Internal analytics
The internal analytics system provides the ability to track user behavior and system status for a GitLab instance to inform customer success services and further product development.
These doc pages provide guides and information on how to leverage internal analytics capabilities of GitLab when developing new features or instrumenting existing ones.
Fundamental concepts
Events and metrics are the foundation of the internal analytics system. Understanding the difference between the two concepts is vital to using the system.
Event
An event is a record of an action that happened within the GitLab instance. An example action would be a user interaction like visiting the issue page or hovering the mouse cursor over the top navigation search. Other actions can result from background system processing like scheduled pipeline succeeding or receiving API calls from 3rd party system. Not every action is tracked and thereby turned into a recorded event automatically. Instead, if an action helps draw out product insights and helps to make more educated business decisions, we can track an event when the action happens. The produced event record, at the minimum, holds information that the action occurred, but it can also contain additional details about the context that accompanied this action. An example of context can be information about who performed the action or the state of the system at the time of the action.
Metric
A single event record is not informative enough and might be caused by a coincidence. We need to look for sets of events sharing common traits to have a foundation for analysis. This is where metrics come into play. A metric is a calculation performed on pieces of information. For example, a single event documenting a paid user visiting the feature’s page after a new feature was released tells us nothing about the success of this new feature. However, if we count the number of page view events happening in the week before the new feature release and then compare it with the number of events for the week following the feature release, we can derive insights about the increase in interest due to the release of the new feature.
This process leads to what we call a metric. An event-based metric counts the number of times an event occurred overall or in a specified time frame. The same event can be used across different metrics and a metric can count either one or multiple events. The count can but does not have to be based on a uniqueness criterion, such as only counting distinct users who performed an event.
Metrics do not have to be based on events. Metrics can also be observations about the state of a GitLab instance itself, such as the value of a setting or the count of rows in a database table.
Instrumentation
- To instrument an event-based metric, see the internal event tracking quick start guide.
- To instrument a metric that observes the GitLab instances state, see the metrics instrumentation.
Data availability
For GitLab there is an essential difference in analytics setup between SaaS and self-managed or GitLab Dedicated instances. On our SaaS instance both individual events and pre-computed metrics are available for analysis. Additionally for SaaS page views are automatically instrumented. For self-managed only the metrics instrumented on the version installed on the instance are available.
Events
Events are collected in real-time but processed in an asynchronous manner. In general events are available in the data warehouse at the latest 48 hours after being fired but can already be available earlier.
Metrics
Metrics are being computed and sent once per week for every instance. On GitLab.com this happens on Sunday and newest values become available throughout Monday. On self-managed this depends on the particular instance. In general, only the metrics instrumented for the installed GitLab version will be sent.
Data discovery
Event and metrics data is ultimately stored in our Snowflake data warehouse. It can either be accessed directly via SQL in Snowflake for ad-hoc analyses or visualized in our data visualization tool Tableau, which has access to Snowflake. Both platforms need an access request (Snowflake, Tableau).
Tableau
Tableau is a data visualization platform and allows building dashboards and GUI based discovery of events and metrics. This method of discovery is most suited for users who are familiar with business intelligence tooling, basic verifications and for creating persisted, shareable dashboards and visualizations. Access to Tableau requires an access request.
Checking events
Visit the Snowplow event exploration dashboard. This dashboard shows you event counts as well as the most fired events. You can scroll down to the “Structured Events Firing in Production Last 30 Days” chart and filter for your specific event action. The filter only works with exact names.
Checking metrics
You can visit the Metrics exploration dashboard.
On the side there is a filter for metric path which is the key_path of your metric and a filter for the installation ID including guidance on how to filter for GitLab.com.
Snowflake
Snowflake allows direct querying of relevant tables in the warehouse within their UI with the Snowflake SQL dialect. This method of discovery is most suited to users who are familiar with SQL and for quick and flexible checks whether data is correctly propagated. Access to Snowflake requires an access request.
Querying events
The following example query returns the number of daily event occurrences for the feature_used event.
SELECT
behavior_date,
COUNT(*) as event_occurences
FROM prod.common_mart.mart_behavior_structured_event
WHERE event_action = 'feature_used'
AND behavior_date > '2023-08-01' --restricted minimum date for performance
AND app_id='gitlab' -- use gitlab for production events and gitlab-staging for events from staging
GROUP BY 1 ORDER BY 1 desc
For a list of other metrics tables refer to the Data Models Cheat Sheet.
Querying metrics
The following example query returns all values reported for count_distinct_user_id_from_feature_used_7d within the last six months and the according instance_id:
SELECT
date_trunc('week', ping_created_at),
dim_instance_id,
metric_value
FROM prod.common.fct_ping_instance_metric_rolling_6_months --model limited to last 6 months for performance
WHERE metrics_path = 'counts.users_visiting_dashboard_weekly' --set to metric of interest
ORDER BY ping_created_at DESC
For a list of other metrics tables refer to the Data Models Cheat Sheet.
Data flow
On SaaS event records are directly sent to a collection system, called Snowplow, and imported into our data warehouse. Self-managed and GitLab Dedicated instances record event counts locally. Every week, a process called Service Ping sends the current values for all pre-defined and active metrics to our data warehouse. For GitLab.com, metrics are calculated directly in the data warehouse.
The following chart aims to illustrate this data flow:
Data Privacy
GitLab only receives event counts or similarly aggregated information from self-managed instances. User identifiers for individual events on the SaaS version of GitLab are pseudonymized. An exact description on what kind of data is being collected through the Internal Analytics system is given in our handbook.