Test Definition Project

From eLinux.org
Revision as of 10:04, 9 August 2019 by Tim Bird (talk | contribs) (Test control)
Jump to: navigation, search

The Test Definition Project is consists of work to categorize and harmonize the attribute of open source tests, so that tests and test artifacts can more easily be shared between different test projects.

Tim Bird is working on a "Test Definition Standards" document, to be presented and discussed at various meetings in the Fall of 2019.

This page has information that is being retained on this wiki as resources for this work.



See Test definition survey

Test definition elements


  • Information about a test
  • Pre-requisites
  • Dependencies
  • Instructions
  • Output format
  • Test variables
  • Results analysis
  • Visualization control

For individual elements of the above:

  • field or item name
  • language (C, Python, sh, etc.)
  • file format (json, xml, etc.)
  • allowed values
    • including APIs, for code items
  • groupings

Information about a test (meta-data)

Also known as test meta-data. Does not affect the test execution, but provides information about it.

Surveyed fields:

  • Name
  • Description
  • License
  • Version
    • test program version
    • test wrapper version
  • Author
  • Maintainer
  • Test format version
  • Package manifest


  • required machine attributes
    • memory
    • cpus
    • storage
    • architecture
    • (specific hardware - for scheduling)
      • network, bus, device
  • required kernel attributes
    • config value (= kernel feature)
    • kernel module
  • required distribution attributes
    • distribution
    • distro version
    • logging system
    • init system
    • installed package
    • installed program
    • installed file
  • test environment
    • root permissions
    • file system type
  • tags


Things that can be installed or modified

  • packages

Test control

test variables

    • params


    • how does each test handle skiplists
    • Tests that are know to handle skiplists:
      • LTP
      • xfstests
        • xfstests also has a mechanism for selecting individual tests (which Fuego does for LTP with LTP_one_test)
      • (does kselftest? - they have a skip mechanism based on CONFIG fragments, is it generalized)

expected duration (timeout)

Expected duration is something a test needs to communicate to the test harness (test manager and scheduler), so that the harness can automatically detect if the test or machine has hung.

The difficulty here is that actual duration may be affected by a lot of factors. So coming up with a value that will work in all circumstances is difficult.

Some systems, like Jenkins, measure the duration of previous runs, but (to my knowledge), do not use that information to stop an instance of a test.


  • source location
  • build instructions
  • run instructions
  • setup instructions
  • teardown/cleanup instructions

Output format

  • parser
  • results output API

Results analysis

  • pass criteria
    • ignore lists
    • pass/fail counts
    • expected values
    • threshholds

Visualization control

  • key fields
  • chart types
  • thresholds
  • groupings
  • summaries

Things that are NOT test part of a test definition

  • build artifact
  • results artifact
  • board scheduling API
  • lab management API (unless test performs board reboot, provision, hardware control, etc.)
  • trigger API
  • scheduling API

element expression

  • file formats
  • file names
  • data format


  • API between test and test framework
  • API between test and target system
  • API between test and external equipment

What languages are supported?


  • beakerlib
  • fuego_board_function_lib.sh
  • sh_test_lib