mirror of
https://github.com/hanzoai/insights.git
synced 2026-10-30 16:15:31 +00:00
zeekay
40d120dbf7
Every first-party Python caller and test now names the mount the API actually answers on. Mechanical: URL literals only, and only where the /api/ was ours -- a preceding word character means it was a module path like insights/api/ or a third-party host like slack.com/api/, and those are untouched. Left saying /api/, each for a reason rather than for safety: - products/warehouse_sources/.../sources/** is a catalogue of vendor endpoints - services/llm-gateway deploys separately and is under a code freeze, so moving its two calls would break it during the skew - the mcp_store and Slack/Teams redirect_uris are registered with third parties and must byte-match what was registered - the Wizard and Raycast client-metadata paths are OAuth client_ids - /api/surveys and /api/feature_flag/local_evaluation are SDK aliases the ingress routes to the Rust fleets, held by released SDK versions Test URLconfs moved with the requests that call them, and the Datastore user-selection fixtures now carry the leading slash that request.path has.
49 lines
2.2 KiB
TypeScript
49 lines
2.2 KiB
TypeScript
import { defineConfig, devices } from '@playwright/test'
|
|
|
|
/**
|
|
* The suite that runs against a DEPLOYED insights, not a dev instance.
|
|
*
|
|
* Separate from playwright.config.ts on purpose. That config's fixtures build a
|
|
* workspace through /v1/setup_test/, which only exists in TEST mode -- pointing
|
|
* it at production makes every test fail on a 404 that has nothing to do with the
|
|
* thing under test. This suite touches no setup endpoint and creates no data; it
|
|
* signs in as an existing account and reads.
|
|
*
|
|
* LIVE_URL selects the deployment. LOGIN_USERNAME/LOGIN_PASSWORD are the account;
|
|
* in CI the password comes from KMS, never from a file or a repo variable.
|
|
*/
|
|
export default defineConfig({
|
|
testDir: '.',
|
|
timeout: 90 * 1000,
|
|
expect: { timeout: 30 * 1000 },
|
|
fullyParallel: false,
|
|
forbidOnly: !!process.env.CI,
|
|
// A deployed target is shared and can be slow under load; a flake here should
|
|
// not read as a broken release. Retries are for the network, not for masking
|
|
// a failure -- a test that only passes on retry still shows up in the report.
|
|
retries: process.env.CI ? 2 : 0,
|
|
// ONE worker. This drives a real deployment serving real users, not a test rig.
|
|
// In parallel the suite was the load: scenes that render in ~2s serially missed
|
|
// a 60s deadline at four workers and again at two, and the failures moved
|
|
// between scenes each run -- the signature of contention, not of a broken
|
|
// scene. A post-deploy check that stresses the thing it is checking reports on
|
|
// the stress. Serial costs a couple of minutes and answers the actual question.
|
|
workers: 1,
|
|
reporter: process.env.CI ? [['list'], ['junit', { outputFile: 'live-results.xml' }]] : [['list']],
|
|
use: {
|
|
baseURL: process.env.LIVE_URL || 'https://insights.hanzo.ai',
|
|
actionTimeout: 30 * 1000,
|
|
trace: 'retain-on-failure',
|
|
screenshot: 'only-on-failure',
|
|
testIdAttribute: 'data-attr',
|
|
},
|
|
projects: [
|
|
{ name: 'setup', testMatch: /auth\.setup\.ts/ },
|
|
{
|
|
name: 'chromium',
|
|
use: { ...devices['Desktop Chrome'], storageState: 'live/.auth/session.json' },
|
|
dependencies: ['setup'],
|
|
},
|
|
],
|
|
})
|