Tines Academy › Lesson

Plan for failure

Keep Sue’s requests saved when notifications fail, test how the tracker handles those failures, and save step tests for the approval routing.

Objective: Define and test how a workflow handles failed notifications, and save checked examples as step tests to rerun after changes.

Keep the request when the message fails

The chat service might be unavailable, or the connector could lose permission to post. Sue can follow up on a missed notification as long as the request is still in the tracker. If the request disappears too, her team may never know someone submitted it.

Save each request before sending its notification, and make sure a failed send can’t undo the save. Track the notification separately from the request’s approval status, so operations can see which messages still need attention.

What Tines 3B can do, and what you decide

Decide what happens when a message fails

If the chat service is temporarily unavailable and the notification wasn’t sent, Sue’s tracker waits about a minute and tries once more. If that attempt fails too, it marks the message as needing attention so operations can follow up. If the tracker can’t tell whether a message was sent, it flags the message for operations to check rather than risk sending it twice.

Some failures need a different response. If the connector doesn’t have permission to post, Sue’s tracker tells operations that a connector’s access needs checking so someone can fix it.

Add failure handling

Choose the tab that matches your setup, then send the prompt in workflow chat. The prompt includes instructions for simulated message sending, which you’ll test next.

Save each vendor request before sending its notification. A failed send must not undo the save. Track notification status separately from approval status.

Handle failed notifications like this:

- If the chat service is temporarily unavailable and the message wasn’t sent, retry once after a minute.

- If the retry fails, flag the message for operations to resend without creating another request.

- If it’s unclear whether the message was sent, flag it for review without retrying automatically.

- If permission to post is missing, don’t retry. Tell operations the connector’s access needs checking.

Add two options to the draft-only test switch: the first attempt fails but the retry succeeds, or both attempts fail. After both fail, let me simulate a successful resend. These tests must not contact the chat service, and their results must be labeled as simulated.

Include a way to turn the simulations off and return to normal behavior. Keep simulated results clearly separate from real delivery results.

Keep the test controls draft-only and prevent simulations from being triggered on live. Preserve the approval rules and access checks. Leave the workflow as a draft.

Summarize what changed and explain how to use the test controls.

Prepare the tracker to handle failed notifications when I add a connector. For now, keep messages labeled "Preview only" and don't send anything. Save requests before creating their previews, and keep notification status separate from approval status.

When a connector is added, handle failed notifications like this:

- If the chat service is temporarily unavailable and the message wasn’t sent, retry once after a minute.

- If the retry fails, flag the message for operations to resend without creating another request.

- If it’s unclear whether the message was sent, flag it for review without retrying automatically.

- If permission to post is missing, don’t retry. Tell operations the connector’s access needs checking.

A failed send must never undo a saved request.

Add two options to the draft-only test switch: the first attempt fails but the retry succeeds, or both attempts fail. After both fail, let me simulate a successful resend. Label all simulated results, keep them distinct from real notification statuses, and don’t contact any chat service.

Include a way to turn the simulations off and return to normal behavior. Keep simulated results clearly separate from real delivery results.

Keep the test controls draft-only and prevent simulations from being triggered on live. Preserve the approval rules and access checks. Leave the workflow as a draft.

Summarize what changed and explain how to use the test controls.

Test failure handling

Read the build summary in workflow chat. Look for how requests stay saved, how notification results are tracked, and how to use the test controls.

These simulations don’t contact the chat service. They check the tracker’s behavior, not real message delivery.

Try a successful retry

In Browser, choose the test option where the first notification attempt fails but the retry succeeds.

Submit a request using made-up vendor details.

Check that:

The request appears in the list.

The notification shows a successful retry after 1 minute.

The result is clearly labeled as simulated.

The notification result shouldn’t change the request’s approval status.

Try two failed attempts

Choose the option where both attempts fail, then submit another test request.

The request should still be saved. Operations should be able to see that its notification needs attention, with the result labeled as simulated.

Check the saved requests

Open Files for the workflow, keeping the draft branch you’re testing selected. Open the file or database where the tracker saves requests and find both test requests by their IDs. Both should still be saved, including the one whose notification failed.

If you’re unsure where to look, ask in workflow chat:

Where does this tracker save requests, and which file or database should I open in Files?

Try a resend

Return to the list and simulate a successful resend for the failed notification.

Its notification result should now show success, still labeled as simulated. The original request should keep the same ID, with no extra request created.

Finish the test

Use the test switch to turn off the simulations.

With a connector, new requests can send real messages to your test channel again. Without a connector, messages should stay labeled “Preview only.”

Keep real delivery marked as unverified if you haven’t tested it with a connector.

If something doesn’t match

Tell Tines 3B which simulation you used, the request ID, what you expected, and what happened. Ask for a focused fix, then repeat the affected check.

Ask it to record unfinished checks in Readme.

Save a step test

A step test saves an example you’ve checked so you can try it again after making any changes. It uses recorded responses rather than contacting the service again. Next, you’ll check two approval-routing results and save them as step tests. Tines 3B runs saved step tests while it builds and before you push live.

Note: Step tests cover the examples you’ve saved. You’ll still need separate checks to confirm that the connector works and that people can only see and approve the requests they’re allowed to.

Add step tests

You’ll save two examples: one request just below $5,000 and one exactly at $5,000.

Ask for the routing tests

Send this in workflow chat:

Test the step that routes requests using amounts of $4,999.99 and $5,000.00.

Use a different made-up vendor website for each, such as routing-one.example.com and routing-two.example.com, and a test requester with an assigned manager.

Test the routing without sending notifications. Show the result of each run and name the step you tested. Don't save the step tests yet.

Check the results

The $4,999.99 request should go to operations. The $5,000.00 request should go to the requester’s manager.

Check both results before saving them.

Save the examples

If both results are correct, send:

Save those two successful runs as step tests, then run both tests and show me the results.

Open the named step in the workflow map and check that both tests show a green dot or “Passed.” After you edit the step, their results will reset to “Not run.”

If something doesn’t match

Ask Tines 3B to explain the difference and fix the step if needed. Check the results again before saving them as tests.

Don’t save an incorrect result just to make a test pass.

What to do if something still feels fuzzy

You tested how the tracker handles failed notifications and saved step tests to help catch changes to the approval routing. Keep unfinished checks in Readme; you’ll use them in the next lesson to decide whether the tracker is ready to go live.

Build it safely: You simulated notification failures and checked that the requests stayed saved. You also checked that operations could see which messages needed attention and resend them without creating another request.