Component Readiness needs to know how to map each test to a particular component and its capabilities. This tool:
- Takes a set of metadata about all tests such as its name and suite
- Maps the test to exactly one component that provides details about the capabilities that test is testing
- Writes the result to a json file comitted to this repo in
data/
- Pushes the result to BigQuery
Teams own their component code under pkg/components/<component_name>
and can handle the mapping as they see fit. New components can copy from
pkg/components/example
and modify it, or write their own
implementation of the interface. The sample Config interface, which you
can include in your component, provides rich filters on substrings,
sigs, operators, etc.
TRT has made a first pass at assigning ownership to teams, but it's likely we haven't correctly assigned all tests. Please re-assign tests to their correct components as needed. Please also create an OWNERS file in your directory so your team can manage components and capabilities without TRT's intervention.
Component owners should return a TestOwnership
struct from their
identification function. See the details in pkg/api/v1
for details
about the TestOwnership
struct.
They should return nil when the test is not theirs. They should ONLY return an error on a fatal error such as inability to read from a file.
A test must only map to one component, but may map to several
capabilities. In the event that two components are vying for a test's
ownership, you may use the Priority
field in the TestOwnership
struct. The highest value wins.
These are the general steps for updating a test's ownership:
- Open the
component.go
for a particular component inpkg/components
. If a test is tied to a namespace, operator, or particular SIG, update that section. If you want to match on the suite or test name, add a new matcher using one of the existing types, such asSuite
, orIncludeAny
, which matches a test substring. - Update the
Priority
field if you need to force a higher or lower priority for this component. - Run
make mapping
, to apply the new rules to the test names. Review the results of the changes by usinggit diff
. Ensure that the changes you see are expected. - Commit the result and open a pull request on GitHub.
If you'd like to annotate a test as having additional capabilities,
update capabilities.go
. A test may have multiple capabilities, but it
can only belong to a single component.
make mapping
enforces this, as well as disallowing any test to move
back to Unknown
. If a test is incorrectly attributed, you must find
the owner and move it to their component.
The unfortunate reality is tests may get renamed, so we need to have a way to compare the test results across renames. To do that, each test has a stable ID which is the current test name stored in the DB as an md5sum.
The test's first stable ID is the one that remains. Component owners are
responsible for ensuring the StableID
function in their component
returns the same ID for all names of a given test. This can be done with
a simple look-up map; see the monitoring component for an example.
If a test is removed, or is refactored in such a way (i.e. one to many)
that it's not reasonable to mark them as renames, then it should be
tracked as an obsolete test. In pkg/obsoletetests
one can manage their
obsolete tests by adding an entry to the set. For OCP, only staff
engineers can approve a test's removal.
Currently the tests used for mapping come from the corpus of tests
previously seen in job results. This list is filtered down to a
smaller quantity by selecting only those in certain suites, jobs, or
matching certain names. This is configurable by specifying a
configuration file. An example is present in
config/openshift-eng.yaml
.
At a mimimum though, for compatibility with component readiness (and all other OpenShift tooling), a test must:
-
always have a result when it runs, indicating success, flake, or failure (historically some tests only report failure)
-
belong to a test suite
-
have a stable name: do not use dynamic names such as full pod names in tests
-
have a reasonable way to map to component/capabilities, such as
[sig-XYZ]
present in the test name, and using[Feature:XYZ]
or[Capability:XYZ]
to make mapping to capabilities easier
See --help for more info.
To find unmapped tests, run make unmapped
.
For production usage we fetch and push data to BigQuery, but for local
testing you can used locally comitted copies of that data by using
--mode local
:
ci-test-mapping map --mode local
For production, use --mode bigquery
and provide credentials:
ci-test-mapping map --mode bigquery \
--google-service-account-credential-file ~/bq.json \
--log-level debug \
--mapping-file mapping.json \
--push-to-bigquery
The BigQuery project, dataset, JUnit table, and component mapping tables are all configurable.
ci-test-mapping map \
--mode bigquery \
--bigquery-project openshift-gce-devel \
--bigquery-dataset ci_analysis_us \
--table-junit junit \
--table-mapping component_mapping
The BigQuery mapping table should be updated only in append mode (aside from older entries being trimmed), so mappings should limit their results to the most recent entry.
The command ./ci-test-mapping jira-verify
will inform you of any Jira
components that have been removed, and report on any new Jira components
that are available.
To create any missing components, run ./ci-test-mapping jira-create
.
You'll need to set the env var JIRA_TOKEN
to your personal API token
(which you can create from your Jira profile page). Then:
- Move any configuration for renamed components
- Delete the obsolete
pkg/components/<component>
directory - Remove references to removed components from
pkg/registry
.