Subset
A subset is a set of tests dynamically selected from a larger test suite using Predictive Test Selection.
The term 'dynamically selected' refers to the fact that the tests returned in subsets change based on the request parameters.

Properties

A subset is the output of a subset request made using the launchable subset CLI command. You make a subset request every time you want to run a subset of tests in your CI pipeline:
High level flow including a subset request
A subset request takes various inputs:
  1. 1.
    The Build being tested
  2. 2.
    A subset optimization target
  3. 3.
    The test runner in use
  4. 4.
    The input test list: the full list of tests that would normally run in a non-subset session (or "full run")
And outputs:
  1. 1.
    A subset list of tests formatted for your test runner
  2. 2.
    [Optional] The remainder list of tests formatted for your test runner

Build being tested

When you request a subset of tests to run in your CI process, you pass in the name of the Buildyou're testing:
launchable subset --build $BUILD_NAME [other options...]
This is important so that the Predictive Test Selection service can analyze the changes in the build and select tests appropriately.

Optimization target

When you request a subset of tests to run in your CI process, you include an optimization target:
launchable subset \
# one of:
--target [PERCENTAGE]
# or
--confidence [PERCENTAGE]
# or
--time [STRING] \
[other options...]
Launchable currently supports three optimization targets which you can read about here: choosing-an-optimization-target

Test runner

When you request a subset of tests, you include the name of the test runner you're going to run tests with. This value should be the same between launchable subset and launchable record tests commands.
The CLI uses this parameter to automatically adjust three things:
  1. 1.
    Input test list format
  2. 2.
    Subset altitude
  3. 3.
    Output test list format

Input test list format

The full list of tests you would normally run is a key input to any subset request. Launchable uses this list to create a subset of tests.
How this list is generated, formatted, and passed into the launchable subset depends on the test runner in use. In general, you don't have to worry about creating this list; the documentation for each test runner goes over the specific flow for your tool.
However, for completeness, we'll outline the various methods used across test runners:
  1. 1.
    Some test runners can generate a list of tests via a special command. The output of this command is then passed into launchable subset.
  2. 2.
    Other test runners don't provide that feature. In that case, you pass the directory/directories containing your tests into launchable subset. The CLI then creates the list of tests by scanning those directories and identifying tests using pattern-matching.
  3. 3.
    Furthermore, some frameworks can list individual tests, but they can only do so after test packages have been compiled. In this case it can be preferable to generate a list of higher-level packages instead of individual test cases. (This relates to the next section.)

Subset altitude

To run a subset of tests, you pass the returned subset list into your test runner for execution.
Each test runner has its own option for specifying a list of tests to run, and these options allow for different 'altitudes' of filtering. For example, some test runners only let you pass in a list of files to run, others support filtering by class, while some support filtering by test case or method.
Based on the test runner used in launchable subset, the CLI automatically outputs a list of tests using the hierarchy level supported by that test runner.
Another factor that impacts subset altitude is the ability of the test runner/CLI to list tests at a low altitude. (See above section for more info)
For example, Maven supports filtering by class, so we say that Maven's subset altitude is class. Test results captured using launchable record tests for Maven will include both class and testcase identifiers, but the output of launchable subset will include a list of classes only.
The Predictive Test Selection service automatically handles aggregation and prioritization at the correct hierarchy level.

Output test list format

To run a subset of tests, you pass the returned subset list into your test runner for execution.
Each test runner has its own method or option for specifying a list of tests to run. For example, one test runner might expect a comma delimited list of tests, whereas another might expect a list separated by spaces, etc.
The CLI adjusts the output format automatically based on the test runner used in the request. In general, you don't need to worry about the output format because you'll pass it directly into your test runner per the documentation for your tool. But this does mean that the contents of subset files/outputs changes based on the test runner value.

Input test list

As described above, the full list of tests you would normally run is a key input to any subset request. Launchable uses this list to create a subset of tests.
This list is important because it can change between requests due to
In general, you don't have to worry about creating the input test list, but it's important to understand this concept because it relates to your optimization target. SeeChoosing a subset optimization target for more on this.
Copy link
On this page
Properties
Build being tested
Optimization target
Test runner
Input test list