Maaz Zindani

Loading the index…

↑↓ move↵ openesc close4072 pages

Help for owners

My n8n workflows hang, stay running or never finish

Short answer

n8n has no workflow timeout by default, so an execution waiting on a slow service can run indefinitely. Long runs are only stopped after the current node finishes, and when concurrency is capped, extra executions wait in a first-in first-out queue. Memory and a very full execution history also freeze instances.

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

Does this sound like your situation?

  • Executions show running for hours.
  • New runs wait and nothing starts.
  • The editor is slow or does not load while workflows run.
  • A workflow works by hand but not on its schedule.
  • Everything unblocks after a restart, then jams again.

Why does this happen?

  1. No timeout is set

    n8n's execution timeout setting defaults to minus one, which means no limit, and a separate maximum can be set per workflow. A call to a service that never answers can hold an execution open.

  2. A timeout stops work only after the current node

    In the main process n8n applies a soft timeout that takes effect after the node that is running finishes. A node stuck inside a slow call keeps going until it ends.

  3. A concurrency cap makes the others wait

    On self-hosted instances an optional cap limits simultaneous production executions. The rest wait in a queue and run first in, first out. Queueing is only visible in the logs, and queued executions cannot be retried.

  4. Memory pressure makes everything slow

    Large data, heavy nodes and many parallel runs can exhaust memory and show up as lost connection or stalled editor.

  5. A huge execution history

    Saved execution data can grow the database until storage runs out. Pruning is on by default, and executions that are running or waiting are not eligible for pruning.

What can I do about it today?

  1. Note which workflows are stuckName, start time and what the last finished node was.
  2. Check the service the stuck node talks toIf it is down or slow, that is the cause. Do not restart n8n over it.
  3. Check whether the work was partly doneBefore you run anything again, look whether orders, emails or invoices were already created. A re-run can repeat them.
  4. Cancel only clearly stuck executionsUse the stop button on an execution that has run far beyond normal. Leave everything else alone.
  5. Restart once, only if neededA single restart to free the instance is acceptable once you know what was running. Tell your developer.

When should I bring someone in?

  • Stuck runs repeat every day or week.
  • A re-run could duplicate payments, emails or orders.
  • You do not know whether work was lost.

How do I handle it?

  1. Identify the node that never returnsFrom the executions list and logs I find the step where each stuck run stopped, and the service behind it.
  2. Set sensible timeoutsA workflow timeout and a maximum are set, and slow external calls get their own time limit and retry rules.
  3. Make re-runs safeSteps that create records check first whether the record exists, so a repeated run does not duplicate.
  4. Tune concurrency and memoryA concurrency cap is considered where parallel runs overload the server, and heavy workflows are cut into batches.
  5. Alert on long runsA message is sent when an execution runs longer than expected, so it is found in minutes rather than days.

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?

Why does a restart help?

It frees memory and stops runs in progress. It does not fix the reason, and the same runs may return, so the cause still needs finding.

Will cancelled executions be run again?

n8n says queued executions cannot be retried, and cancelling one removes it from the queue. Anything cancelled has to be started again by hand or by its trigger.

Is a concurrency limit safe to set?

It stops overload, but executions beyond the cap wait, so slow workflows can delay fast ones. It is set after looking at your workload.

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.