# A deploy broke something important

> When a release breaks a key flow like checkout, find the failing API, the error and the deploy behind it, roll back or fix, and let Relia confirm the fix on real traffic.

Section: Find and fix real problems · Page: https://tryrelia.com/docs/deploy-broke-checkout · Last checked against the product: 2026-11-01
Video walkthrough: https://youtu.be/rRvqe6r_WRU

A new version goes out and minutes later something important stops working: orders fail, sign-ups fail, payments fail. This page shows how Relia ties the failure to the release, and how to confirm your fix.

The example is Relia's demo shop (you can open it at `https://app.tryrelia.com/demo`): at 10:00 the shop deploys api 1.11.0, and placing an order starts failing.

## What you'll see

- **Home** says something needs you, with a headline such as *POST /api/orders is failing* and how many users or requests it hit. The part where it started (usually **APIs**) shows **Needs you**.
- You get an alert on the channels you chose (e-mail on every plan; Slack, ClickUp and Jira on Growth and Pro). See [Choose how Relia reaches you](https://tryrelia.com/docs/notifications.md).
- Relia raises it as **Needs you** only once it has seen the failure at least 3 times, so one stray error doesn't page you.

Relia notices a failing API when its error rate is well above its own normal for that route. A route that normally fails 0.4% of the time and now fails 35% stands out at once.

> **Note:** On the Free plan an incident shows its headline and numbers. The evidence, the cause chain and **Do this next** are on Growth and Pro (and during the 14-day Pro trial).

## Find the cause

1. **Open the incident** from Home. **What happened** gives the numbers, for example *POST /api/orders failed 106 of 302 requests (35%), normally 0.4%*.
2. **Read How it unfolded.** It starts with the likely origin. When a deploy landed just before the failures, **Recent changes** shows it, for example *api deployed 1.11.0 (was 1.10.1)*.
3. **Read the error.** The incident names the exception and, when the stack trace has it, the file and line (for example a `KeyError` in `service.py`, line 88).
4. **Open a failing request** from the evidence to see the full trace: every step of the request and where it broke.
5. **Read Do this next.** When a backend failure starts right after a deploy, it says: *This started right after api 1.11.0 was deployed: check that release or roll back to 1.10.1.*

> **Tip:** watch [How to read a production incident](https://youtu.be/5PDrogf7ZJU) (2 min).

### How Relia knows about your deploys

Relia sees a deploy when a service starts reporting a new version. The Node.js and Python SDKs read the version from `RELIA_RELEASE`, `SERVICE_VERSION` or `OTEL_SERVICE_VERSION`, else from your host's commit variable (for example `GITHUB_SHA`, `VERCEL_GIT_COMMIT_SHA`, `RENDER_GIT_COMMIT`), else from `.git` or your `package.json` (Python: `pyproject.toml`) version. If your version never changes between releases, set `RELIA_RELEASE` in your deploy, or report deploys to Relia yourself. See [Report changes](https://tryrelia.com/docs/report-changes-api.md).

## Fix it

You have two choices:

- **Roll back** to the previous version named in **Do this next**. This is the fastest way to stop the damage.
- **Fix forward.** Click **Show how to fix** on the incident (Growth and Pro). It writes concrete steps from the evidence: where (file and line, table or setting) and what to change. **Copy for your AI assistant** copies a brief you can paste into Claude Code, Cursor or another assistant.

You can also fix it from your editor: connect your AI assistant to Relia and ask it to fix the incident. See [Fix from your editor](https://tryrelia.com/docs/from-your-editor.md).

Relia never changes your code and has no access to your repository.

## Check the fix worked

You don't need to tell Relia about the new deploy. When a new version of the affected service goes live after the incident began, Relia picks it up and starts checking your fix on real traffic. If you fixed it some other way (a config change, a rollback that reuses an old version), click **I fixed it** on the incident.

The incident's fix box then shows **Checking your fix**. Relia confirms the fix once the problem stays away for at least 15 minutes with at least 20 requests on the affected APIs. Then:

- The box says **Fix confirmed**, the incident is resolved, and Home's **This week** lists *Fix confirmed*.
- You get a message: *Your fix worked*.

If the problem shows up again after your fix, you get *Still happening after your fix* instead. If there isn't enough traffic within 24 hours to tell, the box says **Not confirmed yet**.

> **Tip:** watch [From alert to verified fix](https://youtu.be/gtORFurIkow) (2 min).

## Next

- [Did my deploy break anything?](https://tryrelia.com/docs/did-my-deploy-break.md)
- [Fix it and confirm the fix](https://tryrelia.com/docs/fix-and-verify.md)
- [Teach Relia what matters](https://tryrelia.com/docs/teach-relia.md): mark checkout as something that must never break.
- [Write a postmortem](https://tryrelia.com/docs/postmortems.md)

---
All Relia docs for AI assistants: https://tryrelia.com/llms.txt (index) · https://tryrelia.com/llms-full.txt (everything in one file)
