Sentry monitoring in the frontend development of GitLab

The GitLab Frontend team uses Sentry as an observability tool to monitor how the UI performs for users on gitlab.com.

GitLab.com is configured to report to our Sentry instance at Admin > Metrics and profiling > Sentry.

We monitor two kinds of data: Errors and Performance.

note
The Frontend Observability Working Group is looking to improve how we use Sentry. GitLab team members can provide feedback at issue #427402.

Start using Sentry

Our Sentry instance is located at https://new-sentry.gitlab.net/. Only GitLab team members can access Sentry.

After your first log-in you can join the #gitlab team by selecting Join a team. Confirm that #gitlab appears under YOUR TEAMS in the teams page.

Error reporting

Errors, also known as “events” in the Sentry UI, are instances of abnormal or unexpected runtime behavior that users experience in their browser.

GitLab uses the Sentry Browser SDK to report errors to our Sentry instance under the project gitlabcom-clientside.

Reporting known errors

The most common way to report errors to Sentry is to call captureException(error), for example: 

import * as Sentry from '~/sentry/sentry_browser_wrapper';

try {
  // Code that may fail in runtime
} catch (error) {
  Sentry.captureException(error)
}

When should you report an error? We want to avoid reporting errors that we either don’t care about, or have no control over. For example, we shouldn’t report validation errors when a user fills out a form incorrectly. However, if that form submission fails because or a server error, this is an error we want Sentry to know about.

By default your local development instance does not have Sentry configured. Calls to Sentry are stubbed and shown in the console with a [Sentry stub] prefix for debugging.

Unhandled/unknown errors

Additionally, we capture unhandled errors automatically in all of our pages.

Error Monitoring

Once errors are captured, they appear in Sentry. For example you can see the errors reported in the last 24 hours in canary and production.

In the list, select any error to see more details… and ideally propose a solution for it!

note
We suggest filtering errors by the environments gprd and gprd-cny, as there is some spam in our environment data.

Exploring error data

Team members can use Sentry’s Discover page to find unexpected issues.

Additionally, we have created a dashboard to report which feature categories and pages produce most errors, among other data.

Engineering team members are encouraged to explore error data and find ways to reduce errors on our user interface. Sentry also provides alerts for folks interested in getting notified when errors occur.

Filtering errors

We receive several thousands of reports per day, so team members can filter errors based on their work area.

We mark errors with two additional custom tags to help identify their source:

  • feature_category: The feature area of the page. (For example, code_review_workflow or continuous_integration.) Source: gon.feature_category
  • page: Identifier of method called in the controller to render the page. (For example, projects:merge_requests:index or projects:pipelines:index.) Source: body_data_page.

Frontend engineering team members can filter errors relevant to their group and/or page.

Performance Monitoring

We use BrowserTracing to report performance metrics to Sentry.

You can visit our performance data of the last 24 hours and use the filters to drill down and learn more.