• Supported metric types
  • All time total counters
  • Time framed total counters
  • Time framed unique counters
• Event Selection Rules
  • Filters
• Aggregated metrics

Metrics definitions

Metrics are defined in YAML files located in subfolders of config/metrics and ee/config/metrics. The YAML files are called metrics definitions.

This page describes the subsection of metric definitions with data_source: internal_events. You can find a general overview of metric definition files in the Metric Dictionary Guide

Supported metric types

Internal events supports three different metric types which are grouped like this:

1. All time total counters
2. Time framed total counters
3. Time framed unique counters

You can tell if a metric is counting unique values or total values by looking at the event selection rules.

A snippet from a unique metric could look like below. Notice the unique property which defines which identifier of the create_merge_request event is used for counting the unique values.

events:
  - name: create_merge_request
    unique: user_id

Similarly, a snippet from a total count metric can look like below. Notice how there is no unique property.

events:
  - name: create_merge_request

We can track multiple events within one metric via aggregated metrics.

All time total counters

Example: Total visits to /groups/:group/-/analytics/productivity_analytics all time

data_category: optional
key_path: counts.productivity_analytics_views
description: Total visits to /groups/:group/-/analytics/productivity_analytics all time
product_group: optimize
value_type: number
status: active
time_frame: all
data_source: internal_events
events:
- name: view_productivity_analytics
distribution:
- ee
tier:
- premium
- ultimate
performance_indicator_type: []
milestone: "<13.9"

The combination of time_frame: all and the event selection rule under events referring to the view_productivity_analytics event means that this is an “all time total count” metric.

Time framed total counters

An example is: Weekly count of Runner usage CSV report exports

key_path: counts.count_total_export_runner_usage_by_project_as_csv_weekly
description: Weekly count of Runner usage CSV report exports
product_group: runner
performance_indicator_type: []
value_type: number
status: active
milestone: '16.9'
introduced_by_url: https://gitlab.com/gitlab-org/gitlab/-/merge_requests/142328
data_source: internal_events
data_category: optional
distribution:
  - ee
tier:
  - ultimate
time_frame: 7d
events:
  - name: export_runner_usage_by_project_as_csv

The combination of time_frame: 7d and the event selection rule under events referring to the export_runner_usage_by_project_as_csv event means that this is a “timed framed total count” metric.

Time framed unique counters

Example: Count of distinct users who opted to filter out anonymous users on the analytics dashboard view in the last 28 days.

key_path: count_distinct_user_id_from_exclude_anonymised_users_28d
description: Count of distinct users who opted to filter out anonymous users on the analytics dashboard view in the last 28 days.
product_group: product_analytics
performance_indicator_type: []
value_type: number
status: active
milestone: '16.7'
introduced_by_url: https://gitlab.com/gitlab-org/gitlab/-/merge_requests/138150
time_frame: 28d
data_source: internal_events
data_category: optional
distribution:
- ee
tier:
- ultimate
events:
- name: exclude_anonymised_users
  unique: user.id

The combination of time_frame: 28d, the event selection rule under events referring to the exclude_anonymised_users event and the unique value (unique: user.id) means that this is a “timed framed unique count” metric.

Event Selection Rules

Event selection rules are the parts which connects metric definitions and event definitions. They are needed to know which metrics should be updated when an event is triggered.

Each internal event based metric should have a least one event selection rule with the following properties.

An example of a single event selection rule which updates a unique count metric when an event called pull_package with additional property label with the value rubygems occurs:

- name: pull_package
  unique: user.id
  filter:
    label: rubygems

Filters

Filters are used to constrain which events cause an metric to increase.

This filter includes only pull_package events with label: rubygems:

- name: pull_package
  filter:
    label: rubygems

Whereas, this filter is even more restricted and only includes pull_package events with label: rubygems and property: deploy_token:

- name: pull_package
  filter:
    label: rubygems
    property: deploy_token

Filters only support matching of exact values and not wildcards or regular expressions.

Aggregated metrics

A metric definition with several event selection rules can be considered an aggregated metric.

If you want to get total number of pull_package and push_package events you have to add two event selection rules:

events:
- name: pull_package
- name: push_package

To get the number of unique users that have at least pushed or pulled a package once:

events:
- name: pull_package
  unique: user.id
- name: push_package
  unique: user.id

Notice that unique metrics and total count metrics cannot be mixed in a single metric.