Cloudflare Tunnel Monitoring: Status, Metrics, Alerts
Published
- cloudflare tunnel status
- cloudflare tunnel monitoring

To monitor a Cloudflare Tunnel, combine three layers: the tunnel status Cloudflare reports, the metrics and logs cloudflared produces, and a separate check that the application behind the tunnel answers. Status alone is not enough. Cloudflare's documentation says tunnel status only reflects the connection between cloudflared and the Cloudflare network, so a tunnel can show Healthy while users cannot reach the app.
That gap matters most for MSPs and multi-site IT teams, where one tunnel often fronts a customer's internal service and a ticket arrives before any alert fires. At Netverge, we build for that kind of distributed environment: one place for monitoring, ticketing and documentation across sites and clients. This guide covers the Cloudflare side in full, then shows how to fit it into your wider operations.
TL;DR
- Tunnel status (Healthy, Inactive, Down, Degraded) tells you whether
cloudflaredis connected to Cloudflare. It does not tell you whether the origin service works. - Use the dashboard or
cloudflared tunnel listfor a quick check, the Prometheus metrics endpoint for trends, andcloudflared tailor log files for evidence. - Turn on Tunnel Health Alerts and route them by email or webhook, then add an origin-level check so a Healthy tunnel with a dead app still pages someone.
- Run at least two
cloudflaredreplicas per tunnel where uptime matters, and do not use an aggressive liveness probe in Kubernetes. - Feed tunnel signals into the same operations workflow as your other sites so one incident has one ticket and one owner.
Table of Contents
- What Cloudflare Tunnel status actually tells you
- Monitor without sending traffic through the tunnel
- Scrape the cloudflared metrics endpoint
- Use logs for evidence, not for alerting
- Set up alerts that someone will act on
- Verify the application behind a healthy tunnel
- Troubleshoot Down, Degraded and error 1033
- Design for failover so one connector is not a single point of failure
- A practical monitoring checklist
- Bringing tunnel signals into one operations view
- What Netverge offers for this workflow
- FAQ
- Sources
What Cloudflare Tunnel status actually tells you
You can see tunnel state in the Cloudflare dashboard under Networking > Tunnels, or by running cloudflared tunnel list. Cloudflare's Observability documentation defines four states:
| Status | Meaning | First thing to check |
|---|---|---|
| Healthy | The tunnel is active and serving traffic through four connections | Nothing at the tunnel layer. Verify the origin separately. |
| Inactive | The tunnel was created but cloudflared has never run |
Whether the connector was ever started on the host |
| Down | The tunnel connected before, but the cloudflared process has stopped |
Whether cloudflared is running and whether the server has problems |
| Degraded | cloudflared is running but at least one connection failed |
Logs, local network and firewall rules |
The CLI path for Workers-based setups is similar: the wrangler tunnel info command shows a tunnel's ID, name, status and creation time, and wrangler tunnel list shows the same for all non-deleted tunnels, according to Cloudflare's own docs repository.
Why Healthy does not mean working
A tunnel is really three links in a chain: the connector registering with Cloudflare, the hostname routing to the tunnel, and the tunnel reaching the origin. Cloudflare's Common errors page repeats the point that status does not confirm cloudflared can reach internal services. Only the first link feeds the status line.

For an operations team, this changes what "green" means. A green tunnel is a necessary condition. It is not proof of service. Build your alerting so that the origin check, not the tunnel status, decides whether a customer is affected.
Monitor without sending traffic through the tunnel
A common question is whether you can watch a tunnel without making a request through it. You can. Cloudflare's docs describe four methods that need no request to the public hostname:
- Status in the dashboard or via
cloudflared tunnel list. - The Prometheus metrics endpoint exposed by
cloudflared. - Remote log streaming with
cloudflared tail <UUID>. - Dashboard notifications for tunnel health and deployment changes.
The sections below take each in turn.
Scrape the cloudflared metrics endpoint
At startup, cloudflared opens a Prometheus metrics endpoint. In non-containerized environments it binds to 127.0.0.1 on the first free port in the range 20241 to 20245, and picks a random port if all five are taken. In Docker or Kubernetes it binds to 0.0.0.0. You can set your own address with the --metrics flag, per Cloudflare's documentation.
The same page lists metrics worth tracking, including:
cloudflared_tunnel_ha_connections: how many high-availability connections are up. A drop from four is your earliest sign of a Degraded tunnel.cloudflared_tunnel_request_errors: errors on proxied requests, which can point to origin trouble before users report it.- QUIC round-trip time and packet loss counters: signals for a flaky path between the site and Cloudflare's edge.
Two practical notes for multi-site fleets:
- Port drift. The first-free-port behavior was introduced in
cloudflaredversion 2024.12.1. If several processes share a host, pin the address with--metricsso your scraper does not chase a moving port. - No push. Cloudflare's Grafana tutorial states you cannot push metrics straight from
cloudflaredto Grafana. A Prometheus server scrapes the endpoint, and the dashboard reads from Prometheus. Plan for a collector that can reach each site's metrics address.
The collector question is where distributed environments get awkward. Each site's metrics endpoint is local by default, so something on that site has to read it. That is the kind of on-site visibility problem our Vergepoints are built around: a Vergepoint is edge hardware that gives you visibility at the location itself, so you are not relying on a central probe that cannot see inside the site.
Use logs for evidence, not for alerting
Logs answer the question "what happened" once a metric or a user tells you something is wrong. Cloudflare documents two ways to get them.
Remote streaming
cloudflared tail <UUID> streams a tunnel's logs remotely. Cloudflare's changelog also notes that real-time tunnel log streaming is available in the dashboard under Networking > Tunnels, which gives engineers the same live view without shell access to the host.
Server-side files
The run parameters documentation covers the options:
--log-directorywritescloudflared.logto the directory you give, rotates the file at 1 MB and keeps up to five backups. There is no age-based retention, so ship logs elsewhere if you need history.--logfiletakes precedence if both are set, and it is not rotated.--logleveldefaults toinfoand acceptsdebug,info,warn,errorandfatal.--output jsonproduces structured logs. Cloudflare's release notes show JSON output arrived in version 2025.6.1, so older connectors need an upgrade first.
Safety note on debug logging. At debug level, cloudflared records request URLs, methods, protocols, content lengths and all request and response headers, which can expose sensitive information. Use it for a bounded troubleshooting window, restrict access to the file, and switch back to info afterward. This matters most for managed environments where logs may contain customer data.
Set up alerts that someone will act on
Cloudflare supports tunnel notifications by email, webhook or third-party services, and the Tunnel Health Alert is included with all Zero Trust plans. A workable setup looks like this:
- Enable the Tunnel Health Alert for every production tunnel and send it to a webhook, not only to a mailbox.
- Route the webhook into your ticketing or chat workflow so the alert carries an owner. If your team works in Microsoft Teams, our guide to Microsoft Teams alerts covers setting them up and keeping them useful.
- Alert on the metric, not just the state. A
cloudflared_tunnel_ha_connectionsvalue below four for several minutes catches degradation that an email about a Down tunnel will miss. - Separate severity. Down on a customer's only tunnel is an incident. Degraded on a tunnel with two healthy replicas is a ticket.
- Add an origin check (next section) so the alert that matters most, "tunnel up, app down", exists at all.
If you manage many clients, the discipline is the same across all of them: one alert, one ticket, one runbook. Netverge combines monitoring, AI-powered ticketing and documentation in one platform for that reason, so a tunnel alert and the site documentation that explains it sit together.
Verify the application behind a healthy tunnel
This is the check most teams skip. Cloudflare gives two concrete routes.
Load balancer monitors. The public load balancer documentation recommends attaching an HTTPS monitor to tunnel endpoints with path /, port 443, expected status code 200 and the Host header set to the app's hostname. TCP monitors are not supported for tunnel endpoints. If you use the published HTTP_STATUS health-check route instead, it verifies that cloudflared is reachable but not that the upstream service accepts requests. Pick the HTTPS monitor when you need to know the app responds.
Origin troubleshooting. For HTTPS origins, Cloudflare's troubleshooting page tells you to check the route's Service URL, the origin certificate and the SSL/TLS encryption mode. It also tells you to read cloudflared logs for connection refused, malformed HTTP responses, x509 errors and TLS handshake errors. Each of these can occur while the tunnel shows Healthy.
Certificates deserve their own watch. An expired or mismatched origin certificate produces exactly the "connected but not working" symptom, which is why our certificate expiry monitoring guide belongs next to your tunnel checks. DNS is the other usual suspect, and the 60 second DNS monitoring checklist covers silent failures there.

Troubleshoot Down, Degraded and error 1033
Use the status to choose where to start.
Down
Confirm that cloudflared is running on the host, then look for server-side problems such as a stopped service, a rebooted machine or a full disk. Cloudflare's Observability page gives this order.
Degraded
Review the logs and investigate local network and firewall rules, since at least one of the four connections failed. If the same site also shows packet loss or latency, correlate the two. Our guide on correlated packet loss monitoring shows how to do that without drowning in alerts.
Error 1033
This error means Cloudflare's network cannot find a healthy cloudflared instance for the traffic. Cloudflare's troubleshooting guide says to confirm the tunnel is Active in the dashboard or with cloudflared tunnel list. If it is, verify that the DNS record points to the correct tunnel subdomain.
Upgrade and capture debug output
The same page notes that many tunnel issues are resolved by upgrading to the latest cloudflared. To capture evidence, run cloudflared tunnel --loglevel debug run, add --logfile to persist the output, and attach it when you contact Cloudflare support. Since version 2024.12.2, the cloudflared tunnel diag command also collects troubleshooting information in one step.
Idle-stream timeouts
One field report shows a healthy 4/4 tunnel still logging "context canceled" errors because the HTTP/2 transport timed out idle origin streams after about 60 seconds. The reporter switched the transport to QUIC and saw the errors fall. Treat it as a lead to test in your own environment, not a rule.
Design for failover so one connector is not a single point of failure
Cloudflare's load balancer documentation recommends at least two cloudflared replicas per tunnel in each data center for active-passive failover. It also notes that the load balancer treats replicas of the same tunnel UUID as a single endpoint, so you need separate tunnel UUIDs if you want per-host session affinity.
For monitoring, this means you track two things per site: the tunnel as a whole and each replica. If one replica goes down and the tunnel stays Healthy, you still want a ticket, because you have lost your redundancy. To test that failover really behaves as designed, borrow the method in our repeatable WAN failover tests guide: define the metric, trigger the failure, measure the recovery.
Kubernetes: avoid the probe restart loop
In Kubernetes, a liveness probe with failureThreshold: 1 on the cloudflared /ready endpoint can cause a restart loop. Cloudflare periodically rotates edge connections, so a brief non-200 response is normal. One project's commit fix dropped the liveness probe and used startup and readiness probes instead. For the wider picture of what to track in clusters, see our Kubernetes network monitoring guide.
A practical monitoring checklist
Use this per tunnel, and repeat it per client or site.
| Layer | Signal | Where it comes from | Alert when |
|---|---|---|---|
| Tunnel | Healthy, Degraded, Down, Inactive | Dashboard, CLI, health notification | Anything but Healthy beyond your grace period |
| Connector | HA connections, request errors, QUIC loss | Prometheus endpoint | Connections below four, or error rate rising |
| Logs | Origin errors, TLS and x509 failures | cloudflared tail, log files |
Repeated origin errors |
| Origin | HTTP 200 from the app hostname | HTTPS monitor | Any failure, regardless of tunnel state |
| Dependencies | DNS record, certificate expiry | DNS and certificate checks | Mismatch or expiry window reached |
| Version | cloudflared release in use |
Fleet inventory | Connector behind current release |
The last row is easy to forget. Because upgrading resolves many tunnel problems, a fleet inventory that shows which sites run old connectors is cheap insurance, and it is documentation work as much as monitoring work.
Bringing tunnel signals into one operations view
Cloudflare's tools tell you about Cloudflare's side. Your customers, though, experience the whole path: the site network, the host, the connector, the edge, the origin. When these signals live in separate consoles, the person on call has to correlate them by hand at 2 a.m.
Netverge is designed to remove that fragmentation for MSPs and multi-location enterprises. The platform unifies real-time infrastructure monitoring, anomaly detection, ticketing, network documentation and AI-assisted troubleshooting, with multi-tenant management and role-based access for teams that serve many clients. Our hybrid path work, described in the hybrid cloud network monitoring guide, applies the same idea to links between sites and cloud services.
On automation: any workflow that restarts a connector, changes a route or edits a firewall rule should require human approval before it runs, and should log who approved it. Diagnosis and triage can be automated more freely than remediation. Define which actions an agent may take on its own, which need sign-off, and keep that list in your runbook.
What Netverge offers for this workflow
If you run tunnels across many sites, you need monitoring that sees the site, not only the Cloudflare dashboard. Netverge's AI-powered monitoring brings your infrastructure data, tickets and documentation into one interface. Edge visibility comes from Vergepoints, which are available as hardware at $49 per month per device or as software at $29 per month per Vergepoint, as published on the pricing guide. The Starter Package is $299 per month, listed on the pricing page.

You can estimate your own monthly cost with the pricing calculator, or begin with a free trial.
FAQ
How do I check Cloudflare Tunnel status?
Open the Cloudflare dashboard under Networking > Tunnels, or run cloudflared tunnel list. Each tunnel shows Healthy, Inactive, Down or Degraded. Remember that this reflects the connector's link to Cloudflare, not the health of the application behind it.
Why does my tunnel show Healthy when the app is unreachable?
Tunnel status only covers the connection between cloudflared and Cloudflare. A wrong Service URL, an origin certificate problem, a refused connection or a DNS mismatch can all break the app while the tunnel stays Healthy. Check the cloudflared logs and add an HTTPS check against the app hostname.
Can I use Prometheus and Grafana for Cloudflare Tunnel monitoring?
Yes. cloudflared exposes a Prometheus metrics endpoint, and Cloudflare's tutorial describes a Prometheus server scraping it with Grafana reading from Prometheus. Metrics cannot be pushed directly from cloudflared to Grafana, so each site needs a collector that can reach the endpoint. Set --metrics to a fixed address so the port does not change.
What does error 1033 mean?
It means Cloudflare cannot find a healthy cloudflared instance to send the traffic to. Confirm the tunnel is Active in the dashboard or with cloudflared tunnel list. If it is, check that the DNS record points to the correct tunnel subdomain.
Is it safe to leave debug logging on?
No. Debug level records request URLs and all request and response headers, which can contain sensitive data. Use it for a short troubleshooting window, protect the log file, and return to the default info level afterward.
Sources
- Observability, Cloudflare Docs
- Common errors, Cloudflare One docs
- Wrangler tunnel commands, Cloudflare docs repository
- Monitor Cloudflare Tunnel with Grafana, Cloudflare Docs
- Tunnel run parameters, Cloudflare One docs
- cloudflared release notes
- cloudflared change log, version 2024.12.1 port binding
- Public load balancers, Cloudflare One docs
- Troubleshoot HTTPS origins with Cloudflare Tunnel
- Troubleshooting, Cloudflare Docs
- Cloudflare tunnel connection error, GitHub issue
- cloudflared probe loop fix, GitHub commit