Product analytics (Experiment) (ULTIMATE ALL)

  • Introduced in GitLab 15.4 as an Experiment feature with a flag named cube_api_proxy. Disabled by default.
  • cube_api_proxy revised to only reference the Product Analytics API in GitLab 15.6.
  • cube_api_proxy removed and replaced with product_analytics_internal_preview in GitLab 15.10.
  • product_analytics_internal_preview replaced with product_analytics_dashboards in GitLab 15.11.
  • Snowplow integration introduced in GitLab 15.11 with a flag named product_analytics_snowplow_support. Disabled by default.

FLAG: On self-managed GitLab, by default this feature is not available. To make it available per project or for your entire instance, an administrator can enable the feature flag named product_analytics_dashboards. On GitLab.com, this feature is not available. This feature is not ready for production use.

FLAG: On self-managed GitLab, by default the Snowplow integration is not available. To make it available per project or for your entire instance, an administrator can enable the feature flag named product_analytics_snowplow_support. On GitLab.com, this feature is not available. This feature is not ready for production use.

This page is a work in progress, and we're updating the information as we add more features. For more information, see the group direction page.

How product analytics works

Product analytics uses several tools:

  • Snowplow - A developer-first engine for collecting behavioral data, and passing it through to ClickHouse.
  • ClickHouse - A database suited to store, query, and retrieve analytical data.
  • Cube - An analytical graphing library that provides an API to run queries against the data stored in Clickhouse.

The following diagram illustrates the product analytics flow:

---
title: Product Analytics flow
---
flowchart TB
    subgraph Event collection
        A([SDK]) --Send user data--> B[Snowplow Collector]
        B --Pass data through--> C[Snowplow Enricher]
    end
    subgraph Data warehouse
        C --Transform and enrich data--> D([Clickhouse])
    end
    subgraph Data visualization with dashboards
        E([Dashboards]) --Generated from the YAML definition--> F[Panels/Visualizations]
        F --Request data--> G[Product Analytics API]
        G --Run Cube queries with pre-aggregations--> H[Cube]
        H --Get data from database--> D
        D --Return results--> H
        H --Transform data to be rendered--> G
        G --Return data--> F
    end

Enable product analytics

  • Introduced in GitLab 15.6 behind the feature flag named cube_api_proxy. Disabled by default.
  • Moved to be behind the feature flag named product_analytics_admin_settings in GitLab 15.7. Disabled by default.
  • cube_api_proxy removed and replaced with product_analytics_internal_preview in GitLab 15.10.
  • product_analytics_internal_preview replaced with product_analytics_dashboards in GitLab 15.11.

FLAG: On self-managed GitLab, by default this feature is not available. To make it available per project or for your entire instance, an administrator can enable the feature flags named product_analytics_dashboards and product_analytics_admin_settings. On GitLab.com, this feature is not available. This feature is not ready for production use.

To track events in your project applications on a self-managed instance, you must enable and configure product analytics.

Prerequisite:

  • You must be an administrator of a self-managed GitLab instance.
  1. On the left sidebar, expand the top-most chevron ({chevron-down}).
  2. Select Admin Area.
  3. Select Settings > General.
  4. Expand the Analytics tab and find the Product analytics section.
  5. Select Enable product analytics and enter the configuration values.
  6. Select Save changes.

Project-level settings

You can override the instance-level settings defined by the administrator on a per-project basis. This allows you to have a different configured product analytics instance for your project.

Prerequisites:

  • Product analytics must be enabled at the instance-level.
  • You must have at least the Maintainer role for the project or group the project belongs to.
  1. On the left sidebar, at the top, select Search GitLab ({search}) to find your project.
  2. Select Settings > Analytics.
  3. Expand Configure and enter the configuration values.
  4. Select Save changes.

Instrument a GitLab project

To instrument code to collect data, use one or more of the existing SDKs:

Product analytics dashboards

  • Introduced in GitLab 15.5 behind the feature flag named product_analytics_internal_preview. Disabled by default.
  • product_analytics_internal_preview replaced with product_analytics_dashboards in GitLab 15.11.

FLAG: On self-managed GitLab, by default this feature is not available. To make it available per project or for your entire instance, an administrator can enable the feature flag named product_analytics_dashboards. On GitLab.com, this feature is not available. This feature is not ready for production use.

Product analytics dashboards are a subset of dashboards under Analytics dashboards.

Specifically product analytics dashboards and visualizations make use of the cube_analytics data type. The cube_analytics data type connects to the Cube instance defined when product analytics was enabled. All filters and queries are sent to the Cube instance and the returned data is processed by the product analytics data source to be rendered by the appropriate visualizations.

Filling missing data

  • Introduced in GitLab 16.3 behind the feature flag named product_analytics_dashboards. Disabled by default.

When exporting data or viewing dashboards, if there is no data for a given day, the missing data is autofilled with 0.

This approach has the following benefits:

  • The visualization's day axis matches the selected date range, removing ambiguity about missing data.
  • Data exports have rows for the entire date range, making data analysis easier.

However, this approach also has the following limitations:

  • The day granularity must be used. All other granularities are not supported at this time.
  • It only fills a date range defined by the inDateRange filter.
    • The date selector in the UI already uses this filter.
  • The filling of data ignores the query-defined limit. If you set a limit of 10 data points over 20 days, it returns 20 data points, with the missing data filled by 0.

Issue 417231 proposes a solution to this limitation.

Funnel analysis

Use funnel analysis to understand the flow of users through your application, and where users drop out of a predefined flow (for example, a checkout process or ticket purchase).

Each product can also define an unlimited number of funnels. Like dashboards, funnels are defined using the GitLab YAML schema, and stored in the .gitlab/analytics/funnels/ directory of a project repository.

Funnel definitions must include the keys name and seconds_to_convert, and an array of steps.

Key Description
name The name of the funnel.
seconds_to_convert The number of seconds a user has to complete the funnel.
steps An array of funnel steps.

Each step must include the keys name, target, and action.

Key Description
name The name of the step. This should be a unique slug.
action The action performed. (Only pageview is supported.)
target The target of the step. (Because only pageview is supported, this should be a path.)

Example funnel definition

The following example defines a funnel that tracks users who completed a purchase within one hour by going through three target pages:

name: completed_purchase
seconds_to_convert: 3600
steps:
  - name: view_page_1
    target: '/page1.html'
    action: 'pageview'
  - name: view_page_2
    target: '/page2.html'
    action: 'pageview'
  - name: view_page_3
    target: '/page3.html'
    action: 'pageview'

Query a funnel

You can query the funnel data with the REST API. To do this, you can use the example query body below, where you need to replace FUNNEL_NAME with your funnel's name.

NOTE: The afterDate filter is not supported. Please use beforeDate or inDateRange.

{
  "query": {
      "measures": [
        "FUNNEL_NAME.count"
      ],
      "order": {
        "completed_purchase.count": "desc"
      },
      "filters": [
        {
          "member": "FUNNEL_NAME.date",
          "operator": "beforeDate",
          "values": [
            "2023-02-01"
          ]
        }
      ],
      "dimensions": [
        "FUNNEL_NAME.step"
      ]
    }
}

Raw data export

Exporting the raw event data from the underlying storage engine can help you debug and create datasets for data analysis.

Because Cube acts as an abstraction layer between the raw data and the API, the exported raw data has some caveats:

  • Data is grouped by the selected dimensions. Therefore, the exported data might be incomplete, unless including both utcTime and userAnonymousId.
  • Data is by default limited to 10,000 rows, but you can increase the limit to maximum 50,000 rows. If your dataset has more than 50,000 rows, you must paginate through the results by using the limit and offset parameters.
  • Data is always returned in JSON format. If you need it in a different format, you need to convert the JSON to the required format using a scripting language of your choice.

Issue 391683 tracks efforts to implement a more scalable export solution.

Export raw data with Cube queries

You can query the raw data with the REST API, and convert the JSON output to any required format.

To export the raw data for a specific dimension, pass a list of dimensions to the dimensions key. For example, the following query outputs the raw data for the attributes listed:

POST /api/v4/projects/PROJECT_ID/product_analytics/request/load?queryType=multi

{
    "query":{
  "dimensions": [
    "TrackedEvents.docEncoding",
    "TrackedEvents.docHost",
    "TrackedEvents.docPath",
    "TrackedEvents.docSearch",
    "TrackedEvents.eventType",
    "TrackedEvents.localTzOffset",
    "TrackedEvents.pageTitle",
    "TrackedEvents.src",
    "TrackedEvents.utcTime",
    "TrackedEvents.vpSize"
  ],
  "order": {
    "TrackedEvents.apiKey": "asc"
  }
    }
}

If the request is successful, the returned JSON includes an array of rows of results.

Troubleshooting

No events are collected

Check your instrumentation details, and make sure product analytics is enabled and set up correctly.