Traefik 4xx/5xx errors in Zabbix without Prometheus

Traefik sits in front of a lot of applications: as a Docker reverse proxy, as a Kubernetes ingress controller, or as an edge router on a single VM. That makes it the best place to see whether your applications are healthy from the user’s point of view — every failed request passes through it.

Traefik exposes detailed metrics in Prometheus format. The usual advice is “add Prometheus and Grafana”. But if your monitoring is Zabbix, running a second monitoring stack just to read one endpoint is overkill. Zabbix can parse Prometheus metrics itself.

Template: custom-zabbix-templates/traefik-metrics — Zabbix 6.4 and newer, no external scripts.

What you get

For every Traefik service (whoami@docker, api@internal, myapp-web@kubernetescrd, …) the template automatically creates:

Plus two trigger prototypes per service: more than N 4xx and more than N 5xx errors within 5 minutes.

New services are discovered automatically — deploy a container with Traefik labels and it appears in Zabbix without any configuration. Services that disappear are cleaned up automatically.

How it works

1. A Zabbix agent item web.page.get[{$TRAEFIK.API}] fetches the metrics page.
2. Preprocessing converts the Prometheus text (traefik_service_requests_total{service=…, code=…}) into JSON and a small JavaScript step groups status codes into 2xx/4xx/5xx per service.
3. Low-level discovery creates items for each service; simple change preprocessing turns ever-growing counters into “requests since the last check”.

Everything happens inside Zabbix — no exporter, no cron job, no Python.

Setup

1. Enable Prometheus metrics in Traefik, ideally on a separate entry point that isn’t published to the internet:

# traefik.yml (static configuration)
entryPoints:
  metrics:
    address: "127.0.0.1:8082"

metrics:
  prometheus:
    entryPoint: metrics
    addServicesLabels: true

The same with CLI flags: --entryPoints.metrics.address=127.0.0.1:8082 --metrics.prometheus=true --metrics.prometheus.entryPoint=metrics --metrics.prometheus.addServicesLabels=true.

Check it: curl -s http://127.0.0.1:8082/metrics | grep traefik_service_requests_total.

2. Import the template in Zabbix (Data collection → Templates → Import) and link it to the host running Traefik. The host needs a Zabbix agent, because the metrics are fetched by the agent — which means the metrics port can stay bound to localhost.

3. Set the macros:

Choosing thresholds

• 5xx errors are your problem — the backend is down, timing out or crashing. Even a handful in 5 minutes is worth a look on most services.
• 4xx errors are mostly the client’s problem (404s from scanners, 401s from expired sessions), so set the threshold higher and use it to spot unusual spikes: a burst of 401/403 can mean a broken login flow or a credential-stuffing attempt.
• Look at a week of data before setting thresholds. The “Service Requests” items give you the normal traffic level for each service.

Summary

• Traefik already exposes everything you need; Zabbix can read it directly.
• Every service is discovered automatically, with requests, 4xx, 5xx and traffic.
• Keep the metrics entry point on localhost and let the Zabbix agent fetch it.
• Set the 4xx/5xx thresholds — until then, the triggers are off.

Running Traefik in Docker or Kubernetes and want proper monitoring around it? Get in touch.