Honeycomb Kubernetes Agent | Honeycomb

Honeycomb Kubernetes Agent

To add tracing to your Kubernetes cluster, refer to the OpenTelemetry Operator for Kubernetes.

Honeycomb integrates with Kubernetes to collect your applications' logs, cluster logs, and resource metrics so that you can answer all the questions you have.

The Honeycomb Kubernetes agent provides a flexible way to aggregate, structure, and enrich events from applications running on Kubernetes. The Honeycomb Kubernetes agent can also collect resource and status metrics from nodes, pods, containers, and volumes.


To use the Helm package, you will need Helm version 3 or better. You can still install the agent using kubectl though configuration may be more involved. You wil also need your Honeycomb API key.

Install using either Helm, or kubectl directly.

Getting Started Using Helm 

To add tracing to your Kubernetes cluster, refer to the OpenTelemetry Operator for Kubernetes.
By default, this Helm chart will collect metrics from all nodes and pods, and watchers configuration to capture logs from the following system components:

  • kube-controller-manager
  • kube-scheduler

To install the agent using the Helm package manager:

  1. Add the Honeycomb Helm repository
helm repo add honeycomb https://honeycombio.github.io/helm-charts
  1. Install the agent
helm install honeycomb honeycomb/honeycomb --set honeycomb.apiKey=YOUR_API_KEY
For more details about advanced configuration options, refer to the Helm chart.

Getting Started Using Kubectl 

To add tracing to your Kubernetes cluster, refer to the OpenTelemetry Operator for Kubernetes.

To install the agent using kubectl:

  1. Create a honeycomb namespace
kubectl create namespace honeycomb
  1. Store your Honeycomb API Key as a Kubernetes secret
kubectl create secret generic honeycomb --namespace honeycomb \
  1. Add the agent to your cluster
kubectl apply -f https://raw.githubusercontent.com/honeycombio/honeycomb-kubernetes-agent/main/examples/quickstart.yaml

Logs Collection 

The agent will collect logs based on configured watchers. These watchers will match pods based on the configured selection and parse the logs for the pod using the specified parser. The Helm chart exposes a watchers property that you can use. If installed using kubectl, you will need to modify the watchers section of the ConfigMap.

Using modified log watchers configuration 

A configuration to collect and parse logs from the Kubernetes controller manager and scheduler would look like this:

  - dataset: kubernetes-logs
    labelSelector: component=kube-controller-manager
    namespace: kube-system
    parser: glog
  - <dataset: kubernetes-logs
    labelSelector: component=kube-scheduler
    namespace: kube-system
    parser: glog

Capturing events with Heptio Eventrouter 

An optional configuration is to capture Kubernetes events using the Heptio Eventrouter component. Configure the Eventrouter to use the json sink for logs, and you can capture them with this watchers configuration:

  - dataset: k8s-eventrouter
    labelSelector: app=eventrouter
    namespace: olly
    parser: json
      - drop_field:
          field: old_event
You will want to configure the agent to parse logs from your applications, but that depends a lot on your applications! Read more about customizing the agent.

Metrics Collection 

The agent collects metrics from nodes, pods, containers, and volumes. For each collection interval, an event for each resource will be sent to Honeycomb that contains all collected metrics as fields to the event. Kubernetes labels for the resources will be added as fields to the event with a label. prefix. Pods and containers also get additional status and restart metrics. The default collection interval is every 10 seconds, for nodes and pods resources only. Follow details on how to configure the agent to modifying your configuration accordingly.

Metrics List 

Metrics collected depend on the Kubernetes resource type. You can filter metrics on the k8s.resource.type field.

metric node pod container volume
metrics.cpu.usage x x x
metrics.cpu.utilization x x x
metrics.filesystem.available x x x
metrics.filesystem.capacity x x x
metrics.filesystem.usage x x x
metrics.memory.available x x x
metrics.memory.major_page_faults x x x
metrics.memory.page_faults x x x
metrics.memory.rss x x x
metrics.memory.usage x x x
metrics.memory.utilization x x x
metrics.memory.working_set x x x
metrics.network.bytes.receive x x
metrics.network.bytes.send x x
metrics.network.errors.receive x x
metrics.network.errors.send x x
metrics.uptime x x x
metrics.volume.available x
metrics.volume.capacity x
metrics.volume.inodes.free x
metrics.volume.inodes.total x
metrics.volume.inodes.used x
metrics.volume.used x
status.exitcode x
status.message x x
status.phase x
status.ready x
status.reason x x
status.restart x x
status.restart_count x x
status.restart_delta x x
status.state x

How the Honeycomb Kubernetes Agent Works 

The Honeycomb Kubernetes Agent runs as a DaemonSet. That is, one copy of the agent runs on each node in the cluster.

Honeycomb Kubernetes Agent as DaemonSet

Metrics for resources running on the node are collected by communicating with the local node’s kubelet directly.

Logs from containers' stdout and stderr are written by the Docker daemon to the local node filesystem. The Honeycomb Kubernetes Agent reads these logs, augments them with metadata from the Kubernetes API, and ships them to Honeycomb so that you can observe what is going on.

Agent inside node

The Honeycomb Kubernetes Agent’s configuration file describes which pods' logs to process, and how to handle them, as well as which resource types to collect metrics from.


What metadata does the agent add to these logs?

  • pod.labels
  • pod.name
  • pod.namespace
  • pod.resourceVersion
  • pod.UID
  • pod.nodeName
  • pod.nodeSelector
  • pod.serviceAccountName
  • pod.subdomain
  • pod.annotations
  • container.args
  • container.command
  • container.name
  • container.env
  • container.image
  • container.ports
  • container.VolumeMounts
  • container.workingDir
  • container.resources

The above metadata is added by default. Node metadata can be added by specifying includeNodeLabels: true within the agent’s metrics configuration.

Modifying Configuration 

You can modify the agent’s configuration via the ConfigMap deployed. By default, Kubernetes resource metrics and logs from the controller and scheduler will be collected.

Applications in Kubernetes tend to use different logging formats. In our opinion, your own applications should use a structured, self-describing log format such as JSON. But Kubernetes system components use the glog format, reverse proxies and ingress controllers may use a combined log format, and so on.

You might also want to aggregate events only from specific services, rather than from everything that might be running in a cluster. Or you might want to send logs from different services to different datasets.

To accommodate these real-world use cases, you can customize the Honeycomb Kubernetes Agent’s behavior with a YAML configuration file. Ordinarily, you will create this file as a Kubernetes ConfigMap that will be mounted inside the agent container.

Metrics Collection Configuration 

Metrics collection can be configured using the metrics configuration property.

  enabled: true
  dataset: kubernetes-metrics
  clusterName: MyCluster
    - node
    - pod

A clusterName should be specified for each cluster. By default, only node and pod metrics will be collected.

The following table describes all properties for metrics configuration:

key required? type description
enabled yes bool Enables metrics to be collected and sent to Honeycomb. default: false
dataset yes string Name of dataset to sent events to. default: kubernetes-metrics
clusterName yes string Name of Kubernetes cluster. Will be emitted as a field to Honeycomb. default: k8s-cluster
interval no string Collection interval in time duration format, which is specified with a duration suffix. Valid time units are ns, us (or µs), ms, s, m, h. default: 10s
metricGroups no list Resource groups to collected metrics from. Valid values are: node, pod, container, volume. default: node, pod
omitLabels no list Labels in this list will not be collected and sent as fields to Honeycomb. default: nil
additionalFields no map A map of fields name and values to apply to each metric event. default: nil
includeNodeLabels no bool If enabled, attaches node metadata to metric events. Node labels will respect the omitLabels list.

Example Metrics Configuration 

This configuration will collect metrics from the node, pods, containers, and volumes, every 10 seconds. The auto generated controller-revision-hash label will be omitted, and additional fields for region and az will be added.

  enabled: true
  dataset: kubernetes-metrics
  clusterName: k8s-cluster
  interval: 10s
    - node
    - pod
    - container
    - volume
    - controller-revision-hash
    region: us-east
    az: us-east-1a

Log Watchers Configuration 

Logs parsing and collection can be configured using the watchers configuration property.

- labelSelector: "app=nginx"
  parser: nginx
  dataset: kubernetes-nginx

- labelSelector: "app=frontend"
  parser: json
  dataset: kubernetes-frontend

Each block in the watchers list describes a set of pods whose logs you want to handle in a specific way, and has the following keys:

key required? type description
labelSelector yes* string A Kubernetes label selector identifying the set of pods to watch.
parser yes string Describes how this watcher should parse events.
dataset yes string The dataset that this watcher should send events to.
containerName no string If you only want to consume logs from one container in a multi-container pod, the name of the container to watch.
processors no list A list of processors to apply to events after they are parsed
namespace no string The Kubernetes namespace the pods are located in. If not supplied, default namespace is used.
paths no string array Glob-style* paths to the log files. If not supplied, the default Kubernetes log paths and filenames are used.
exclude no string array Glob-style* paths for files to exclude from consideration. If a given file matches an exclude, it will not be watched. If not supplied, no files are excluded.

“Glob-style” means:

  • * matches any sequence of non-path-separators
  • /**/ matches zero or more directories
  • ? matches any single non-path-separator character

Validating a Configuration File 

To check a configuration file without needing to deploy it into the cluster, you can run the Honeycomb Kubernetes Agent container locally with the --validate flag:

docker run -v /FULL/PATH/TO/YOUR/config.yaml:/etc/honeycomb/config.yaml honeycombio/honeycomb-kubernetes-agent:head --validate

Uploading a Configuration File to a Cluster 

To make a configuration file visible to the Honeycomb Kubernetes Agent inside a Kubernetes cluster, you will need to create a Kubernetes ConfigMap from it.

To create a brand-new ConfigMap from a local file config.yaml, run:

kubectl create configmap honeycomb-agent-config --from-file=config.yaml

To replace an existing ConfigMap, you can run:

kubectl create configmap honeycomb-agent-config \
    --from-file=config.yaml --output=yaml \
    --dry-run | kubectl replace --filename=-

Then restart running agent pods:

kubectl delete pod --selector k8s-app=honeycomb-agent


Currently, the following parsers are supported:


Does no parsing on logs, and submits an event with the entire contents of the log line in a "log" field, plus the aforementioned kubernetes metadata. Use this if you just want the “raw” log line, or if your log line structure does not match one of the parsers below. You can still query datasets with raw log lines to some degree using string filters and derived columns, but structuring your logs is strongly encouraged.


JSON is a great format for structured logs. With the JSON parser, we map JSON key/value pairs to event fields.


Parses NGINX access logs.

If you are using a custom NGINX log format, you can specify the format using the following configuration:

- labelSelector: "io.kompose.service=nginx"
    name: nginx
    dataset: nginx-inner
      log_format: '$remote_addr - $remote_user [$time_local] $host "$request" $status $bytes_sent $body_bytes_sent $request_time "$http_referer" "$http_user_agent" $request_length "$http_authorization" "$http_x_forwarded_proto" "$http_x_forwarded_for" $server_name'
This uses the enhanced additional log fields from our using NGINX with Honeytail guide. You may need to modify the log format in the watcher YAML, or your NGINX config file / ConfigMap, to match.


Parses logs produced by glog, which look like this:

I0719 23:09:54.422170       1 kube.go:118] Node controller sync successful

This format is commonly used by Kubernetes system components, such as the API server.


Parses logs produced by redis 3.0+, which look like this:

1:M 08 Aug 22:59:58.739 * Background saving started by pid 43

Thanks to MacRae Linton for contributing the Redis parser.


Parses logs in key=value format, such as:

time=2022-05-15T05:43:19Z msg="server response - time 12ms code: 401 - request: GET /hello ..."

Key-value formatted logs often have a special prefix, such as a log level.

INFO: time=2022-05-15T05:43:19Z msg="server response - time 12ms code: 401 - request: GET /hello ..."

For parsing lines that have additional fields without an equal sign, and to ensure fields are extracted properly into fields in the Honeycomb UI, specify a regular expression to parse that prefix in the configuration.

- labelSelector: "com.myco.logging.keyvalformat=true"
    name: keyval
      prefixRegex: "(?P<loglevel>[A-Z]+): (?P<timestamp>time=[0-9]{4}-[0-9]{2}-[0-9]{2}T[0-9]{2}:[0-9]{2}:[0-9]{2}Z )"

If whitespace-separated text is found that does not contain =, then it will be interpreted as a key without a value. For example, if a log contains text:

This is a test

Then, the keyval parser will create an event that looks similar to:

  "This": true,
  "is": true,
  "a": true,
  "test": true

This will cause an explosion of columns in Honeycomb. As this is a deliberate choice by the keyval parser, extra care should be taken when using it.


Processors transform events after they are parsed. Currently, the following processors are supported:


The additional_fields processor accepts a static map of field names and values and appends those to every event it processes. These values will overwrite existing fields of the same name, if they exist.

For example, with the following configuration:

  - additional_fields:
      environment: production
      owner: me@example.com

The field’s environment and owner will be added to the event.


The sample processor will only send a subset of events to Honeycomb. Honeycomb natively supports sampled event streams, allowing you to send a representative subset of events while still getting high-fidelity query results.


key type description
type "static" or "dynamic" How events should be sampled.
rate integer The rate at which to sample events. Specifying a sample rate of 20 will cause one in 20 events to be sent.
keys list of strings The list of field keys to use when doing dynamic sampling.
windowsize integer How often to refresh estimated sample rates during dynamic sampling, in seconds. Default: 30s.
minEventsPerSec integer Whenever the number of events per second being processed falls below this value for a time window (see windowSize), sampling will be disabled for the next time window (all events will be sent with a sample rate of 1). Default: 50. The minimum possible value is 1.


The drop_field processor will remove the specified field from all events before sending them to Honeycomb. This is useful for removing sensitive information from events.


key value description
field string The name of the field to drop.


The request_shape processor will take a field representing an HTTP request, such as GET /api/v1/users?id=22 HTTP/1.1, and unpack it into its constituent parts.


key value description
field string The name of the field containing the HTTP request (for example, "request")
patterns list of strings A list of URL patterns to match when unpacking the request
queryKeys list of strings An allowlist of keys in the URL query string to unpack
prefix string A prefix to prepend to the unpacked field names

For example, with the following configuration:

- request_shape:
    field: request
    - /api/:version/:resource
    - id

the request_shape processor will expand the event

{"request": "GET /api/v1/users?id=22 HTTP/1.1", ...}


    "request": "GET /api/v1/users?id=22 HTTP/1.1",
    "request_method": "GET",
    "request_protocol_version": "HTTP/1.1",
    "request_uri": "/api/v1/users?id=22",
    "request_path": "/api/v1/users",
    "request_query": "id=22",
    "request_shape": "/api/:version/:resource?id=?",
    "request_path_version": "v1",
    "request_path_resource": "users",
    "request_pathshape": "/api/:version/:resource",
    "request_queryshape": "id=?",
    "request_query_id": "22",


The timefield processor will replace the default timestamp in an event with one extracted from a specific field in the event.


key value description
field string The name of the field containing the timestamp
format string The format of the timestamp found in timefield, in strftime or Golang format

Note: This processor is not generally necessary when collecting pod logs. The Honeycomb Kubernetes Agent will automatically use the timestamp recorded by the Docker json-log driver. It is useful when parsing logs that live at a particular path on the node filesystem, such as Kubernetes audit logs.

Sample Configurations 

Here are some example Kubernetes Honeycomb Agent configurations.

Parse logs from pods labelled with app: nginx:

writekey: "YOUR_API_KEY"
  - labelSelector: app=nginx
    parser: nginx
    dataset: nginx-kubernetes

    - request_shape:
        field: request

Send logs from different services to different datasets:

writekey: "YOUR_API_KEY"
  - labelSelector: "app=nginx"
    parser: nginx
    dataset: nginx-kubernetes

  - labelSelector: "app=frontend-web"
    parser: json
    dataset: frontend

Sample events from a frontend-web deployment: only send one in 20 events from the prod namespace, and one in 10 events from the staging namespace.

writekey: "YOUR_API_KEY"
  - labelSelector: "app=frontend-web"
    namespace: prod
    parser: json
    dataset: frontend

      - sample:
          type: static
          rate: 20
      - drop_field:
        field: user_email

  - labelSelector: "app=frontend-web"
    namespace: staging
    parser: json
    dataset: frontend

      - sample:
          type: static
          rate: 10

Get logs from a multi-container pod, but only from the sidecar container:

writekey: "YOUR_API_KEY"
  - labelSelector: "app=frontend-web"
    containerName: sidecar
    parser: json
    dataset: frontend

Getting Help 

Have questions? Encountering difficulties? Missing features? Join our Pollinators Community Slack to ask questions and learn more.