I Audited 50,669 n8n Runs. 85% of Failures Alerted Nobody.

A reporter examined the error handling in n8n, a workflow automation tool, after discovering that 85% of failures were going unnoticed in their production instance. After auditing 50,669 runs over a six-day period, the reporter found that 106 executions failed, with a failure rate of 0.21%. However, 90 of those failures (85%) occurred in workflows without any error handling.

The reporter discovered that error handling in n8n is an opt-in feature, requiring a specific checkbox to be enabled for each workflow. This led to a decay in error handling coverage, with the reporter's instance only having 43% error handling coverage. The longest period between failures was 70.9 hours for a WhatsApp bot gate, which went unnoticed for three days.

The reporter found that most of the unguarded workflows failed multiple times, with the longest window being 1.2 hours. This highlighted the value of the Error Trigger node, which alerts via Telegram when an error occurs. However, the reporter found that all three of their error handlers were unguarded, meaning that the same 13 failures went unnoticed.

To address these issues, the reporter suggested four key points. First, a single error handling workflow should be used instead of multiple handlers. Second, coverage should be audited regularly using a query against the n8n Postgres database. Third, the handler should be loud about the workflows it monitors, not just the errors. Fourth, the handler should be monitored from outside n8n to catch any failures that occur before an execution starts.

The reporter emphasized that while these changes may seem simple, they can significantly improve the reliability of error handling in n8n workflows.