← All questions

net::ERR_ABORTED on an authenticated route is our single biggest source of failures

page.goto: net::ERR_ABORTED at https://staging-app.example.com/workspace/<uuid>/home

await this.page.goto(`/workspace/${workspaceId}/home`, { waitUntil: 'domcontentloaded' });
await this.page.waitForURL(new RegExp(`/workspace/${workspaceId}`), { timeout: 30_000 });

3 answers

• Accepted answer

  The wait is not the problem. The goto is.

  Client-side auth attaches after the first authenticated route has loaded. Deep-linking an internal route lands before that bootstrap has finished, the client sees no user yet and issues a redirect to the login route, and the navigation you started is aborted by the one the client started. ERR_ABORTED is the precise symptom of that, and your two screenshots are the two halves of the race: blank if the abort came early, "Loading session…" if it came mid-bootstrap.

  waitForURL cannot rescue it because it runs after goto resolves, and here goto is what rejects. And a retry around the navigation would only hide it.

  There is an open issue. Nothing has been fixed, but the fix on your side is a page-object change and it is worth making now:

  • make the first authenticated navigation the only goto — enter at the app root once, let

  the session attach

  • reach the deep target the way a person does, through the switcher or the sidebar
  • if you must keep a deep link somewhere, wait on a test id that only exists once the shell has

  rendered, not on domcontentloaded

  tsCopy

  async openFixtureWorkspace(workspaceId: string) {
    if (!this.page.url().includes('/workspace')) {
      await this.page.goto('/workspace', { waitUntil: 'domcontentloaded' });
      await this.page.waitForURL(/\/workspace/, { timeout: 30_000 });
    }
    await this.workspaceSwitcher.click();
    await this.page.getByTestId(`workspace-option-${workspaceId}`).click();
    await this.page.waitForURL(new RegExp(`/workspace/${workspaceId}`), { timeout: 30_000 });
  }

  That needs a stable test id on the switcher options. If your app does not have one yet, that request is the highest-value thing you can put in front of your developers this week.

  TBanswered by Tomas BrekkeSDODS team

  Reply

• When you do this, audit every page.goto on an authenticated path, not just the one in the page object:

  bashCopy

  grep -rn "page.goto('/workspace" projects/<slug>/steps/ projects/<slug>/pages/

  The ones inside step files are the easy ones to miss, and they are on the paths that hurt — deep links into a builder or a settings surface. The single legitimate goto is the first entry after login; everything else should be navigation.

  Also worth a review rule, because a comment does not hold. We had this documented in the page object itself, nine lines above the call that broke it.

  RNanswered by Reeta NandalSDODS team

  Reply

• We added a workspace-shell test id and waited on that instead of domcontentloaded, which cut it a long way down but not to zero — the very first entry of the run still has to happen somewhere, and that one is still a goto. Doing it once per worker in a fixture rather than once per scenario is what got the last of it.

  Also: do not re-baseline anything until this is fixed. Several hundred of our scenarios go through that step, so the reported state of most UI modules was a report on this bug.

  ISanswered by Ingrid Solberg

  Reply

Have a different question? Ask it.