Production & Error Handling
The workflows in this guide will all break in the same handful of predictable ways at some point. Knowing the pattern on sight turns a 30-minute panic into a 30-second fix — which matters a lot more once a client, not just you, is depending on the workflow running tonight.
Why this layer matters
This is not an exhaustive list of every error you'll ever see — it's the handful that account for the overwhelming majority of them. Recognizing the shape of the problem is most of the battle; the fix is almost always small once you know where to look.
Credential / authorization errors
"Authorization failed" almost always means one of two things: the credential was never created for this node, or an API key referenced by that credential was later deleted or rotated at the source (in OpenAI's or another provider's own dashboard). Re-open the credential, re-authenticate or paste in a fresh key, and re-test the single node before assuming anything else is wrong.
Data type mismatches
Covered first in Layer 00 because it is, by a wide margin, the most common error you'll hit — comparing a string "1000" against a number 1000 in a Filter, IF, or Switch condition. Two fixes:
- Enable Convert types where required on the condition, letting n8n coerce both sides automatically, or
- Explicitly change the condition's type dropdown (Number vs. String vs. Boolean) to match the incoming field, which you can confirm at a glance from its color and icon in the Input panel.
Undefined references
An expression referencing a field that doesn't exist — often because a node earlier in the chain
was renamed, deleted, or duplicated (which appends "1" to its name, silently breaking any
expression that still points at the original name) — returns undefined or
null instead of erroring loudly. Re-drag the field fresh from the current Input panel
rather than trying to patch the old reference by hand.
Invalid JSON
The most common cause, by far: a trailing comma after the last item in a JSON object or
array (every item needs a comma except the last one), or a raw line break sitting inside a
string value where JSON requires an escaped \n. Paste the exact broken JSON into
ChatGPT with "fix the syntax in this JSON" — it will find the stray comma or unescaped line break
immediately.
Timeouts & continue-on-fail
An HTTP Request or scraping call to a slow or unreachable target can hang indefinitely without an explicit limit. In the node's Settings tab:
- Add a Timeout (in milliseconds — 10,000 = 10 seconds is a sane default for most external calls).
- Set On Error → Continue so, when looping over a batch, one bad item doesn't kill the run for every other item in it.
Errors inside sub-workflows
An error surfacing from a Call n8n Workflow or agent tool is often confusingly generic in the parent workflow's log. Open the sub-workflow directly, check its Executions tab for the actual failing run, and use Debug in Editor to pull that exact failing data back onto the canvas so you can iterate on the fix using the real input that broke it, rather than guessing.
HTTP status codes, quickly
| Code | Meaning |
|---|---|
2xx | Success |
3xx | Redirect — you're being pointed at a different URL |
400 | Invalid parameters — check your request body against the API docs |
401 | Missing or expired auth token |
403 | Authenticated, but insufficient permissions |
429 | Rate limited — see the looping/wait pattern in Layer 03 |
5xx | Server-side error on the API's end — usually transient, retry with backoff |
Wire an error-handling workflow
Don't wait for a client to tell you their automation broke. Build one small workflow, trigger → Error Trigger, that sends you a Slack message or email whenever any other workflow fails. Assign it under every important workflow's Settings → Error Workflow so failures page you the moment they happen instead of surfacing days later as an angry email.
Version history & rollback
Every save is checkpointed in a workflow's version history (accessible from the clock icon in the top bar), but that window is typically 24 hours on lower-tier plans — treat it as an "oops, undo the last change" tool, not a long-term backup. For anything client-facing, keep your own periodic JSON exports (Download from the workflow menu) or use Git-based version control, available on n8n's Business and Enterprise tiers — see Layer 11.
Sources
| Source | Used for |
|---|---|
| Jono Catliff (Automatable), "n8n 8 Hour Masterclass — Build & Sell AI Agents (Automate Anything)" | Primary source for this layer's error patterns |
| n8n docs, "Handle errors gracefully" | Error workflows, continue-on-fail, retries |
| n8n docs, HTTP Request node | Timeout settings & status-code behavior |