tree: 15a0f0e0cbdf3a6de81ff61a3b0e4ba74c9b00da [path history] [tgz]
  1. meta/
  2. BUILD.gn
  3. converter.cc
  4. converter.h
  5. converter_unittest.cc
  6. main.cc
  7. README.md
garnet/bin/catapult_converter/README.md

Catapult performance results converter

This directory contains the catapult_converter command line tool which takes perf test results in our format and converts them to the Catapult Dashboard's JSON “HistogramSet” format.

Parameters

The Catapult dashboard requires the following parameters (called “diagnostics”) to be present in each HistogramSet that is uploaded to the dashboard:

  • pointId: This parameter is taken from the --execution-timestamp-ms argument. The dashboard uses this value to order results from different builds in a graph.

  • benchmarks: This parameter is taken from the test_suite field in the JSON input file. This is often the name of the executable containing the perf tests, prefixed by “fuchsia.”, e.g. “fuchsia.zircon_benchmarks”.

  • masters: The term “master” is an outdated term from when Buildbot was used by Chrome infrastructure. The convention now is to use the name of the bucket containing the builder for this parameter.

  • bots: The term “bot” is an outdated term from when Buildbot was used by Chrome infrastructure. The convention now is to use the name of the builder for this parameter.

  • logUrls: This parameter is taken from the --log-url argument.

    This should contain a link to the LUCI build log page (or a page on any similar continuous integration system) for the build that produced the perf test results. The Catapult dashboard will show this link if you select a point on a performance results graph.

    Note: Although the converter requires the --log-url argument, the Catapult dashboard does not require the logUrls field in the HistogramSet data.

For more information on Catapult's format, see How to Write Metrics.