Right now the functional tests are run against a distributable of Kibana in CI, but that distributable is running with `--env.name=development`. That causes the optimizer to run again before the tests can start and prevents the functional tests from running against the actual version of the application users will end up getting. This seems necessary for some tests, but not all of them, but I would like to get all of the tests running against the production version of Kibana soon. This PR implements a second ftr config, `test/functional_production` that uses basically a copy of the `test/functional` config but with a few minor adjustments, removing the `--env.name=development` kbnServerArg and using a unique junit report name. To accomplish this I needed to modify the `@kbn/test` module to only pass the `--dev` flag to the Kibana server if it is being run in development mode, which it currently does by testing the args for `--env.name=development` or two args next to each other: `'--env.name'` and `'development'`. It does this by converting the `extraKbnOpts` option into an `addExtraKbnArgs` function, which is called with the final args just before passing them to the proc runner and given a chance to modify then after all other args are resolved (pulling from different places in config based on the build type, etc.) Over the next couple weeks I'll push up PRs for individual test suites, migrating them over to the new production config, hopefully in a short period of time we will have all the function test suites back under that `test/function` config and can run them all against the Kibana server in production mode. |
||
---|---|---|
.. | ||
src | ||
.babelrc | ||
package.json | ||
README.md | ||
yarn.lock |
Kibana Testing Library
The @kbn/test package provides ways to run tests. Currently only functional testing is provided by this library, with unit and other testing possibly added here.
Functional Testing
Dependencies
Functional testing methods exist in the src/functional_tests
directory. They depend on the Functional Test Runner, which is found in {KIBANA_ROOT}/src/functional_test_runner
. Ideally libraries provided by kibana packages such as this one should not depend on kibana source code that lives in {KIBANA_ROOT}/src
. The goal is to start pulling test and development utilities out into packages so they can be used across Kibana and plugins. Accordingly the Functional Test Runner itself will be pulled out into a package (or part of a package), and this package's dependence on it will not be an issue.
Exposed methods
runTests(configPaths: Array)
For each config file specified in configPaths, starts Elasticsearch and Kibana once, runs tests specified in that config file, and shuts down Elasticsearch and Kibana once completed. (Repeats for every config file.)
configPaths
: array of strings, each an absolute path to a config file that looks like this, following the config schema specified here.
Internally the method that starts Elasticsearch comes from kbn-es.
startServers(configPath: string)
Starts Elasticsearch and Kibana servers given a specified config.
configPath
: absolute path to a config file that looks like this, following the config schema specified here.
Allows users to start another process to run just the tests while keeping the servers running with this method. Start servers and run tests using the same config file (see how).
Rationale
Single config per setup
We think it makes sense to specify the tests to run along with the particular server configuration for Elasticsearch and Kibana servers, because the tests expect a particular configuration. For example, saml api integration tests expect certain xml files to exist in Elasticsearch's config directory, and certain saml specific options to be passed in via the command line (or alternatively via the .yml
config file) to both Elasticsearch and Kibana. It makes sense to keep all these config options together with the list of test files.
Multiple configs running in succession
We also think it makes sense to have a test runner intelligently (but simply) start servers, run tests, tear down servers, and repeat for each config, uninterrupted. There's nothing special about each kind of config that specifies running some set of functional tests against some kind of Elasticsearch/Kibana servers. There doesn't need to be a separate job to run each kind of setup/test/teardown. These can all be orchestrated sequentially via the current runTests
implementation. This is how we envision tests to run on CI.
This inherently means that grouping test files in configs matters, such that a group of test files that depends on a particular server config appears together in that config's testFiles
list. Given how quickly and easily we can start servers using @kbn/es, it should not impact performance to logically group tests by domain even if multiple groups of tests share the same server config. We can think about how to group test files together across domains when that time comes.