Hints annotations based autodiscover

Standalone Elastic Agent supports autodiscover based on hints from the provider. The hints mechanism looks for hints in Kubernetes Pod annotations that have the prefix co.elastic.hints. As soon as the container starts, Elastic Agent checks it for hints and launches the proper configuration for the container. Hints tell Elastic Agent how to monitor the container by using the proper integration. This is the full list of supported hints:

The package to use for monitoring.

The host to use for metrics retrieval. If not defined, the host will be set as the default one: <pod-ip>:<container-port>.

The list of data streams to enable. If not specified, the integration’s default data streams are used. To find the defaults, refer to the Elastic integrations documentation.

If data streams are specified, additional hints can be defined per data stream. For example, co.elastic.hints/info.period: 5m if the data stream specified is info for the Redis module.

apiVersion: v1
kind: Pod
metadata:
  name: redis
  annotations:
    co.elastic.hints/package: redis
    co.elastic.hints/data_streams: info
    co.elastic.hints/info.period: 5m

If data stream hints are not specified, the top level hints will be used in its configuration.

The path to retrieve the metrics from.

The time interval for metrics retrieval, for example, 10s.

Metrics retrieval timeout, for example, 3s.

The username to use for authentication.

The password to use for authentication. It is recommended to retrieve this sensitive information from an ENV variable and avoid placing passwords in plain text.

The stream to use for logs collection, for example, stdout/stderr.

If the specified package has no logs support, a generic container’s logs input will be used as a fallback. See the Hints autodiscovery for kubernetes log collection example below.

Define a processor to be added to the input configuration. See Define processors for the list of supported processors.

If the processors configuration uses list data structure, object fields must be enumerated. For example, hints for the rename processor configuration below

processors:
  - rename:
      fields:
        - from: "a.g"
          to: "e.d"
      fail_on_error: true

will look like:

co.elastic.hints/processors.rename.fields.0.from: "a.g"
co.elastic.hints/processors.rename.fields.1.to: "e.d"
co.elastic.hints/processors.rename.fail_on_error: 'true'

If the processors configuration uses map data structure, enumeration is not needed. For example, the equivalent to the add_fields configuration below

processors:
  - add_fields:
      target: project
      fields:
        name: myproject

is

co.elastic.hints/processors.1.add_fields.target: "project"
co.elastic.hints/processors.1.add_fields.fields.name: "myproject"

In order to provide ordering of the processor definition, numbers can be provided. If not, the hints builder will do arbitrary ordering:

co.elastic.hints/processors.1.dissect.tokenizer: "%{key1} %{key2}"
co.elastic.hints/processors.dissect.tokenizer: "%{key2} %{key1}"

In the above sample the processor definition tagged with 1 would be executed first.

When a pod has multiple containers, the settings are shared unless you put the container name in the hint. For example, these hints configure processors.decode_json_fields for all containers in the pod, but set a specific stream hint for the container called sidecar.

annotations:
  co.elastic.hints/processors.decode_json_fields.fields: "message"
	co.elastic.hints/processors.decode_json_fields.add_error_key: true
	co.elastic.hints/processors.decode_json_fields.overwrite_keys: true
	co.elastic.hints/processors.decode_json_fields.target: "team"
	co.elastic.hints.sidecar/stream: "stderr"

The available packages that are supported through hints can be found here.

To enable hints autodiscovery, you must add hints.enabled: true to the provider’s configuration:

providers:
  kubernetes:
    hints.enabled: true

Then ensure that the proper volumes and volumeMounts are specified by uncommenting the appropriate sections in the Elastic Agent manifest:

volumeMounts:
- name: external-inputs
  mountPath: /etc/elastic-agent/inputs.d
...
volumes:
- name: external-inputs
  emptyDir: {}
...

An init container is also required to download the hints templates. The init container is already defined, so uncomment the respective section:

initContainers:
- name: k8s-templates-downloader
  image: busybox:1.28
  command: ['sh']
  args:
    - -c
    - >-
      mkdir -p /etc/elastic-agent/inputs.d &&
      wget -O - https://github.com/elastic/elastic-agent/archive/8.13.tar.gz | tar xz -C /etc/elastic-agent/inputs.d --strip=5 "elastic-agent-8.13/deploy/kubernetes/elastic-agent-standalone/templates.d"
  volumeMounts:
    - name: external-inputs
      mountPath: /etc/elastic-agent/inputs.d

Enabling hints allows users deploying Pods on the cluster to automatically turn on Elastic monitoring at Pod deployment time. For example, to deploy a Redis Pod on the cluster and automatically enable Elastic monitoring, add the proper hints as annotations on the Pod manifest file:

...
apiVersion: v1
kind: Pod
metadata:
  name: redis
  annotations:
    co.elastic.hints/package: redis
    co.elastic.hints/data_streams: info
    co.elastic.hints/host: '${kubernetes.pod.ip}:6379'
    co.elastic.hints/info.period: 5s
  labels:
    k8s-app: redis
    app: redis
...

After deploying this Pod, the data will start flowing in automatically. You can find it on the index metrics-redis.info-default.

The log collection for Kubernetes autodiscovered pods can be supported by using container_logs.yml template. Elastic Agent needs to emit a container_logs mapping so as to start collecting logs for all the discovered containers even if no annotations are present in the containers.

The previous default behaviour can be disabled with hints.default_container_logs: false. So this will disable the automatic logs collection from all discovered pods. Users need specifically to annotate their pod with following annotations:

annotations:
  co.elastic.hints/package: "container_logs"

providers.kubernetes:
  node: ${NODE_NAME}
  scope: node
  hints:
    enabled: true
    default_container_logs: false
...

In the following sample nginx manifest, we will additionally provide specific stream annotation, in order to configure the filestream input to read only stderr stream:

apiVersion: apps/v1
kind: Deployment
metadata:
  labels:
    app: nginx
  name: nginx
  namespace: default
spec:
  selector:
    matchLabels:
      app: nginx
  template:
    metadata:
      labels:
        app: nginx
      annotations:
        co.elastic.hints/package: "container_logs"
        co.elastic.hints/stream: "stderr"
    spec:
      containers:
      - image: nginx
        name: nginx
...

Users can monitor the final rendered Elastic Agent configuration:

kubectl exec -ti -n kube-system elastic-agent-7fkzm -- bash


/usr/share/elastic-agent# /elastic-agent inspect -v --variables --variables-wait 2s

inputs:
- data_stream.namespace: default
  id: hints-container-logs-3f69573a1af05c475857c1d0f98fc55aa01b5650f146d61e9653a966cd50bd9c-kubernetes-1780aca0-3741-4c8c-aced-b9776ba3fa81.nginx
  name: filestream-generic
  original_id: hints-container-logs-3f69573a1af05c475857c1d0f98fc55aa01b5650f146d61e9653a966cd50bd9c
  [output truncated ....]
  streams:
  - data_stream:
      dataset: kubernetes.container_logs
      type: logs
    exclude_files: []
    exclude_lines: []
    parsers:
    - container:
        format: auto
        stream: stderr
    paths:
    - /var/log/containers/*3f69573a1af05c475857c1d0f98fc55aa01b5650f146d61e9653a966cd50bd9c.log
    prospector:
      scanner:
        symlinks: true
    tags: []
  type: filestream
  use_output: default
outputs:
  default:
    hosts:
    - https://elasticsearch:9200
    password: changeme
    type: elasticsearch
    username: elastic
providers:
  kubernetes:
    hints:
      default_container_logs: false
      enabled: true
    node: control-plane
    scope: node

Based on the previous example, users might want to perform extra processing on specific logs, for example to decode specific fields containing JSON strings. Use of decode_json_fields is advisable as follows:

You need to have enabled hints autodiscovery, as described in the previous Hints autodiscovery for Kubernetes log collection example.

The pod that will produce JSON logs needs to be annotated with:

 annotations:
        co.elastic.hints/package: "container_logs"
        co.elastic.hints/processors.decode_json_fields.fields: "message"
        co.elastic.hints/processors.decode_json_fields.add_error_key: 'true'
        co.elastic.hints/processors.decode_json_fields.overwrite_keys: 'true'
        co.elastic.hints/processors.decode_json_fields.target: "team"

The following log entry:

{"myteam": "ole"}

Will produce both fields: the original message field and also the target field team.

"team": {
      "myteam": "ole"
    },

"message": "{\"myteam\": \"ole\"}",

When things do not work as expected, you may need to troubleshoot your setup. Here we provide some directions to speed up your investigation: