BLOG
BlogGuides

Run your key journeys against production every night

Deploys are not the only thing that breaks an app. Email providers, payment settings and third-party scripts change too. A nightly schedule catches what no deploy caused.

Breakage without a deploy#

Post-deploy tests catch what your change broke. A lot of breakage comes from changes that are not yours:

  • An email provider rotates a setting and confirmation emails stop.
  • A payment provider deprecates a field.
  • A third-party script (chat, analytics, consent) starts throwing.
  • A certificate, a DNS record or a feature flag changes.
  • Data grows until a page that was fast times out.

None of these goes through your pipeline, so nothing runs your tests. A schedule does.

Add a schedule#

On a test's page, under When it runs, choose Add a schedule: the days (Monday to Sunday) and a time, in your browser's time zone. The default is every day at 02:00. A test can have up to five schedules, from the schedules docs.

A test page: the last result, its runs, what it checks and when it runs.
A test page: the last result, its runs, what it checks and when it runs.

Over the API or MCP, create_schedule also takes a day of the month (1 to 28) for monthly checks:

{ "time": "02:00", "days": ["mon", "tue", "wed", "thu", "fri"], "timezone": "Europe/London" }

Each schedule starts within nine minutes after its time, so tests set for the same minute do not all start at once.

Choose what runs at night#

Night runs are for journeys whose failure you want to know about before customers wake up:

  1. Sign-up with the email confirmation. It depends on your email provider, which changes without you.
  2. Sign-in and the dashboard. The first thing every customer does.
  3. Checkout in test mode, or as far as the payment page on production.
  4. The one page that makes money: the booking form, the quote request, the export your customers pay for.

Point production tests at made-up accounts and avoid steps with real side effects (orders, emails to real people).

Hear about it#

Scheduled runs that fail, are blocked or are inconclusive email the workspace's owners and admins by default. Send them to a Slack channel instead, or as well: Send test failures to Slack. A failed run in the morning comes with its replay, so whoever reads it can see what broke without reproducing it.

What it costs#

Each scheduled run counts toward your plan like any other. Four tests every night is about 120 runs a month. If a schedule could not start because the test's previous run was still going, the test page says so ("Last time: skipped").

This post is part of Release testing for product teams.

Filed underWebReleases and regressionsCI, schedules and agents