Systems that use Fuchsia performance test results

This page lists the systems that consume results from Fuchsia's performance tests.

  • Chromeperf (the Chrome Performance Dashboard): Chromeperf maintains a database of performance test results. Fuchsia Infra uploads results to that database.

    Chromeperf has three main features that are used by the Fuchsia project:

    • Post-commit regression alerting: Chromeperf can detect performance regressions and report them by filing bugs in Fuchsia's Monorail issue tracker. Chromeperf is only able to detect regressions post-commit, i.e. after the CL that causes a regression has landed. Chromeperf only uses performance results data from CI builds and not from CQ builds.

      An example of a regression alert issue filed by Chromeperf is

    • Web UI for graphing performance results: Chromeperf's graphs show results by revision/time.

    • BigQuery database: Chromeperf publishes BigQuery tables containing performance results data and details of regressions that Chromeperf has detected. This can be useful for doing further analysis and graphing of the data. These tables are not publicly accessible at the moment.

    Chromeperf is used by the Chrome project and is open source. However, currently all the results uploaded to Chromeperf from Fuchsia builds are not made publicly visible.

    Chromeperf is sometimes referred to as the “Catapult Performance Dashboard” in the Fuchsia codebase, because its code lives in the Catapult project Git repo.

    Chromeperf uses a “push” model: Fuchsia builds have a step that uploads performance results to Chromeperf.

    See Chromeperf uploading and configuration for a description of the upload code path and configuration.

  • Per-build results summary pages: Each Fuchsia build that runs performance tests successfully contains a link to a table of results from those tests. See “summary of perf test results” -> “stdout” on the build page.

  • Perfcompare try builder: Perfcompare is an optional try builder that can be used to measure the performance impact of a CL before landing it. It runs performance tests with and without the CL applied and compares their results to check for performance improvements and regressions.

  • Culprit finder tools: This system graphs and analyzes results from both CI and CQ builds. It is used as part of the performance regression triage process for identifying which CL introduced each regression.

    This system is currently Google-internal. Googlers can refer to the Google-internal documentation for more details.

    This system uses a “pull” model: It enumerates Fuchsia builds using Buildbucket and fetches data from them.