Metrics Dictionary Guide

Service Ping metrics are defined in individual YAML files definitions from which the Metrics Dictionary is built. This guide describes the dictionary and how it's implemented.

Metrics Definition and validation

We are using JSON Schema to validate the metrics definition.

This process is meant to ensure consistent and valid metrics defined for Service Ping. All metrics must:

Comply with the defined JSON schema.
Have a unique key_path .
Have an owner.

All metrics are stored in YAML files:

config/metrics

WARNING: Only metrics with a metric definition YAML are added to the Service Ping JSON payload.

Each metric is defined in a separate YAML file consisting of a number of fields:

Field	Required	Additional information
`key_path`	yes	JSON key path for the metric, location in Service Ping payload.
`name`	no	Metric name suggestion. Can replace the last part of `key_path`.
`description`	yes
`product_section`	yes	The section.
`product_stage`	no	The stage for the metric.
`product_group`	yes	The group that owns the metric.
`product_category`	no	The product category for the metric.
`value_type`	yes	`string`; one of `string`, `number`, `boolean`, `object`.
`status`	yes	`string`; status of the metric, may be set to `active`, `removed`, `broken`.
`time_frame`	yes	`string`; may be set to a value like `7d`, `28d`, `all`, `none`.
`data_source`	yes	`string`; may be set to a value like `database`, `redis`, `redis_hll`, `prometheus`, `system`.
`data_category`	yes	`string`; categories of the metric, may be set to `operational`, `optional`, `subscription`, `standard`. The default value is `optional`.
`instrumentation_class`	no	`string`; the class that implements the metric.
`distribution`	yes	`array`; may be set to one of `ce, ee` or `ee`. The distribution where the tracked feature is available.
`performance_indicator_type`	no	`array`; may be set to one of `gmau`, `smau`, `paid_gmau`, or `umau`.
`tier`	yes	`array`; may contain one or a combination of `free`, `premium` or `ultimate`. The tier where the tracked feature is available. This should be verbose and contain all tiers where a metric is available.
`milestone`	no	The milestone when the metric is introduced.
`milestone_removed`	no	The milestone when the metric is removed.
`introduced_by_url`	no	The URL to the merge request that introduced the metric.
`repair_issue_url`	no	The URL of the issue that was created to repair a metric with a `broken` status.
`options`	no	`object`: options information needed to calculate the metric value.
`skip_validation`	no	This should not be set. Used for imported metrics until we review, update and make them valid.

Metric statuses

Metric definitions can have one of the following statuses:

active: Metric is used and reports data.
broken: Metric reports broken data (for example, -1 fallback), or does not report data at all. A metric marked as broken must also have the repair_issue_url attribute.
removed: Metric was removed, but it may appear in Service Ping payloads sent from instances running on older versions of GitLab.

Metric value_type

Metric definitions can have one of the following values for value_type:

boolean
number
string
object: A metric with value_type: object must have value_json_schema with a link to the JSON schema for the object. In general, we avoid complex objects and prefer one of the boolean, number, or string value types. An example of a metric that uses value_type: object is topology (/config/metrics/settings/20210323120839_topology.yml), which has a related schema in /config/metrics/objects_schemas/topology_schema.json.

Metric time_frame

7d: The metric data applies to the most recent 7-day interval. For example, the following metric counts the number of users that create epics over a 7-day interval: ee/config/metrics/counts_7d/20210305145820_g_product_planning_epic_created_weekly.yml.
28d: The metric data applies to the most recent 28-day interval. For example, the following metric counts the number of unique users that create issues over a 28-day interval: config/metrics/counts_28d/20210216181139_issues.yml.
all: The metric data applies for the whole time the metric has been active (all-time interval). For example, the following metric counts all users that create issues: /config/metrics/counts_all/20210216181115_issues.yml.
none: The metric collects a type of data that's not tracked over time, such as settings and configuration information. Therefore, a time interval is not applicable. For example, uuid has no time interval applicable: config/metrics/license/20210201124933_uuid.yml.

Metric name

To improve metric discoverability by a wider audience, each metric with instrumentation added at an appointed key_path receives a name attribute filled with the name suggestion, corresponding to the metric data_source and instrumentation. Metric name suggestions can contain two types of elements:

User input prompts: Enclosed by <>, these pieces should be replaced or removed when you create a metrics YAML file.
Fixed suggestion: Plaintext parts generated according to well-defined algorithms. They are based on underlying instrumentation, and should not be changed.

For a metric name to be valid, it must not include any prompt, and no fixed suggestions should be changed.

Data category

We use the following categories to classify a metric:

operational: Required data for operational purposes.
optional: Default value for a metric. Data that is optional to collect. This can be enabled or disabled in the Admin Area.
subscription: Data related to licensing.
standard: Standard set of identifiers that are included when collecting data.

Metric name suggestion examples

Metric with `data_source: database`

For a metric instrumented with SQL:

SELECT COUNT(DISTINCT user_id) FROM clusters WHERE clusters.management_project_id IS NOT NULL

Suggested name: count_distinct_user_id_from_<adjective describing: '(clusters.management_project_id IS NOT NULL)'>_clusters
Prompt: <adjective describing: '(clusters.management_project_id IS NOT NULL)'> should be replaced with an adjective that best represents filter conditions, such as project_management
Final metric name: For example, count_distinct_user_id_from_project_management_clusters

For metric instrumented with SQL:

SELECT COUNT(DISTINCT clusters.user_id)
FROM clusters_applications_helm
INNER JOIN clusters ON clusters.id = clusters_applications_helm.cluster_id
WHERE clusters_applications_helm.status IN (3, 5)

Suggested name: count_distinct_user_id_from_<adjective describing: '(clusters_applications_helm.status IN (3, 5))'>_clusters_<with>_<adjective describing: '(clusters_applications_helm.status IN (3, 5))'>_clusters_applications_helm
Prompt: <adjective describing: '(clusters_applications_helm.status IN (3, 5))'> should be replaced with an adjective that best represents filter conditions
Final metric name: count_distinct_user_id_from_clusters_with_available_clusters_applications_helm

In the previous example, the prompt is irrelevant, and user can remove it. The second occurrence corresponds with the available scope defined in Clusters::Concerns::ApplicationStatus. It can be used as the right adjective to replace prompt.

The <with> represents a suggested conjunction for the suggested name of the joined relation. The person documenting the metric can use it by either:

Removing the surrounding <>.
Using a different conjunction, such as having or including.

Metric with `data_source: redis` or `redis_hll`

For metrics instrumented with a Redis-based counter, the suggested name includes only the single prompt to be replaced by the person working with metrics YAML.

Prompt: <please fill metric name, suggested format is: {subject}_{verb}{ing|ed}_{object} eg: users_creating_epics or merge_requests_viewed_in_single_file_mode>
Final metric name: We suggest the metric name should follow the format of {subject}_{verb}{ing|ed}_{object}, such as user_creating_epics, users_triggering_security_scans, or merge_requests_viewed_in_single_file_mode

Metric with `data_source: prometheus` or `system`

For metrics instrumented with Prometheus or coming from the operating system, the suggested name includes only the single prompt by person working with metrics YAML.

Prompt: <please fill metric name>
Final metric name: Due to the variety of cases that can apply to this kind of metric, no naming convention exists. Each person instrumenting a metric should use their best judgment to come up with a descriptive name.

Example YAML metric definition

The linked uuid YAML file includes an example metric definition, where the uuid metric is the GitLab instance unique identifier.

key_path: uuid
description: GitLab instance unique identifier
product_category: collection
product_section: growth
product_stage: growth
product_group: group::product intelligence
value_type: string
status: active
milestone: 9.1
instrumentation_class: UuidMetric
introduced_by_url: https://gitlab.com/gitlab-org/gitlab/-/merge_requests/1521
time_frame: none
data_source: database
distribution:
- ce
- ee
tier:
- free
- premium
- ultimate

Create a new metric definition

The GitLab codebase provides a dedicated generator to create new metric definitions.

For uniqueness, the generated files include a timestamp prefix in ISO 8601 format.

The generator takes a list of key paths and 3 options as arguments. It creates metric YAML definitions in the corresponding location:

--ee, --no-ee Indicates if metric is for EE.
--dir=DIR Indicates the metric directory. It must be one of: counts_7d, 7d, counts_28d, 28d, counts_all, all, settings, license.
--class_name=CLASS_NAME Indicates the instrumentation class. For example UsersCreatingIssuesMetric, UuidMetric

Single metric example

bundle exec rails generate gitlab:usage_metric_definition counts.issues --dir=7d --class_name=CountIssues
create  config/metrics/counts_7d/issues.yml

Multiple metrics example

bundle exec rails generate gitlab:usage_metric_definition counts.issues counts.users --dir=7d --class_name=CountUsersCreatingIssues
create  config/metrics/counts_7d/issues.yml
create  config/metrics/counts_7d/users.yml

NOTE: To create a metric definition used in EE, add the --ee flag.

bundle exec rails generate gitlab:usage_metric_definition counts.issues --ee --dir=7d --class_name=CountUsersCreatingIssues
create  ee/config/metrics/counts_7d/issues.yml

Metrics added dynamic to Service Ping payload

The Redis HLL metrics are added automatically to Service Ping payload.

A YAML metric definition is required for each metric. A dedicated generator is provided to create metric definitions for Redis HLL events.

The generator takes category and event arguments, as the root key is redis_hll_counters, and creates two metric definitions for weekly and monthly time frames:

bundle exec rails generate gitlab:usage_metric_definition:redis_hll issues count_users_closing_issues
create  config/metrics/counts_7d/count_users_closing_issues_weekly.yml
create  config/metrics/counts_28d/count_users_closing_issues_monthly.yml

To create a metric definition used in EE, add the --ee flag.

bundle exec rails generate gitlab:usage_metric_definition:redis_hll issues users_closing_issues --ee
create  config/metrics/counts_7d/i_closed_weekly.yml
create  config/metrics/counts_28d/i_closed_monthly.yml