This PR cuts GitHub Action build times from 14-15 minutes, down to just
under 7 minutes. It does this by:
- caching `grafana-plugins/node_modules` and `pip` dependencies based on
their respective dependency files (eg. `requirements.txt` &
`yarn.lock`). This step alone saves ~3 minutes.
- get rid of the "build-engine-docker-image" and
"backend-integration-tests" jobs in the old "Integration Tests"
workflow. This was split out this way so that we could build the backend
docker image once, upload the artifact, and then reuse it across the
backend and e2e tests. We no longer need these backend integration tests
because we are testing the same thing in the e2e tests. This saves ~45
seconds of having to upload the image artifact.
- few improvements within the integration tests themselves:
- move plugin configuration to the `globalSetup.ts`. This means that
every test does not need to check if the plugin has been configured
because it is done once before all the tests are run.
- cache the plugin frontend build. If your commit doesn't change
anything to `grafana-plugin/src` or `grafana-plugin/yarn.lock` it should
be safe to reuse a previously built/cached version of the plugin
frontend. This saves ~3 minutes
- cache playwright binaries/dependencies. Only re-install them if the
version of `@playwright/test` in `grafana-plugin/yarn.lock` changes.
This saves ~3 minutes.
**Other things to mention**
Once we refactor the `GSelect` component to not call the `onChange`
callback on every keyDown event (#1628), this should allow us to
parallelize the integration tests, and cut the time required to execute
the tests themselves in half
# What this PR does
This PR makes the Grafana login portion of the e2e tests much
faster/more reliable.
Currently we use CSS selectors to go to the login form, input the
username/password, and proceed as such. This PR refactors to instead
make a call to `POST ${GRAFANA_API_URL}/login` and then stores that
authentication state that is then reused by subsequent browsers. This
was inspired by [how the Incident team does their playwright
authentication](https://github.com/grafana/incident/blob/main/plugin/e2e/global-setup.ts)
+ the recommendation from the [Playwright
docs](https://playwright.dev/docs/auth#basic-shared-account-in-all-tests)
## Which issue(s) this PR fixes
Slow/flaky Grafana login flow
## Checklist
- [x] Tests updated
- [ ] Documentation added (N/A)
- [ ] `CHANGELOG.md` updated (N/A)
**What this PR does**:
Adds our first UI integration test using
[Playwright](https://playwright.dev/) and runs the test on CI. Right now
the test:
- logs into Grafana
- configures the plugin (if it isn't already)
- creates an OnCall schedule, where the current user will be OnCall
- creates an escalation chain to notify based on the newly created
OnCall schedule
- creates a webhook integration, attached to the created escalation
chain
- sends a demo alert for the new integration
- goes to the alert groups page and validates that the escalation step
to alert the OnCall user actually happened
Currently the Playwright tests are run against the 3 default headless
browsers, chromium, Firefox, and webkit. The CI job that runs these
tests is run as a matrix against 3 tagged versions of `grafana`; `main`,
`latest`, and `9.2.6`.
Secondly, it adds most of the logic for a second test which:
- logs into Grafana
- configures the plugin (if it isn't already)
- goes to the user's settings, verifies their phone number (using a tool
called [MailSlurp](https://www.mailslurp.com/))
- configures the current user's default escalation policy to send alerts
via SMS
- creates an escalation policy and configures it to send alerts to our
current user
- creates an integration and assigns the created escalation policy
- triggers a test alert + verifies that we receive the SMS alert text
(again, using MailSlurp)
**Which issue(s) this PR fixes**:
Closes#873
**Checklist**
- [x] Tests updated
- [ ] Documentation added (N/A)
- [ ] `CHANGELOG.md` updated (N/A)
