fluent-plugin-kubernetes_metadata_filter, a plugin for Fluentd

The Kubernetes metadata plugin filter enriches container log records with pod and namespace metadata.

This plugin derives basic metadata about the container that emitted a given log record using the source of the log record. Records from journald provide metadata about the container environment as named fields. Records from JSON files encode metadata about the container in the file name. The initial metadata derived from the source is used to lookup additional metadata about the container's associated pod and namespace (e.g. UUIDs, labels, annotations) when the kubernetes_url is configured. If the plugin cannot authoritatively determine the namespace of the container emitting a log record, it will use an 'orphan' namespace ID in the metadata. This behaviors supports multi-tenant systems that rely on the authenticity of the namespace for proper log isolation.

Requirements

fluent-plugin-kubernetes_metadata_filter	fluentd	ruby
>= 2.5.0	>= v1.10.0	>= 2.5
>= 2.0.0	>= v0.14.20	>= 2.1
< 2.0.0	>= v0.12.0	>= 1.9

NOTE: For v0.12 version, you should use 1.x.y version. Please send patch into v0.12 branch if you encountered 1.x version's bug.

NOTE: This documentation is for fluent-plugin-kubernetes_metadata_filter-plugin-elasticsearch 2.x or later. For 1.x documentation, please see v0.12 branch.

Installation

gem install fluent-plugin-kubernetes_metadata_filter

Configuration

Configuration options for fluent.conf are:

kubernetes_url - URL to the API server. Set this to retrieve further kubernetes metadata for logs from kubernetes API server. If not specified, environment variables KUBERNETES_SERVICE_HOST and KUBERNETES_SERVICE_PORT will be used if both are present which is typically true when running fluentd in a pod.
apiVersion - API version to use (default: v1)
ca_file - path to CA file for Kubernetes server certificate validation
verify_ssl - validate SSL certificates (default: true)
client_cert - path to a client cert file to authenticate to the API server
client_key - path to a client key file to authenticate to the API server
bearer_token_file - path to a file containing the bearer token to use for authentication
tag_to_kubernetes_name_regexp - the regular expression used to extract kubernetes metadata (pod name, container name, namespace) from the current fluentd tag. This must used named capture groups for container_name, pod_name & namespace default: See code
cache_size - size of the cache of Kubernetes metadata to reduce requests to the API server (default: 1000)
cache_ttl - TTL in seconds of each cached element. Set to negative value to disable TTL eviction (default: 3600 - 1 hour)
watch - set up a watch on pods on the API server for updates to metadata (default: true)
de_dot - replace dots in labels and annotations with configured de_dot_separator, required for ElasticSearch 2.x compatibility (default: true)
de_dot_separator - separator to use if de_dot is enabled (default: _)
DEPRECATED use_journal - If false, messages are expected to be formatted and tagged as if read by the fluentd in_tail plugin with wildcard filename. If true, messages are expected to be formatted as if read from the systemd journal. The MESSAGE field has the full message. The CONTAINER_NAME field has the encoded k8s metadata (see below). The CONTAINER_ID_FULL field has the full container uuid. This requires docker to use the --log-driver=journald log driver. If unset (the default), the plugin will use the CONTAINER_NAME and CONTAINER_ID_FULL fields if available, otherwise, will use the tag in the tag_to_kubernetes_name_regexp format.
container_name_to_kubernetes_regexp - The regular expression used to extract the k8s metadata encoded in the journal CONTAINER_NAME field default: See code
- This corresponds to the definition in the source
annotation_match - Array of regular expressions matching annotation field names. Matched annotations are added to a log record.
allow_orphans - Modify the namespace and namespace id to the values of orphaned_namespace_name and orphaned_namespace_id when true (default: true)
orphaned_namespace_name - The namespace to associate with records where the namespace can not be determined (default: .orphaned)
orphaned_namespace_id - The namespace id to associate with records where the namespace can not be determined (default: orphaned)
lookup_from_k8s_field - If the field kubernetes is present, lookup the metadata from the given subfields such as kubernetes.namespace_name, kubernetes.pod_name, etc. This allows you to avoid having to pass in metadata to lookup in an explicitly formatted tag name or in an explicitly formatted CONTAINER_NAME value. For example, set kubernetes.namespace_name, kubernetes.pod_name, kubernetes.container_name, and docker.id in the record, and the filter will fill in the rest. (default: true)
ssl_partial_chain - if ca_file is for an intermediate CA, or otherwise we do not have the root CA and want to trust the intermediate CA certs we do have, set this to true - this corresponds to the openssl s_client -partial_chain flag and X509_V_FLAG_PARTIAL_CHAIN (default: false)
skip_labels - Skip all label fields from the metadata.
skip_container_metadata - Skip some of the container data of the metadata. The metadata will not contain the container_image and container_image_id fields.
skip_master_url - Skip the master_url field from the metadata.
skip_namespace_metadata - Skip the namespace_id field from the metadata. The fetch_namespace_metadata function will be skipped. The plugin will be faster and cpu consumption will be less.
watch_retry_interval - The time interval in seconds for retry backoffs when watch connections fail. (default: 10)

NOTE: As of the release 2.1.x of this plugin, it no longer supports parsing the source message into JSON and attaching it to the payload. The following configuration options are removed:

merge_json_log
preserve_json_log

One way of preserving JSON logs can be through the parser plugin

NOTE As of this release, the use of use_journal is DEPRECATED. If this setting is not present, the plugin will attempt to figure out the source of the metadata fields from the following:

If lookup_from_k8s_field true (the default) and the following fields are present in the record: docker.container_id, kubernetes.namespace_name, kubernetes.pod_name, kubernetes.container_name, then the plugin will use those values as the source to use to lookup the metadata
If use_journal true, or use_journal is unset, and the fields CONTAINER_NAME and CONTAINER_ID_FULL are present in the record, then the plugin will parse those values using container_name_to_kubernetes_regexp and use those as the source to lookup the metadata
Otherwise, if the tag matches tag_to_kubernetes_name_regexp, the plugin will parse the tag and use those values to lookup the metdata

Reading from the JSON formatted log files with in_tail and wildcard filenames while respecting the CRI-o log format with the same config you need the fluent-plugin "multi-format-parser":

fluent-gem install fluent-plugin-multi-format-parser

The config block could look like this:

<source>
  @type tail
  path /var/log/containers/*.log
  pos_file fluentd-docker.pos
  read_from_head true
  tag kubernetes.*
  <parse>
    @type multi_format
    <pattern>
      format json
      time_key time
      time_type string
      time_format "%Y-%m-%dT%H:%M:%S.%NZ"
      keep_time_key false
    </pattern>
    <pattern>
      format regexp
      expression /^(?<time>.+) (?<stream>stdout|stderr)( (?<logtag>.))? (?<log>.*)$/
      time_format '%Y-%m-%dT%H:%M:%S.%N%:z'
      keep_time_key false
    </pattern>
  </parse>
</source>

<filter kubernetes.var.log.containers.**.log>
  @type kubernetes_metadata
</filter>

<match **>
  @type stdout
</match>

Reading from the systemd journal (requires the fluentd fluent-plugin-systemd and systemd-journal plugins, and requires docker to use the --log-driver=journald log driver):

<source>
  @type systemd
  path /run/log/journal
  pos_file journal.pos
  tag journal
  read_from_head true
</source>

# probably want to use something like fluent-plugin-rewrite-tag-filter to
# retag entries from k8s
<match journal>
  @type rewrite_tag_filter
  rewriterule1 CONTAINER_NAME ^k8s_ kubernetes.journal.container
  ...
</match>

<filter kubernetes.**>
  @type kubernetes_metadata
  use_journal true
</filter>

<match **>
  @type stdout
</match>

Log content as JSON

In former versions this plugin parsed the value of the key log as JSON. In the current version this feature was removed, to avoid duplicate features in the fluentd plugin ecosystem. It can parsed with the parser plugin like this:

<filter kubernetes.**>
  @type parser
  key_name log
  <parse>
    @type json
    json_parser json
  </parse>
  replace_invalid_sequence true
  reserve_data true # this preserves unparsable log lines
  emit_invalid_record_to_error false # In case of unparsable log lines keep the error log clean
  reserve_time # the time was already parsed in the source, we don't want to overwrite it with current time.
</filter>

Environment variables for Kubernetes

If the name of the Kubernetes node the plugin is running on is set as an environment variable with the name K8S_NODE_NAME, it will reduce cache misses and needless calls to the Kubernetes API.

In the Kubernetes container definition, this is easily accomplished by:

env:
- name: K8S_NODE_NAME
  valueFrom:
    fieldRef:
      fieldPath: spec.nodeName

Example input/output

Kubernetes creates symlinks to Docker log files in /var/log/containers/*.log. Docker logs in JSON format.

Assuming following inputs are coming from a log file named /var/log/containers/fabric8-console-controller-98rqc_default_fabric8-console-container-df14e0d5ae4c07284fa636d739c8fc2e6b52bc344658de7d3f08c36a2e804115.log:

{
  "log": "2015/05/05 19:54:41 \n",
  "stream": "stderr",
  "time": "2015-05-05T19:54:41.240447294Z"
}

Then output becomes as belows

{
  "log": "2015/05/05 19:54:41 \n",
  "stream": "stderr",
  "docker": {
    "id": "df14e0d5ae4c07284fa636d739c8fc2e6b52bc344658de7d3f08c36a2e804115",
  }
  "kubernetes": {
    "host": "jimmi-redhat.localnet",
    "pod_name":"fabric8-console-controller-98rqc",
    "pod_id": "c76927af-f563-11e4-b32d-54ee7527188d",
    "pod_ip": "172.17.0.8",
    "container_name": "fabric8-console-container",
    "namespace_name": "default",
    "namespace_id": "23437884-8e08-4d95-850b-e94378c9b2fd",
    "namespace_annotations": {
      "fabric8.io/git-commit": "5e1116f63df0bac2a80bdae2ebdc563577bbdf3c"
    },
    "namespace_labels": {
      "product_version": "v1.0.0"
    },
    "labels": {
      "component": "fabric8Console"
    }
  }
}

If using journal input, from docker configured with --log-driver=journald, the input looks like the journalctl -o export format:

# The stream identification is encoded into the PRIORITY field as an
# integer: 6, or github.com/coreos/go-systemd/journal.Info, marks stdout,
# while 3, or github.com/coreos/go-systemd/journal.Err, marks stderr.
PRIORITY=6
CONTAINER_ID=b6cbb6e73c0a
CONTAINER_ID_FULL=b6cbb6e73c0ad63ab820e4baa97cdc77cec729930e38a714826764ac0491341a
CONTAINER_NAME=k8s_registry.a49f5318_docker-registry-1-hhoj0_default_ae3a9bdc-1f66-11e6-80a2-fa163e2fff3a_799e4035
MESSAGE=172.17.0.1 - - [21/May/2016:16:52:05 +0000] "GET /healthz HTTP/1.1" 200 0 "" "Go-http-client/1.1"

Contributing

Fork it
Create your feature branch (git checkout -b my-new-feature)
Commit your changes (git commit -am 'Add some feature')
Test it (GEM_HOME=vendor bundle install; GEM_HOME=vendor bundle exec rake test)
Push to the branch (git push origin my-new-feature)
Create new Pull Request