n8n: METRIC_MAP contains three entries that don't match n8n's actual Prometheus output

[kerr-bighealth]

Summary

Three entries in n8n/datadog_checks/n8n/metrics.py reference Prometheus metric names that n8n does not emit. As a result, the corresponding DataDog metrics never receive data, and one important histogram (workflow execution duration) is silently dropped.

Affected lines

metrics.py#L74-L76:

'workflow_executions_active': 'workflow.executions.active',
'workflow_executions_duration_seconds': 'workflow.executions.duration.seconds',
'workflow_executions': 'workflow.executions',

What n8n actually emits

Verified against n8n v2.19.4 (/metrics endpoint, with N8N_METRICS=true):

# HELP n8n_workflow_execution_duration_seconds Workflow execution duration in seconds.
# TYPE n8n_workflow_execution_duration_seconds histogram
n8n_workflow_execution_duration_seconds_bucket{...}
n8n_workflow_execution_duration_seconds_sum{...}
n8n_workflow_execution_duration_seconds_count{...}

# TYPE n8n_workflow_started_total counter
n8n_workflow_started_total{...}

# TYPE n8n_workflow_failed_total counter
n8n_workflow_failed_total{...}

# TYPE n8n_workflow_success_total counter
n8n_workflow_success_total{...}

# TYPE n8n_active_workflow_count gauge
n8n_active_workflow_count 40

No n8n_workflow_executions_* (plural) metric is emitted.

Diagnosis

• L75 workflow_executions_duration_seconds — misspelling. n8n emits the singular workflow_execution_duration_seconds, consistent with Prometheus convention (singular subject + plural unit, e.g., http_request_duration_seconds). This is the bug with the highest user impact: workflow_execution_duration_seconds is the canonical workflow latency histogram.
• L74 workflow_executions_active — n8n does not emit this. The closest concept (active workflows) is already mapped at L9 as active_workflow_count → active.workflow.count.
• L76 workflow_executions — n8n does not emit a single counter for executions; it splits this into workflow_started_total / workflow_failed_total / workflow_success_total, all already correctly mapped at L77–L79.

Impact

With the n8n integration enabled, the agent's debug log shows:

DEBUG | (transform.py:79) | Skipping metric `n8n_workflow_execution_duration_seconds` as it is not defined in `metrics`

The execution duration histogram is silently dropped. Workflow latency dashboards/alerts can't be built from this integration without a workaround.

Proposed fix

1. L75: rename key to workflow_execution_duration_seconds. The value should also be singular (workflow.execution.duration.seconds) to be consistent with the singular workflow_started/failed/success entries at L77–L79.
2. L74 and L76: remove. They reference metrics that don't exist.

Reproduce

# n8n container with N8N_METRICS=true
curl -s http://<n8n-host>:5678/metrics | grep -E '^n8n_(workflow|active)'

# Then on the DD agent sidecar
agent check n8n --log-level debug 2>&1 | grep workflow_execution

[AAraKKe]

Hi @kerr-bighealth, thanks for flagging this. We are looking into that and hopefully we can have an update soon.