentation.

• For a full list of configuration options, refer to Python specific configuration.

• For Python specific limitations when using the OpenTelemetry operator, refer to Python-specific topics.

Instrument a Python app with EDOT Python SDK on Kubernetes

In this example, you’ll learn how to:

• Enable auto-instrumentation of a Python application using one of the following supported methods:
  • Adding an annotation to the deployment Pods.
  • Adding an annotation to the namespace.
• Verify that auto-instrumentation libraries are injected and configured correctly.
• Confirm data is flowing to Kibana Observability.

For this example, we assume the application you’re instrumenting is a deployment named python-app running in the python-ns namespace.

1. Ensure you have successfully installed the OpenTelemetry Operator, and confirm that the following Instrumentation object exists in the system:

$ kubectl get instrumentation -n opentelemetry-operator-system
NAME                      AGE    ENDPOINT
elastic-instrumentation   107s   http://opentelemetry-kube-stack-daemon-collector.opentelemetry-operator-system.svc.cluster.local:4318

[!NOTE] If your Instrumentation object has a different name or is created in a different namespace, you will have to adapt the annotation value in the next step.

1. Enable auto-instrumentation of the Python application using one of the following methods:

• Edit your application workload definition and include the annotation under spec.template.metadata.annotations:

  spec:
  ...
    template:
      metadata:
        labels:
          app: python-app
        annotations:
          instrumentation.opentelemetry.io/inject-python: opentelemetry-operator-system/elastic-instrumentation
  ...
  

• Alternatively, add the annotation at namespace level to apply auto-instrumentation in all Pods of the namespace:

  kubectl annotate namespace python-ns instrumentation.opentelemetry.io/inject-python=opentelemetry-operator-system/elastic-instrumentation
  

1. Restart application:

Once the annotation has been set, restart the application to create new Pods and inject the instrumentation libraries:

```bash
kubectl rollout restart deployment python-app -n python-ns
```

1. Verify the auto-instrumentation resources are injected in the Pod:

Run a kubectl describe of one of your application pods and check:

• There should be an init container named opentelemetry-auto-instrumentation-python in the Pod:

  $ kubectl describe pod python-app-8d84c47b8-8h5z2 -n python-ns
  ...
  ...
  Init Containers:
    opentelemetry-auto-instrumentation-python:
      Container ID:  containerd://fdc86b3191e34ef5ec872853b14a950d0af1e36b0bc207f3d59bd50dd3caafe9
      Image:         docker.elastic.co/observability/elastic-otel-python:0.3.0
      Image ID:      docker.elastic.co/observability/elastic-otel-python@sha256:de7b5cce7514a10081a00820a05097931190567ec6e18a384ff7c148bad0695e
      Port:          <none>
      Host Port:     <none>
      Command:
        cp
        -r
        /autoinstrumentation/.
        /otel-auto-instrumentation-python
      State:          Terminated
        Reason:       Completed
  ...
  

• The main container has new environment variables, including PYTHONPATH:

  ...
  Containers:
    python-app:
  ...
      Environment:
  ...
        PYTHONPATH:                          /otel-auto-instrumentation-python/opentelemetry/instrumentation/auto_instrumentation:/otel-auto-instrumentation-python
        OTEL_EXPORTER_OTLP_PROTOCOL:         http/protobuf
        OTEL_TRACES_EXPORTER:                otlp
        OTEL_METRICS_EXPORTER:               otlp
        OTEL_SERVICE_NAME:                   python-app
        OTEL_EXPORTER_OTLP_ENDPOINT:         http://opentelemetry-kube-stack-daemon-collector.opentelemetry-operator-system.svc.cluster.local:4318
  ...
  

• The Pod has an EmptyDir volume named opentelemetry-auto-instrumentation-python mounted in both the main and the init containers in path /otel-auto-instrumentation-python:

  Init Containers:
    opentelemetry-auto-instrumentation-python:
  ...
      Mounts:
        /otel-auto-instrumentation-python from opentelemetry-auto-instrumentation-python (rw)
  Containers:
    python-app:
  ...
      Mounts:
        /otel-auto-instrumentation-python from opentelemetry-auto-instrumentation-python (rw)
  ...
  Volumes:
  ...
    opentelemetry-auto-instrumentation-python:
      Type:        EmptyDir (a temporary directory that shares a pod's lifetime)
  

Ensure the environment variable OTEL_EXPORTER_OTLP_ENDPOINT points to a valid endpoint and there’s network communication between the Pod and the endpoint.

1. Confirm data is flowing to Kibana:

• Open Observability -> Applications -> Service inventory, and determine if:
  • The application appears in the list of services.
  • The application shows transactions and metrics.
  • If python logs instrumentation is enabled, the application logs should appear in the Logs tab.
• For application container logs, open Kibana Discover and filter for your Pods’ logs. In the provided example, we could filter for them with either of the following:
  • k8s.deployment.name: "python-app" (adapt the query filter to your use case)
  • k8s.pod.name: python-app* (adapt the query filter to your use case)

Note that the container logs are not provided by the instrumentation library, but by the DaemonSet collector deployed as part of the operator installation.

Troubleshooting

• Refer to troubleshoot auto-instrumentation for further analysis.