Help › Recipes
Catch a broken contact flow after a change
Hear about a jump in flow errors before your callers tell you.
This is for admins who publish changes to contact flows and want a safety net. You'll end up with a rule that pages someone when flows start taking their error branch far more than usual.
What Standby can see
The rule counts how many times any contact flow took its error branch, or failed to run, in the last 5 minutes. It uses Amazon Connect's own figures, read about every 5 minutes. The count covers the whole instance: it can't watch one flow, and the alert doesn't say which flow failed. It also doesn't know when you publish a change, so it watches all the time.
What you'll set up
- One alert rule on contact flow errors, with a threshold above your normal level.
Steps
- Go to Rules and choose New rule. Name it something like "Flow errors spiking".
- In Watch for, choose Contact flow errors (whole instance). There's no queue to choose; the form says the count covers the whole instance.
- Set Above to a starting threshold. The form suggests 5 errors in 5 minutes. Many flows take an error branch now and then on purpose, so expect to adjust it.
- Leave For at least on Any single reading. A broken flow fails every caller, so one bad reading is enough.
- Under When this fires, choose Push, then call, or Push, then call, even during quiet hours if a broken flow at night should wake someone. In Pages, choose the rotation that can fix flows. Choose Create rule.
Try it safely first
Set this up well before your next change, and let it run Pending for the full 7 days. Choose Review to see how often it would have paged, and open the list of firings to see each reading. If it fired on ordinary days with no change behind it, raise the threshold above those readings. Changing the threshold restarts the 7 days, and on a live rule it sends the rule back to Pending, so settle it before your next change. Then choose Promote to live.
Simulate doesn't cover flow errors. To check who it would reach, book a practice drill on the rule instead.
Variations
- Very sensitive, never a call. A second rule with a low threshold and Push only — no call, ever tells you about smaller blips without ringing anyone.
- Watch the queues too. A broken flow often shows up as callers who never reach a queue, or as callers piling up in one. A rule on Contacts waiting in a queue or Longest wait catches the second. See Get alerted when a caller waits too long.
- Fix the flow, then resolve. Standby never closes an incident by itself. When errors drop back, paging pauses; resolve the incident once you've confirmed the fix. See When you're paged.