Switch branches/tags
Find file History
hkaj Enable docker config provider if docker.sock exists (#2544)
* Enable docker config provider if docker.sock exists

Signed-off-by: Haissam Kaj <haissam.kaj@gmail.com>

* add reno
Latest commit 91f62af Nov 12, 2018


Agent 6 docker image

This is how the official agent 6 image available here is built.

How to run it

Head over to datadoghq.com to get the official installation guide.

For a simple docker run, you can quickly get started with:

docker run -d -v /var/run/docker.sock:/var/run/docker.sock:ro \
              -v /proc/:/host/proc/:ro \
              -v /sys/fs/cgroup/:/host/sys/fs/cgroup:ro \
              -e DD_API_KEY=<YOUR_API_KEY> \

Environment variables

The agent is highly customizable, here are the most used environment variables:

Global options

  • DD_API_KEY: your API key (required)
  • DD_HOSTNAME: hostname to use for metrics (if autodetection fails)
  • DD_TAGS: host tags, separated by spaces. For example: simple-tag-0 tag-key-1:tag-value-1
  • DD_CHECK_RUNNERS: the agent runs all checks in sequence by default (default value = 1 runner). If you need to run a high number of checks (or slow checks) the collector-queue component might fall behind and fail the healthcheck. You can increase the number of runners to run checks in parallel


Starting with Agent v6.4.0, the agent proxy settings can be overridden with the following environment variables:

  • DD_PROXY_HTTP: an http URL to use as a proxy for http requests.
  • DD_PROXY_HTTPS: an http URL to use as a proxy for https requests.
  • DD_PROXY_NO_PROXY: a space-separated list of URLs for which no proxy should be used.

Note: At the moment, the trace-agent only supports setting a proxy in datadog.yaml, and does not support these proxy environment variables. If you wish to set a proxy for trace-agent, please refer to the section on mounting a custom datadog.yaml.

For more information: https://docs.datadoghq.com/agent/proxy/#agent-v6

Optional collection agents

These features are disabled by default for security or performance reasons, you need to explicitly enable them:

  • DD_APM_ENABLED: run the trace-agent along with the infrastructure agent, allowing the container to accept traces on 8126/tcp
  • DD_LOGS_ENABLED: run the log-agent along with the infrastructure agent. See below for details
  • DD_PROCESS_AGENT_ENABLED: enable live process collection in the process-agent. The Live Container View is already enabled by default if the Docker socket is available

Dogstatsd (custom metrics)

Send custom metrics via the statsd protocol:

  • DD_DOGSTATSD_NON_LOCAL_TRAFFIC: listen to dogstatsd packets from other containers, required to send custom metrics
  • DD_HISTOGRAM_PERCENTILES: histogram percentiles to compute, separated by spaces. The default is "0.95"
  • DD_HISTOGRAM_AGGREGATES: histogram aggregates to compute, separated by spaces. The default is "max median avg count"
  • DD_DOGSTATSD_SOCKET: path to the unix socket to listen to. Must be in a rw mounted volume.
  • DD_DOGSTATSD_ORIGIN_DETECTION: enable container detection and tagging for unix socket metrics. Running in host PID mode (e.g. with --pid=host) is required.


We automatically collect common tags from Docker, Kubernetes, ECS, Swarm, Mesos, Nomad and Rancher, and allow you to extract even more tags with the following options:

  • DD_DOCKER_LABELS_AS_TAGS : extract docker container labels
  • DD_DOCKER_ENV_AS_TAGS : extract docker container environment variables
  • DD_KUBERNETES_POD_LABELS_AS_TAGS : extract pod labels
  • DD_KUBERNETES_POD_ANNOTATIONS_AS_TAGS : extract pod annotations

You can either define them in your custom datadog.yaml, or set them as JSON maps in these envvars. The map key is the source (label/envvar) name, and the map value the Datadog tag name.


You can use shell patterns in label names to define simple rules for mapping labels to Datadog tag names using the same simple template system used by Autodiscovery. This is only supported by DD_KUBERNETES_POD_LABELS_AS_TAGS.

To add all pod labels as tags to your metrics where tags names are prefixed by kube_, you can use the following:


To add only pod labels as tags to your metrics that start with app, you can use the following:


Using secret files (BETA)

Integration credentials can be stored in Docker / Kubernetes secrets and used in Autodiscovery templates. See the setup instructions for the helper script and the agent documentation for more information.

Ignore containers

You can exclude containers from the metrics collection and autodiscovery, if these are not useful for you. We already exclude Kubernetes and OpenShift pause containers by default. See the datadog.yaml.example file for more documentation, and examples.

  • DD_AC_INCLUDE: whitelist of containers to always include
  • DD_AC_EXCLUDE: blacklist of containers to exclude

The format for these option is space-separated strings. For example, if you only want to monitor two images, and exclude the rest, specify:

DD_AC_EXCLUDE = "image:.*"
DD_AC_INCLUDE = "image:cp-kafka image:k8szk"

Please note that the docker.containers.running, .stopped, .running.total and .stopped.total metrics are not affected by these settings and always count all containers. This does not affect your per-container billing.

Datadog Cluster Agent

The DCA is a beta feature, if you are facing any issues please reach out to our support team Starting with Agent v6.3.2, you can use the Datadog Cluster Agent.

Cluster level features are now handled by the cluster agent, and you will find a [DCA] notation next to the affected features. Please refer to the below user documentation as well as the technical documentation here for further details on the instrumentation.

Kubernetes integration

Please refer to the dedicated section about the Kubernetes integration for more details.

  • DD_KUBERNETES_COLLECT_METADATA_TAGS: configures the agent to collect Kubernetes metadata (service names) as tags.
  • DD_KUBERNETES_METADATA_TAG_UPDATE_FREQ: set the collection frequency in seconds for the Kubernetes metadata (service names) from the API Server (or the Datadog Cluster Agent if enabled).
  • DD_COLLECT_KUBERNETES_EVENTS [DCA]: configures the cluster agent to collect Kubernetes events. See Event collection for more details.
  • DD_LEADER_ELECTION [DCA]: activates the leader election. Will be activated if the DD_COLLECT_KUBERNETES_EVENTS is set to true. The expected value is a bool: true/false.
  • DD_LEADER_LEASE_DURATION [DCA]: only used if the leader election is activated. See the details here. The expected value is a number of seconds.
  • DD_KUBE_RESOURCES_NAMESPACE [DCA]: configures the namespace where the Cluster Agent creates the configmaps required for the Leader Election, the Event Collection (optional) and the Horizontal Pod Autoscaling.


  • DD_JMX_CUSTOM_JARS: space-separated list of custom jars to load in jmxfetch (only for the -jmx variants)
  • DD_ENABLE_GOHAI: enable or disable the system information collector gohai (enabled by default if not set)
  • DD_EXPVAR_PORT: change the port for fetching expvar public variables from the agent. (defaults to 5000, you may then also have to change the agent_stat.yaml)

Some options are not yet available as environment variable bindings. To customize these, the agent supports mounting a custom /etc/datadog-agent/datadog.yaml configuration file (based on the docker or kubernetes base configurations) for these options, and using environment variables for the rest.

Optional volumes

To run custom checks and configurations without building your own image, you can mount additional files in these folders:

  • /checks.d/ : custom checks in this folder will be copied over and used, if a corresponding configuration is found
  • /conf.d/ : check configurations and Autodiscovery templates in this folder will be copied over in the agent's configuration folder. You can mount a host folder, kubernetes configmaps, or other volumes. Note: autodiscovery templates now are directly stored in the main conf.d folder, not in an auto_conf subfolder.

Going further

For more information about the container's lifecycle, see SUPERVISION.md.


Without the DCA

This sub-section is only valid for the agent versions < 6.3.2 or when not using the Datadog Cluster Agent.

To deploy the Agent in your Kubernetes cluster, you can use the manifest in manifests. Firstly, make sure you have the correct RBAC in place. You can use the files in manifests/rbac that contain the minimal requirements to run the Kubernetes Cluster level checks and perform the leader election. kubectl create -f manifests/rbac

Please note that with the above RBAC, every agent will have access to the API Server, to list the pods, services ... These accesses vanish when using the Datadog Cluster Agent. Indeed, the agents will only have access to the local kubelet and only the Cluster Agent will be able to access cluster level insight (nodes, services...).

Once the RBAC is in place, you can then create the agents with: kubectl create -f manifests/agent.yaml

The manifest for the agent has the KUBERNETES environment variable enabled, which will enable the collection of local kubernetes metrics via the kubelet's API. For the event collection and the API server check please read below. If you want the event collection to be resilient, you can create a ConfigMap datadogtoken that agents will use to save and share a state reflecting which events where pulled last.

To create such a ConfigMap, you can use the following command: kubectl create -f manifests/datadog_configmap.yaml See details in Event Collection.

With the DCA

This sub-section is only valid for agent versions > 6.3.2 and when using the Datadog Cluster Agent.

Event collection is handled by the cluster agent and the RBAC for the agent is slimmed down to the kubelet's API access. There is now a dedicated Clusterrole for the agent which should be as follows:

apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRole
  name: datadog-agent
- apiGroups:  # Kubelet connectivity
  - ""
  - nodes/metrics
  - nodes/spec
  - nodes/proxy
  - get

It goes along the ClusterRoleBinding and the Service Account, dedicated to the datadog-agents.

Event Collection [DCA]

Similarly to Agent 5, Agent 6 collects events from the Kubernetes API server.

1/ Set the collect_kubernetes_events variable to true in the datadog.yaml file, you can use the environment variable DD_COLLECT_KUBERNETES_EVENTS for this. 2/ Give the agents proper RBACs to activate this feature. See the RBAC section. 3/ A ConfigMap can be used to store the event.tokenKey and the event.tokenTimestamp. It has to be deployed in the default namespace and be named datadogtoken. Run kubectl create configmap datadogtoken --from-literal="event.tokenKey"="0" . You can also use the example in [manifests/datadog_configmap.yaml][https://github.com/DataDog/datadog-agent/blob/master/Dockerfiles/manifests/datadog_configmap.yaml].

Note: When the ConfigMap is used, if the agent in charge (via the Leader election) of collecting the events dies, the next leader elected will use the ConfigMap to identify the last events pulled. This is in order to avoid duplicate the events collected, as well as putting less stress on the API Server.

Leader Election [DCA]

Datadog Agent 6 supports built in leader election option for the Kubernetes event collector and the Kubernetes cluster related checks (i.e. Controle Plane service check).

This feature relies on Endpoints, you can enable it by setting the DD_LEADER_ELECTION environment variable to true the Datadog Agents will need to have a set of actions allowed prior to its deployment nevertheless. See the RBAC section for more details and keep in mind that these RBAC entities will need to be created before the option is set.

Agents coordinate by performing a leader election among members of the Datadog DaemonSet through kubernetes to ensure only one leader agent instance is gathering events at a given time.

This functionality is disabled by default, enabling the event collection will activate it (see Event collection) to avoid duplicating collecting events and stress on the API server. The leaderLeaseDuration is the duration for which a leader stays elected. It should be > 30 seconds and is 60 seconds by default. The longer it is, the less frequently your agents hit the apiserver with requests, but it also means that if the leader dies (and under certain conditions), events can be missed until the lease expires and a new leader takes over. It can be configured with the environment variable DD_LEADER_LEASE_DURATION.


If you are using the DCA, find all the RBAC for the agent as well as the Cluster agent here

In the context of using the Kubernetes integration, and when deploying agents in a Kubernetes cluster, a set of rights are required for the agent to integrate seamlessly.

You will need to allow the agent to be allowed to perform a few actions:

  • get and update of the Configmaps named datadogtoken to update and query the most up to date version token corresponding to the latest event stored in ETCD.
  • list and watch of the Events to pull the events from the API Server, format and submit them.
  • get, update and create for the Endpoint. The Endpoint used by the agent for the Leader election feature is named datadog-leader-election.
  • list the componentstatuses resource, in order to submit service checks for the Controle Plane's components status.

You can find the templates in manifests/rbac here. This will create the Service Account in the default namespace, a Cluster Role with the above rights and the Cluster Role Binding.

Node label collection

The agent can collect node labels from the APIserver and report them as host tags. This feature is disabled by default, as it is usually redundant with cloud provider host tags. If you need to do so, you can provide a node label -> host tag mapping in the DD_KUBERNETES_NODE_LABELS_AS_TAGS environment variable. The format is the inline JSON described in the tagging section.

Kubernetes node name as aliases

By default, the agent is using the kubernetes node name as an alias that can be used to forward metrics and events. This allows to submit events and metrics from remote hosts. However, if you have several clusters where some nodes could have similar node names, some host alias collisions could occur. To prevent those, the agent supports the use of a cluster-unique identifier (such as the actual cluster name), through the environment variable DD_CLUSTER_NAME. That identifier will be added to the node name as a host alias, and avoid collision issues altogether.

Legacy Kubernetes Versions

Our default configuration targets Kubernetes 1.7.6 and later, as we rely on features and endpoints introduced in this version. More installation steps are required for older versions:

  • RBAC objects (ClusterRoles and ClusterRoleBindings) are available since Kubernetes 1.6 and OpenShift 1.3, but are available under different apiVersion prefixes:
    • rbac.authorization.k8s.io/v1 in Kubernetes 1.8+ (and OpenShift 3.9+), the default apiVersion we target
    • rbac.authorization.k8s.io/v1beta1 in Kubernetes 1.5 to 1.7 (and OpenShift 3.7)
    • v1 in Openshift 1.3 to 3.6

You can apply our yaml manifests with the following sed invocations:

sed "s%authorization.k8s.io/v1%authorization.k8s.io/v1beta1%" clusterrole.yaml | kubectl apply -f -
sed "s%authorization.k8s.io/v1%authorization.k8s.io/v1beta1%" clusterrolebinding.yaml | kubectl apply -f -

or for Openshift 1.3 to 3.6:

sed "s%rbac.authorization.k8s.io/v1%v1%" clusterrole.yaml | oc apply -f -
sed "s%rbac.authorization.k8s.io/v1%v1%" clusterrolebinding.yaml | oc apply -f -
  • The kubelet check retrieves metrics from the Kubernetes 1.7.6+ (OpenShift 3.7.0+) prometheus endpoint. You need to enable cAdvisor port mode for older versions.

  • Our default daemonset makes use of the downward API to pass the kubelet's IP to the agent. This only works on versions 1.7 and up. For older versions, here are other ways to enable kubelet connectivity:

    • On versions 1.6, use fieldPath: spec.nodeName and make sure your node name is resolvable and reachable from the pod
    • If DD_KUBERNETES_KUBELET_HOST is unset, the agent will retrieve the node hostname from docker and try to connect there. See docker info | grep "Name:" and make sure the name is resolvable and reachable
    • If the IP of the docker default gateway is constant across your cluster, you can directly pass that IP in the DD_KUBERNETES_KUBELET_HOST envvar. You can retrieve the IP with the ip addr show | grep docker0 command.
  • Our default configuration relies on bearer token authentication to the APIserver and kubelet. On 1.3, the kubelet does not support bearer token auth, you will need to setup client certificates for the datadog-agent serviceaccount and pass them to the agent.

Log collection

The Datadog Agent can collect logs from containers starting at the version 6. Two installations are possible:

  • on the host: where the agent is external to the Docker environment
  • or by deploying its containerized version in the Docker environment


First let’s create two directories on the host that we will later mount on the containerized agent:

  • /opt/datadog-agent/run: to make sure we do not lose any logs from containers during restarts or network issues we store on the host the last line that was collected for each container in this directory
  • /opt/datadog-agent/conf.d: this is where you will provide your integration instructions. Any configuration file added there will automatically be picked up by the containerized agent when restarted. For more information about this check here.

To run a Docker container which embeds the Datadog Agent to monitor your host use the following command:

docker run -d --name dd-agent -h `hostname` -e DD_API_KEY=<YOUR_API_KEY> -e DD_LOGS_ENABLED=true -v /var/run/docker.sock:/var/run/docker.sock:ro -v /proc/:/host/proc/:ro -v /opt/datadog-agent/run:/opt/datadog-agent/run:rw -v /sys/fs/cgroup/:/host/sys/fs/cgroup:ro -v /opt/datadog-agent/conf.d:/conf.d:ro datadog/agent:latest

Important notes:

  • The Docker integration is enabled by default, as well as autodiscovery in auto config mode ((remove the listeners: -docker section in datadog.yaml to disable it).

  • You can find here the list of available images for agent 6 and we encourage you to always pick the latest version.

The parameters specific to log collection are the following:

  • -e DD_LOGS_ENABLED=true: this parameter enables the log collection when set to true. The agent now looks for log instructions in configuration files.
  • -v /opt/datadog-agent/run:/opt/datadog-agent/run:rw : mount the directory we created to store pointer on each container logs to make sure we do not lose any.
  • -v /opt/datadog-agent/conf.d:/conf.d:ro : mount the configuration directory we previously created to the container

Configuration file example

Now that the agent is ready to collect logs, you need to define which containers you want to follow. To start collecting logs for a given container filtered by image or label, you need to update the log section in an integration or custom .yaml file. Add a new yaml file in the conf.d directory with the following parameters:



#Log section

   - type: docker
     image: my_image_name    #or label: mylabel
     service: my_application_name
     source: java #tells Datadog what integration it is
     sourcecategory: sourcecode

For more examples of configuration files or agent capabilities (such as filtering, redacting, multiline, …) read this documentation.

How to build this image

On debian-based systems

You can build your own debian package using inv agent.omnibus-build

Then you can call inv agent.image-build that will take the debian package generated above and use it to build the image

On other systems

To build the image you'll need the agent debian package that can be found on this APT listing here.

You'll need to download one of the datadog-agent*_amd64.deb package in this directory, it will then be used by the Dockerfile and installed within the image.

You can then build the image using docker build -t datadog/agent:master .

To build the jmx variant, add --build-arg WITH_JMX=true to the build command