n8n Error Workflow Not Triggering? We Tested 16 Cases

automation
A white smoke detector on an olive plaster wall with its battery door hanging open and the battery slot empty, while a black 9-volt battery sits on an oak shelf below, a metaphor for an n8n error workflow that exists but never fires
Most n8n error workflows that never fire are installed correctly and simply not powered: the handler isn't published, the failure was swallowed, or the run was manual.

If your n8n error workflow isn't triggering, check three things first: the failing run was a production execution and not a manual test, the execution really ended in error instead of a node continuing past it, and the error workflow itself is published. We tested 16 failure cases on n8n 2.46.6 on November 8, 2026, in a throwaway Docker container. The handler stayed silent in 7 of them. The surprise was the third check. n8n's own docs say an Error Trigger workflow doesn't have to be published, but a handler we never published didn't run once.

Every result below comes from that run. Statuses and log lines are copied from the container's database and log file.

How does the n8n error workflow actually get called?

When an execution finishes with a status other than success or waiting, n8n looks at the failing workflow's settings. If an Error Workflow is selected there, n8n loads that workflow, finds its Error Trigger node, and starts a new execution in error mode. That execution receives the failed workflow's name and ID, the execution ID and URL, the last node that ran, and the error message.

If no Error Workflow is selected but the failing workflow contains an Error Trigger node itself, n8n uses the failing workflow as its own handler. We confirmed this as well: a published workflow with a webhook, a node that throws, and a separate Error Trigger branch handled its own failure.

That's the whole mechanism. Each cause below breaks one link in it.

What we tested

We built 19 small workflows: three handlers, 13 failing workflows behind production webhooks, one failing workflow with a Manual Trigger, and two sub-workflows. Each failing workflow had a Code node that throws an error, or a Stop and Error node. We published them with the n8n CLI, started the server, called each production webhook once, and then read execution_entity and the debug log.

Here's what fired the handler:

  • Plain error in a published workflow: fired once.
  • Retry On Fail, 3 tries, all failed: fired once, after the last try.
  • Stop and Error node: fired, with our custom message.
  • Workflow timeout (2 s limit on a 6 s node): fired. The execution was saved as canceled, not error.
  • Save failed production executions set to "Do not save": still fired.
  • Workflow with its own Error Trigger and no setting: handled itself.
  • Parent fails because a child sub-workflow failed: fired for the parent.

And here's what stayed silent:

  • Handler never published: no run.
  • Handler unpublished after it had been working: no run.
  • Manual execution from the editor: no run.
  • On Error set to Continue: no run. The execution was saved as success.
  • On Error set to Continue (using error output): no run. Also saved as success.
  • Error Workflow setting pointing to an ID that doesn't exist: no run.
  • Handler's caller policy set so no other workflow may call it: the handler started and failed straight away, so no alert went out.

One case behaved differently from what most people expect. A draft edit to the handler didn't take effect. The failure ran the old published version until we published again.

Cause 1: You're testing with a manual run

This is the most common one by far, and n8n's documentation says it plainly: you can't test error workflows when running workflows manually. In our run, clicking Execute workflow on a failing workflow saved an execution with mode manual and status error, and the handler never started. The source is explicit. The lifecycle hook only calls the error workflow when the mode isn't manual.

Fix: publish the failing workflow and trigger it the way production does. For a webhook, call the production URL (/webhook/...), not the test URL (/webhook-test/...). For a schedule, wait for the next tick or set a temporary one-minute schedule.

Cause 2: The error workflow isn't published

The Error Trigger docs currently say: "If a workflow uses the Error Trigger node, you don't have to publish the workflow." That didn't match what we saw on 2.46.6. The workflow failed, n8n tried to start the handler, and logged this at error level:

Calling Error Workflow for "t02unpub000001". Workflow "ehdraft0000001" is not active and cannot be executed

The code matches the log. loadErrorWorkflowData fetches the handler's published version and returns nothing if there isn't one. When we unpublished a handler that had been working a minute earlier, every workflow pointing to it went silent the same way.

Fix: open the handler and publish it. An Error Trigger counts as a trigger, so the button is there. Then look at the third case below, because publishing once isn't the end of it.

Cause 3: You edited the handler but didn't republish

n8n 2.x separates the saved draft from the published version, and error workflows follow the same rule. We changed the handler's Code node from v1 to v2-draft and saved without publishing. The next failure still produced "handler": "v1". After we published, the next one produced v2-draft.

This is how teams end up with a handler that "works" but sends alerts to an old Slack channel or uses an expired credential for weeks. Republish the handler after every edit, and check that the alert actually arrives.

Cause 4: Continue On Fail swallowed the error

If the node that fails has On Error set to Continue or Continue (using error output), the execution doesn't fail. In our test both variants were saved with status success. With the first, the next node got an item holding the error message. With the second, the error went to an output branch that nothing was connected to. The response we got back was No item to return was found.

From n8n's point of view, nothing went wrong, so there's nothing to report. Use Continue only when another branch handles the problem, and make that branch send the same alert your error workflow would. Retry On Fail doesn't have this problem: after the third failed try, the handler fired exactly once.

Cause 5: The Error Workflow setting points to the wrong ID

The setting stores a workflow ID, not a name. Import workflows from another instance, restore from a backup, or rebuild the handler, and that ID can point to something that doesn't exist. In our test the handler simply never ran. The log line is confusing, too. A missing ID produces the same message as an unpublished one:

Calling Error Workflow for "t09missing0001". Workflow "doesnotexist01" is not active and cannot be executed

Fix: after any import or migration, open each production workflow's Settings and re-select the handler from the dropdown. If you manage dozens of workflows, pull them with the n8n API and check that settings.errorWorkflow matches your handler's current ID. This is a step on our upgrade and rollback checklist for self-hosted instances.

Cause 6: The handler's caller policy blocks the call

The handler's own Settings include This workflow can be called by. n8n checks that policy before it runs a handler, just as it does for sub-workflows. We set a handler to allow no callers. The failing workflow triggered it, and the handler's execution list showed a run that failed straight away with this error:

Another workflow: (ID t13policy00001) tried to invoke this workflow to handle errors.

The main log only recorded it at info level: Error workflow execution blocked due to subworkflow settings. This one bites agencies that run several projects on one instance, because a handler in one project may not accept calls from workflows owned by another.

Fix: set the handler to accept the workflows that point to it, or keep one handler per project.

What happens when a sub-workflow fails?

A sub-workflow is a workflow in its own right, so its Error Workflow setting is separate from the parent's. Here's what we saw:

  • Child without a setting, parent with one: one alert. It named the parent workflow and the Execute Workflow node as the failed node. The message was the child's error, boom from T08 child no setting.
  • Child and parent both set: two alerts for one failure. The first came from the child in mode integrated, the second from the parent in mode webhook.

Neither case is wrong, but the second one doubles your alert volume. Set the Error Workflow on the parent and leave it off sub-workflows that are only ever called by a parent. Set it on a child when it also runs on its own trigger. If double alerts are already a problem, dedupe in the handler on the execution chain. We use the same idea for duplicate webhook executions.

What an error workflow will never catch

Even when it's set up perfectly, an error workflow only reacts to executions that end in failure. It won't tell you about:

  • Workflows that never start. A webhook URL that the sender stopped calling, a schedule on an instance that was down, or a trigger that was unpublished produce no execution, so nothing fails.
  • Executions stuck on running. A run that hangs never finishes, so the handler is never called. Our guide to n8n executions stuck in running covers how to catch those.
  • Wrong but successful output. An AI node that returns nonsense, or an API call that returns 200 with an empty body, is a success as far as n8n is concerned.

For those, you need a heartbeat check that alerts when a workflow hasn't succeeded for longer than normal. We explain how we layer retries, one shared error workflow and heartbeats in our n8n error handling and monitoring guide.

The 5-minute checklist we run

When a client tells us their alerts went quiet, we go through these in order:

  1. Reproduce in production. Trigger the failing workflow through its production webhook or schedule, never with Execute workflow.
  2. Read the failed execution. If it says Success, find the node set to Continue.
  3. Open the handler. It must start with an Error Trigger, it must be published, and the published version must be the one you think it is.
  4. Open the failing workflow's Settings. Re-select the handler under Error Workflow, especially after an import.
  5. Check the handler's caller policy if workflows live in more than one project.
  6. Search the logs. Calling Error Workflow and Error workflow execution blocked tell you exactly why it didn't run.

Then make the handler useful once it does fire. Put the workflow name, the failed node, the error message and the execution URL in the alert. That URL comes in the Error Trigger's output, and it saves the first five minutes of every incident.

Want someone to check your alerting before the next silent failure?

Most instances we review have an error workflow. Far fewer have one that's published, current, actually selected on every production workflow, and backed by a heartbeat. Get a Free Automation Audit: we'll go through your n8n instance and email you a free initial audit report within 24 hours, listing which workflows would fail silently today. If you'd rather hand the whole thing over, see how we build and run n8n workflow automation for clients.