From Fedora Project Wiki

Revision as of 14:56, 22 January 2018 by Psss (talk | contribs) (Add links to the initial proof of concept)

Introduction

In order to keep test execution efficient when number of test cases grows, it is crucial to maintain corresponding metadata, which define some aspects of how the test coverage is executed.

This document proposes a flexible format for defining metadata in plain text files which can be stored close to the test code and structured in a hiearchical way with support for inheritance.

Although the proposal initially originated from user stories centered around test execution, the format is general and thus can be used in broader scenarios, e.g. test coverage mapping.

Using this approach it's also possible to combine both test execution metadata and test coverage information. Thanks to elasticity and hiearchy it provides ability to organize data into well-sized text documents while preventing duplication.

Contact: Petr Šplíchal

Stones

These are essential corner stones for the design:

  • Text files under version control
  • Keep common uses cases simple
  • Use hiearchy to organize content
  • Prevent duplication where possible
  • Metadata close to the test code
  • Solution should be open source
  • Focus on essential use cases

Stories

Important user stories to be covered:

  • As a tester or developer I want to easy read and modify metadata and see history.
  • As a tester I want to select a subset of test cases for execution by specifying a tag.
  • As a tester I want to define a maximum time for a test case to run.
  • As a tester I want to specify which environment is relevant for testing.
  • As a user I want to easily define common metadata for multiple cases to simplify maintenance.
  • As a user I want to provide specific metadata for selected tests to complement common metadata.
  • As an individual tester and test contributor I want to execute specific single test case.
  • As an automation tool I need a metadata storage with good api, extensible, quick for reading.

Choices

  • Use git for version control and history of changes.
  • Yaml format easily readable for both machines and humans.

Naming

A dedicated file name extension fmf as an abbreviation of Flexible Metadata Format to easily find all metadata files:

  • main.fmf
  • smoke.fmf

The format does not define attribute naming in any way. This is up to individual projects. The only exception is the special name main which is reserved for main directory index.

Attribute namespacing can be introduced as needed to prevent collisions between similar attributes. For example:

  • test-description, requirement-description
  • test:description, requirement:description
  • test_description, requirement_description

Objects

Objects are identified by path from the git root directory. Special file name main.fmf works similarly as index.html:

Location Identifier
wget/main.fmf wget
wget/download/main.fmf wget/download
wget/download/smoke.fmf wget/download/smoke

Features

Let's demonstrate the features on a simple wget example with the following directory structure:

wget
├── download
├── protocols
│   ├── ftp
│   ├── http
│   └── https
├── recursion
└── smoke

Simple

The most common use cases super simple to read & write. Test metadata for a single test look like this:

description: Check basic download options
tester: Petr Šplíchal <psplicha@redhat.com>
tags: [Tier2, TierSecurity]
test: runtest.sh
time: 3 min

Hiearchy

Hiearchy is defined by directory structure (see example above) and explicit nesting using attributes starting with /. Defining metadata for several tests in a single file is straightforward:

/download:
    description: Check basic download options
    tester: Petr Šplíchal <psplicha@redhat.com>
    tags: [Tier2, TierSecurity]
    test: runtest.sh
    time: 3 min
/recursion:
    description: Check recursive download options
    tester: Petr Šplíchal <psplicha@redhat.com>
    tags: [Tier2, TierSecurity]
    test: runtest.sh
    time: 20 min

Content above would be stored in wget/main.fmf file.

Inheritance

Metadata is inherited from parent objects:

tester: Petr Šplíchal <psplicha@redhat.com>
tags: [Tier2, TierSecurity]
test: runtest.sh

/download:
    description: Check basic download options
    time: 3 min
/recursion:
    description: Check recursive download options
    time: 20 min

This nicely prevents unnecessary duplication.

Elasticity

Use a single file or scatter metadata across the hiearchy, whatever is more desired for the project.

File wget/main.fmf:

tester: Petr Šplíchal <psplicha@redhat.com>
tags: [Tier2, TierSecurity]
test: runtest.sh

File wget/download/main.fmf:

description: Check basic download options
time: 3 min

File: wget/recursion/main.fmf:

description: Check recursive download options
time: 20 min

This allows reasonable structure for both small and large projects.

Scatter

Thanks to elasticity, metadata can be scattered across several files. For example wget/download metadata can be defined in the following three files:

File wget/main.fmf:

/download:
    description: Check basic download options
    test: runtest.sh

File wget/download.fmf:

description: Check basic download options
test: runtest.sh

File wget/download/main.fmf:

description: Check basic download options
test: runtest.sh

Parsing is done from top to bottom (in the order of examples above). Later/lower defined attributes replace values defined earlier/higher in the structure.

Leaves

When searching, key content is used to define which leaves from the metadata tree will be selected. For example, every test case to be executed must have the test attribute defined, every requirement to be considered for test coverage evaluation must have the requirement attribute defined. Otherwise object data is used for inheritance only.

description: Check basic download options
test: runtest.sh
time: 3 min

The key content attributes are not supposed to be hard-coded in the Flexible Metadata Format but freely configurable. Multiple key content attributes (e.g. script & backend) could be used as well.

Virtual

Using a single test code for testing multiple scenarios can be easily implemented using leaves inheriting from the same parent:

description: Check basic download options
test: runtest.sh

/fast:
    description: Check basic download options (quick smoke test)
    environment: MODE=fast
    tags: [Tier1]
    time: 1 min
/full:
    description: Check basic download options (full test set)
    environment: MODE=full
    tags: [Tier2]
    time: 3 min

In this way we can efficiently create virtual test cases.


Examples

Relevancy

Test Case Relevancy can be naturaly integrated:

description: Check basic download options
tags: [Tier2, TierSecurity]
relevancy:
- "distro < rhel-7: False"
- "arch = s390x: False"

Note that, because of YAML parsing, relevancy rules have to be enclosed in quotes. Another option is to use text format:

description: Check basic download options
tags: [Tier2, TierSecurity]
relevancy: |
    distro < rhel-7: False
    arch = s390x: False

Which seems a bit more clear.

Coverage

Test coverage information can be stored in a single file, for example wget/requirements.fmf:

/protocols:
    priority: high
    /ftp:
        requirement: Download a file using the ftp protocol.
        coverage: wget/protocols/ftp
    /http:
        requirement: Download a file using the http protocol.
        coverage: wget/protocols/http
    /https:
        requirement: Download a file using the https protocol.
        coverage: wget/protocols/https

/download:
    priority: medium
    /output-document-pipe:
        requirement: Save content to pipe.
        coverage: wget/download
    /output-document-file:
        requirement: Save content to a file.
        coverage: wget/download

/upload:
    priority: medium
    /post-file:
        requirement: Upload a file to the server
        coverage: wget/protocols/http
    /post-data:
        requirement: Upload a string to the server
        coverage: wget/protocols/http

Or split by functionality area into separate files as desired, for example wget/download/requirements.fmf:

priority: medium
/output-document-pipe:
    requirement: Save content to pipe.
    coverage: wget/download
/output-document-file:
    requirement: Save content to a file.
    coverage: wget/download

Or integrated with test case metadata, e.g. wget/download/main.fmf:

description: Check basic download options
tags: [Tier2, TierSecurity]
test: runtest.sh
time: 3 min

/requirements
    requirement: Various download options working correctly
    priority: low
    /get-file:
        coverage: wget/download
    /output-document:
        coverage: wget/download
    /continue:
    /timestamping:
    /tries:
    /no-clobber:
        coverage: wget/download
    /progress:
    /quota:
    /server-response:
    /bind-address:
    /spider:

In the example above three requirements are already covered, the rest still await for test coverage (attributes value is null).

Strategist

Here's an example implementation of test-strategist data for openscap using the Flexible Metadata Format:

/probes:
    description: Probes
    /offline:
        description: Offline scanning
    /online:
        description: Online scanning
/scanning:
    description: Reading and understanding source datastreams
    /oval:
        influencers:
        - openscap/probes/offline
        - openscap/probes/online
    /ds:
        influencers:
        - openscap/scanning/oval
        - openscap/scanning/cpe
    /cpe:
        influencers:
        - openscap/scanning/oval

Implementation

A Python module and a simple command line tool implementing the features described in this draft are available for experimenting: