Promtail Scraping (Service Discovery)

File Target Discovery

Promtail discovers locations of log files and extract labels from them through the scrape_configs section in the config YAML. The syntax is identical to what Prometheus uses.

scrape_configs contains one or more entries which are executed for each discovered target (i.e., each container in each new pod running in the instance):

  - job_name: local
      - ...

  - job_name: kubernetes
      - ...

If more than one scrape config section matches your logs, you will get duplicate entries as the logs are sent in different streams likely with slightly different labels.

There are different types of labels present in Promtail:

  • Labels starting with __ (two underscores) are internal labels. They usually come from dynamic sources like service discovery. Once relabeling is done, they are removed from the label set. To persist internal labels so they’re sent to Loki, rename them so they don’t start with __. See Relabeling for more information.

  • Labels starting with __meta_kubernetes_pod_label_* are “meta labels” which are generated based on your Kubernetes pod’s labels.

    For example, if your Kubernetes pod has a label name set to foobar, then the scrape_configs section will receive an internal label __meta_kubernetes_pod_label_name with a value set to foobar.

  • Other labels starting with __meta_kubernetes_* exist based on other Kubernetes metadata, such as the namespace of the pod (__meta_kubernetes_namespace) or the name of the container inside the pod (__meta_kubernetes_pod_container_name). Refer to the Prometheus docs for the full list of Kubernetes meta labels.

  • The __path__ label is a special label which Promtail uses after discovery to figure out where the file to read is located. Wildcards are allowed, for example /var/log/*.log to get all files with a log extension in the specified directory, and /var/log/**/*.log for matching files and directories recursively. For a full list of options check out the docs for the library promtail uses.

  • The label filename is added for every file found in __path__ to ensure the uniqueness of the streams. It is set to the absolute path of the file the line was read from.

Kubernetes Discovery

Note that while Promtail can utilize the Kubernetes API to discover pods as targets, it can only read log files from pods that are running on the same node as the one Promtail is running on. Promtail looks for a __host__ label on each target and validates that it is set to the same hostname as Promtail’s (using either $HOSTNAME or the hostname reported by the kernel if the environment variable is not set).

This means that any time Kubernetes service discovery is used, there must be a relabel_config that creates the intermediate label __host__ from __meta_kubernetes_pod_node_name:

  - source_labels: ['__meta_kubernetes_pod_node_name']
    target_label: '__host__'

See Relabeling for more information.

Journal Scraping (Linux Only)

On systems with systemd, Promtail also supports reading from the journal. Unlike file scraping which is defined in the static_configs stanza, journal scraping is defined in a journal stanza:

  - job_name: journal
      json: false
      max_age: 12h
      path: /var/log/journal
        job: systemd-journal
      - source_labels: ['__journal__systemd_unit']
        target_label: 'unit'

All fields defined in the journal section are optional, and are just provided here for reference. The max_age field ensures that no older entry than the time specified will be sent to Loki; this circumvents “entry too old” errors. The path field tells Promtail where to read journal entries from. The labels map defines a constant list of labels to add to every journal entry that Promtail reads.

When the json field is set to true, messages from the journal will be passed through the pipeline as JSON, keeping all of the original fields from the journal entry. This is useful when you don’t want to index some fields but you still want to know what values they contained.

By default, Promtail reads from the journal by looking in the /var/log/journal and /run/log/journal paths. If running Promtail inside of a Docker container, the path appropriate to your distribution should be bind mounted inside of Promtail along with binding /etc/machine-id. Bind mounting /etc/machine-id to the path of the same name is required for the journal reader to know which specific journal to read from. For example:

docker run \
  -v /var/log/journal/:/var/log/journal/ \
  -v /run/log/journal/:/run/log/journal/ \
  -v /etc/machine-id:/etc/machine-id \
  grafana/promtail:latest \

When Promtail reads from the journal, it brings in all fields prefixed with __journal_ as internal labels. Like in the example above, the _SYSTEMD_UNIT field from the journal was transformed into a label called unit through relabel_configs. See Relabeling for more information, also look at the systemd man pages for a list of fields exposed by the journal.

Here’s an example where the SYSTEMD_UNIT, HOSTNAME, and SYSLOG_IDENTIFIER are relabeled for use in Loki.

Keep in mind that labels prefixed with __ will be dropped, so relabeling is required to keep these labels.

- job_name: systemd-journal
      cluster: ops-tools1
      job: default/systemd-journal
    path: /var/log/journal
  - source_labels:
    - __journal__systemd_unit
    target_label: systemd_unit
  - source_labels:
    - __journal__hostname
    target_label: nodename
  - source_labels:
    - __journal_syslog_identifier
    target_label: syslog_identifier

Windows Event Log

On Windows Promtail supports reading from the event log. Windows event targets can be configured using the windows_events stanza:

- job_name: windows
    use_incoming_timestamp: false
    bookmark_path: "./bookmark.xml"
    eventlog_name: "Application"
    xpath_query: '*'
      job: windows
    - source_labels: ['computer']
      target_label: 'host'

When Promtail receives an event it will attach the channel and computer labels and serialize the event in json. You can relabel default labels via Relabeling if required.

Providing a path to a bookmark is mandatory, it will be used to persist the last event processed and allow resuming the target without skipping logs.

see the configuration section for more information.

Gcplog scraping

Promtail supports scraping cloud resource logs(say GCS bucket logs, Load Balancer logs, Kubernetes Cluster logs) from GCP. Configs are set in gcplog section in scrape_config

  - job_name: gcplog
      project_id: "my-gcp-project"
      subscription: "my-pubsub-subscription"
      use_incoming_timestamp: false # default rewrite timestamps.
        job: "gcplog"
      - source_labels: ['__project_id']
        target_label: 'project'

Here project_id and subscription are the only required fields.

  • project_id is the GCP project id.
  • subscription is the GCP pubsub subscription where promtail can consume log entries from.

Before using gcplog target, GCP should be configured with pubsub subscription to receive logs from.

It also support relabeling and pipeline stages just like other targets.

When Promtail receives GCP logs the labels that are set on the GCP resources are available as internal labels. Like in the example above, the __project_id label from a GCP resource was transformed into a label called project through relabel_configs. See Relabeling for more information.

Log entries scraped by gcplog will add an additional label called promtail_instance. This label uniquely identifies each promtail instance trying to scrape gcplog (from a single subscription_id). We need this unique identifier to avoid out-of-order errors from Loki servers. Because say two promtail instances rewrite timestamp of log entries(with same labelset) at the same time may reach Loki servers at different times can cause Loki servers to reject it.

Syslog Receiver

Promtail supports receiving IETF Syslog (RFC5424) messages from a tcp stream. Receiving syslog messages is defined in a syslog stanza:

  - job_name: syslog
      idle_timeout: 60s
      label_structured_data: yes
        job: "syslog"
      - source_labels: ['__syslog_message_hostname']
        target_label: 'host'

The only required field in the syslog section is the listen_address field, where a valid network address should be provided. The idle_timeout can help with cleaning up stale syslog connections. If label_structured_data is set, structured data in the syslog header will be translated to internal labels in the form of __syslog_message_sd_<ID>_<KEY>. The labels map defines a constant list of labels to add to every journal entry that Promtail reads.

Note that it is recommended to deploy a dedicated syslog forwarder like syslog-ng or rsyslog in front of Promtail. The forwarder can take care of the various specifications and transports that exist (UDP, BSD syslog, …). See recommended output configurations for syslog-ng and rsyslog.

When Promtail receives syslog messages, it brings in all header fields, parsed from the received message, prefixed with __syslog_ as internal labels. Like in the example above, the __syslog_message_hostname field from the journal was transformed into a label called host through relabel_configs. See Relabeling for more information.

Syslog-NG Output Configuration

destination d_loki {
  syslog("localhost" transport("tcp") port(<promtail_port>));

Rsyslog Output Configuration

action(type="omfwd" protocol="tcp" port="<promtail_port>" Template="RSYSLOG_SyslogProtocol23Format" TCP_Framing="octet-counted")


Each scrape_configs entry can contain a relabel_configs stanza. relabel_configs is a list of operations to transform the labels from discovery into another form.

A single entry in relabel_configs can also reject targets by doing an action: drop if a label value matches a specified regex. When a target is dropped, the owning scrape_config will not process logs from that particular source. Other scrape_configs without the drop action reading from the same target may still use and forward logs from it to Loki.

A common use case of relabel_configs is to transform an internal label such as __meta_kubernetes_* into an intermediate internal label such as __service__. The intermediate internal label may then be dropped based on value or transformed to a final external label, such as __job__.


  • Drop the target if a label (__service__ in the example) is empty:
  - action: drop
    regex: ''
    - __service__
  • Drop the target if any of the source_labels contain a value:
  - action: drop
    regex: .+
    separator: ''
    - __meta_kubernetes_pod_label_name
    - __meta_kubernetes_pod_label_app
  • Persist an internal label by renaming it so it will be sent to Loki:
  - action: replace
    - __meta_kubernetes_namespace
    target_label: namespace
  • Persist all Kubernetes pod labels by mapping them, like by mapping __meta_kube__meta_kubernetes_pod_label_foo to foo.
  - action: labelmap
    regex: __meta_kubernetes_pod_label_(.+)

Additional reading:

HTTP client options

Promtail uses the Prometheus HTTP client implementation for all calls to Loki. Therefore it can be configured using the clients stanza, where one or more connections to Loki can be established:

  - [ <client_option> ]

Refer to client_config from the Promtail Configuration reference for all available options.