Maaz Zindani

Loading the index…

↑↓ move↵ openesc close4072 pages

Help for owners

My n8n workflows fail and nobody finds out until a customer complains

Short answer

n8n can run a separate error workflow whenever another workflow fails, so it sends you an email or message. Each workflow has to be pointed at it in its settings. That alert does not tell you when n8n itself is down or a scheduled workflow never starts, so the instance needs its own health check as well.

  • Written for business owners
  • Updated
  • By Maaz Zindani, Senior Software Engineer

Does this sound like your situation?

  • A workflow was failing for days and nobody noticed.
  • I only learn about problems when a customer writes to me.
  • I check the executions list by hand every morning.
  • n8n was down over the weekend and nothing warned me.
  • I get alerts, but there are so many that nobody reads them.

Why does this happen?

  1. Alerts are not on by default

    n8n shows failed executions in its own list, but sends nothing to you. To be told, a workflow has to be set up as an error workflow. It must start with the Error Trigger node and can send, for example, an email or Slack message.

  2. Each workflow must be connected to it

    The error workflow is chosen per workflow, under the workflow settings. One error workflow can serve many workflows, but a new workflow that nobody connected stays silent.

  3. Some failures carry less information

    The alert holds the execution link and ID only if the execution was saved to the database. If the failure is in the trigger of the main workflow, the workflow never ran, and the error workflow receives different, shorter data.

  4. A workflow can be wrong without failing

    A run that finishes with the wrong result is not a failure to n8n. The Stop And Error node lets a workflow fail on purpose under conditions you choose, which then starts the error workflow.

  5. The whole instance needs its own check

    An error workflow cannot report that n8n is down. n8n offers a health endpoint that returns 200 when the instance is reachable, and a readiness endpoint that also confirms the database is connected and migrated. Metrics are off by default and not available on n8n Cloud.

What can I do about it today?

  1. Decide who should be toldOne named person or a shared mailbox that someone really reads. An alert that goes to nobody is the same as none.
  2. List the workflows that must never fail silentlyOrders, invoices, payments and customer messages. These get alerts first.
  3. Ask whether an error workflow existsOpen a few important workflows and look in their settings for an error workflow. If the field is empty, you have found the gap.
  4. Check the last week of failuresLook at the executions list for failed runs and note which workflows they belong to.
  5. Ask what watches the server itselfIf the answer is nothing, a simple outside check of the health address is the cheapest fix.

When should I bring someone in?

  • A failure cost you money or a customer before you found it.
  • Alerts exist but arrive in the hundreds.
  • n8n has been down and nobody was told.

How do I handle it?

  1. Rank the workflowsCritical ones first, so alerts match the cost of a failure.
  2. Build one error workflowIt sends a short message that names the workflow, the failed step and a link, to the right person.
  3. Connect every workflowI go through each one and set the error workflow, and make it part of the checklist for new workflows.
  4. Add the checks n8n cannot do itselfAn outside check of the health and readiness address, and a check that important scheduled workflows have run.
  5. Cut the noiseRepeated alerts are grouped and fixed at their cause, so the real ones stand out.

The price depends on what I find, so I quote in writing after a short discovery call. Tell me what is happening and I reply within 24 hours.

What do owners ask about this?

Will I be told if n8n itself stops?

Not by an error workflow. You need an outside check of n8n's health address.

Can the alert go to Slack or email?

n8n's documentation gives email and Slack messages as examples of what an error workflow can send.

Does a successful run mean the result was right?

Not necessarily. n8n only knows whether a node errored. Checks for wrong results have to be designed into the workflow.

Sources

How was this page made?

I drafted this page with AI assistance from the official documents listed under Sources. Found a mistake? Tell me.

For your developer: the technical fixes

Next step

Want this off your desk?

Send me a few lines on where things stand. I reply within 24 hours.