The most common failure in production n8n is not a code bug. It is a credential that stopped working — usually an OAuth token that expired while the workflow kept trying to use it. The workflow fails, the error message says something about authorisation, and if you run workflows for clients, the client finds out before you do.
This page covers why it happens, what the error messages actually mean, how to fix each case, and how to stop being the last person to know.
OAuth uses two tokens:
n8n handles access token refresh automatically — you do not need to do anything for that.
The problem is when the refresh token itself dies. Then n8n cannot get
a new access token, and every API call fails with a 401 or invalid_grant.
| Service | Refresh token lifespan | Common trap |
|---|---|---|
| Google (Testing mode) | 7 days | Google Cloud app is still in "Testing" — the #1 cause of weekly re-auth |
| Google (Published) | 6 months of inactivity | A workflow that runs daily is fine; one that runs monthly may lose its token |
| Microsoft | 90 days (default) | Password change invalidates all tokens immediately |
| Slack | Does not expire | User revoking the app in Slack settings |
| HubSpot | 6 months | Changing OAuth app scopes invalidates existing tokens |
n8n credential errors come in several shapes. They all mean roughly the same thing, but the fix depends on which one you see:
| Error message | What it means | Fix |
|---|---|---|
invalid_grant | Refresh token expired or revoked | Reconnect the credential from scratch |
401 Unauthorized | Access token invalid and refresh failed | Reconnect the credential |
Unable to sign without access token | n8n tried to sign the request but had no token at all — it failed before the API call was ever made | Reconnect or delete and recreate the credential |
Token has been expired or revoked | Google explicitly says the token is dead | Reconnect + check Google Cloud testing status |
credentials not found | The credential was deleted or never finished setup | Create a new credential and select it in the node |
Unable to sign without access token has no HTTP status code.
It happens before n8n sends the request, at the signing step. If you are matching on
HTTP 401 to detect dead credentials, you will miss this entire class. It is one of the
most common Google OAuth failures in real production data.
Fixing a credential takes 2 minutes. The damage is not the fix — it is the 3 days between the credential expiring and someone noticing. In those 3 days, the workflow keeps failing, the client's data stops updating, and the first person to notice is the client, not you.
The client doesn't experience three weeks of silent failure as three weeks. They experience the thirty seconds where they ask what happened and you don't have an answer.
An error workflow helps — but only for the failures that throw an error. Some credential failures are silent (the node continues after error), and an error workflow cannot tell you whose credential expired or which client is affected.
This is the part that matters if you run workflows for clients. A dead credential is not always the client's fault, and it is not always yours:
| Situation | Whose side | What to do |
|---|---|---|
| OAuth token expired (was working, stopped) | Not your bug | Re-authorise in n8n, notify client if their action caused it |
| Credential was never finished (setup incomplete) | Your setup issue | Open the node, complete the OAuth flow |
| Client changed their password | Client's action | Tell the client, re-authorise together |
| Client revoked the app | Client's action | Tell the client, re-authorise together |
| Google Cloud app still in "Testing" | Your setup | Publish the app, re-authorise |
Getting this distinction right — quickly, without digging through logs — is the difference between "your OAuth token expired on the 4th, here is the trail" and "I am not sure yet." The first keeps the relationship. The second turns a technical failure into doubt about whether the agency is in control.