How to inject the library locally, without touching the application code at all, varies depending on where and how your Agent and application are installed. Select the scenario that represents your environment:
With the Admission Controller approach, the Agent uses the Kubernetes Admission Controller to intercept requests to the Kubernetes API and mutate new pods to inject the specified instrumentation library.
Library injection is applied on new pods only and does not have any impact on running pods.
Datadog Admission Controller enabled. Note: In Helm chart v2.35.0 and later, Datadog Admission Controller is activated by default in the Cluster Agent.
For Python, uWSGI applications are not supported at this time.
For Ruby, library injection support is in Preview. Instrumentation is only supported for Ruby on Rails applications with Bundler version greater than 2.3 and without vendored gems (deployment mode or BUNDLE_PATH).
Applications in Java, JavaScript, Python, .NET, or Ruby deployed on Linux with a supported architecture. Check the corresponding container registry for the complete list of supported architectures by language.
Docker Hub is subject to image pull rate limits. If you are not a Docker Hub customer, Datadog recommends that you update your Datadog Agent and Cluster Agent configuration to pull from GCR or ECR. For instructions, see Changing your container registry.
Datadog publishes instrumentation libraries images on gcr.io, Docker Hub, and Amazon ECR:
The DD_ADMISSION_CONTROLLER_AUTO_INSTRUMENTATION_CONTAINER_REGISTRY environment variable in the Datadog Cluster Agent configuration specifies the registry used by the Admission Controller. The default value is gcr.io/datadoghq.
You can pull the tracing library from a different registry by changing it to docker.io/datadog, public.ecr.aws/datadog, or another URL if you are hosting the images in a local container registry.
For your Kubernetes applications whose traces you want to send to Datadog, configure the Datadog Admission Controller to inject Java, JavaScript, Python, .NET or Ruby instrumentation libraries automatically. From a high level, this involves the following steps, described in detail below:
Enable Datadog Admission Controller to mutate your pods.
Annotate your pods to select which instrumentation library to inject.
Tag your pods with Unified Service Tags to tie Datadog telemetry together and navigate seamlessly across traces, metrics, and logs with consistent tags.
Apply your new configuration.
You do not need to generate a new application image to inject the library. The library injection is taken care of adding the instrumentation library, so no change is required in your application image.
By default, Datadog Admission controller mutates only pods labeled with a specific label. To enable mutation on your pods, add the label admission.datadoghq.com/enabled: "true" to your pod spec.
Note: You can configure Datadog Admission Controller to enable injection config without having this pod label by configuring the Cluster Agent with clusterAgent.admissionController.mutateUnlabelled (or DD_ADMISSION_CONTROLLER_MUTATE_UNLABELLED) to true.
apiVersion:apps/v1kind:Deploymentmetadata:labels:# (...)spec:template:metadata:labels:admission.datadoghq.com/enabled:"true"# Enable Admission Controller to mutate new pods part of this deploymentspec:containers:- # (...)
Note: For .NET library injection on musl-based Linux distributions (such as Alpine), append -musl to the pod annotation tag for versions 2.55 and earlier. For example, use v2.29.0-musl instead of v2.29.0. The -musl suffix is not required for versions after 2.55, including v3.0+.
Note: If you already have an application instrumented using version X of the library, and then use library injection to instrument using version Y of the same tracer library, the tracer does not break. Rather, the library version loaded first is used. Because library injection happens at the admission controller level prior to runtime, it takes precedence over manually configured libraries.
Note: Using the latest tag is supported, but use it with caution because major library releases can introduce breaking changes.
For example, to inject a Java library:
apiVersion:apps/v1kind:Deploymentmetadata:labels:# (...)spec:template:metadata:labels:admission.datadoghq.com/enabled:"true"# Enable Admission Controller to mutate new pods in this deploymentannotations:admission.datadoghq.com/java-lib.version:"<CONTAINER IMAGE TAG>"spec:containers:- # (...)
With Unified Service Tags, you can tie Datadog telemetry together and navigate seamlessly across traces, metrics, and logs with consistent tags. Set the Unified Service Tagging on both the deployment object and the pod template specs.
Set Unified Service tags by using the following labels:
Note: It is not necessary to set the environment variables for universal service tagging (DD_ENV, DD_SERVICE, DD_VERSION) in the pod template spec, because the Admission Controller propagates the tag values as environment variables when injecting the library.
For example:
apiVersion:apps/v1kind:Deploymentmetadata:labels:tags.datadoghq.com/env:"prod"# Unified service tag - Deployment Env tagtags.datadoghq.com/service:"my-service"# Unified service tag - Deployment Service tagtags.datadoghq.com/version:"1.1"# Unified service tag - Deployment Version tag# (...)spec:template:metadata:labels:tags.datadoghq.com/env:"prod"# Unified service tag - Pod Env tagtags.datadoghq.com/service:"my-service"# Unified service tag - Pod Service tagtags.datadoghq.com/version:"1.1"# Unified service tag - Pod Version tagadmission.datadoghq.com/enabled:"true"# Enable Admission Controller to mutate new pods part of this deploymentannotations:admission.datadoghq.com/java-lib.version:"<CONTAINER IMAGE TAG>"spec:containers:- # (...)
Library injection leverages the injection of a dedicated init container in pods.
If the injection was successful you can see an init container called datadog-lib-init in your pod:
Or run kubectl describe pod <my-pod> to see the datadog-lib-init init container listed.
The instrumentation also starts sending telemetry to Datadog (for example, traces to APM).
Problem: The pod annotation for the dotnet library version included a -musl suffix, but the application container runs on a Linux distribution that uses glibc.
Solution: Remove the -musl suffix from the dotnet library version.
Problem: The application container runs on a Linux distribution that uses musl-libc (for example, Alpine), but the pod annotation does not include the -musl suffix.
Solution: Add the -musl suffix to the dotnet library version.
In Python < 1.20.3, Python injection logs output to stderr. Upgrade to 1.20.3 or above to suppress the logs by default. The logs can be enabled by setting the environment variable DD_TRACE_DEBUG to 1.
Problem: The ddtrace library is already installed on the system so the injection logic aborts injecting the library to avoid introducing a breaking change in the application.
Solution: Remove the installation of ddtrace if library injection is desired. Otherwise, use the installed library (see documentation) instead of library injection.
Problem: Since Cluster Agent v1.20, the API is only served using TLS v1.3 by default. If the Kubernetes cluster is configured with TLS v1.2 or older, library injection fails.
Solution: Set DD_CLUSTER_AGENT_ALLOW_LEGACY_TLS to true for Cluster Agent.
Tracing Library Injection on a host is in Preview.
When both the Agent and your services are running on a host, real or virtual, Datadog injects the tracing library by using a preload library that overrides calls to execve. Any newly started processes are intercepted and the specified instrumentation library is injected into the services.
If the host does not yet have a Datadog Agent installed, or if you want to upgrade your Datadog Agent installation, use the Datadog Agent install script to install both the injection libraries and the Datadog Agent:
By default, running the script installs support for Java, Node.js, Python, Ruby, and .NET all pinned to the latest major version. If you want to specify which language support is installed, also set the DD_APM_INSTRUMENTATION_LIBRARIES environment variable. The valid values are java, js, python, dotnet, and php. Use a comma-separated list to specify more than one language:
The following environment variables configure library injection. You can pass these in by export through the command line (export DD_CONFIG_SOURCES=BASIC), shell configuration, or launch command.
Each of the fields in the config file corresponds to an environment variable. This environment variable is read from the environment of the process that’s being launched and affects only the process currently being launched.
Config file property
Environment Variable
log_level
DD_APM_INSTRUMENTATION_DEBUG
output_paths
DD_APM_INSTRUMENTATION_OUTPUT_PATHS
env
DD_ENV
The DD_APM_INSTRUMENTATION_DEBUG environment variable is limited to the values true and false (default value false). Setting it to true sets log_level to debug and setting it to false (or not setting it at all) uses the log_level specified in the configuration file. The environment variable can only set the log level to debug, not any other log level values.
The DD_INSTRUMENT_SERVICE_WITH_APM environment variable controls whether or not injection is enabled. It defaults to TRUE. Set it to FALSE to turn off library injection altogether.
By default, the following settings are enabled in an instrumented process:
Tracing
Log injection, assuming the application uses structured logging (usually JSON). For traces to appear in non-structured logs, you must change your application’s log configuration to include placeholders for trace ID and span ID. See Connect Logs and Traces for more information.
Health metrics
Runtime metrics
Exercise your application to start generating telemetry data, which you can see as traces in APM.
Tracing Library Injection on hosts and containers is in Preview.
When your Agent is running on a host, and your services are running in containers, Datadog injects the tracing library by intercepting container creation and configuring the Docker container.
Any newly started processes are intercepted and the specified instrumentation library is injected into the services.
If the host does not yet have a Datadog Agent installed, or if you want to upgrade your Datadog Agent installation, use the Datadog Agent install script to install both the injection libraries and the Datadog Agent:
By default, running the script installs support for Java, Node.js, Python, Ruby, and .NET. If you want to specify which language support is installed, also set the DD_APM_INSTRUMENTATION_LIBRARIES environment variable. The valid values are java, js, python, dotnet, and php. Use a comma-separated list to specify more than one language:
If the environment variables DD_ENV, DD_SERVICE, or DD_VERSION are specified in a service container image, those values are used to tag telemetry from the container.
If they are not specified, DD_ENV uses the env value set in the /etc/datadog-agent/inject/docker_config.yaml config file, if any. DD_SERVICE is derived from the name of the Docker image. An image with the name my-service:1.0 is tagged with DD_SERVICE of my-service.
Start your Agent and launch your containerized services as usual.
Exercise your application to start generating telemetry data, which you can see as traces in APM.
Tracing Library Injection in containers is in Preview.
When your Agent and services are running in separate Docker containers on the same host, Datadog injects the tracing library by intercepting container creation and configuring the Docker container.
Any newly started processes are intercepted and the specified instrumentation library is injected into the services.
This installs language libraries for all supported languages. To install specific languages, set the DD_APM_INSTRUMENTATION_LIBRARIES variable. The valid values are java, js, python, dotnet, and php:
In the Docker compose file that launches your containers, use the following settings for the Agent, securely setting your own Datadog API key for ${DD_API_KEY}:
If the environment variables DD_ENV, DD_SERVICE, or DD_VERSION are specified in a service container image, those values are used to tag telemetry from the container.
If they are not specified, DD_ENV uses the env value set in the /etc/datadog-agent/inject/docker_config.yaml config file, if any. DD_SERVICE is derived from the name of the Docker image. An image with the name my-service:1.0 is tagged with DD_SERVICE of my-service.
The supported features and configuration options for the tracing library are the same for library injection as for other installation methods, and can be set with environment variables. Read the Datadog library configuration page for your language for more details.
