Ports and Network Flows
Plan Management and appliance access separately. Management’s public hostname, appliance application hostname, cluster API endpoint, and individual node addresses identify different destinations.
Operator and Integration Access
| Source | Destination | Protocol / port | Purpose and applicability |
|---|---|---|---|
Operator browser |
Management public hostname |
TCP / configured HTTPS port, default 443 |
Management, authentication, downloads, and operator tools exposed through Management. |
Browser or ACME certificate authority |
Management host |
TCP / 80 |
Published HTTP listener. ACME HTTP-01 validation requires the CA to reach it. |
Host administrator |
Management host |
TCP / 22 |
SSH host administration. Separate from Management browser authentication. |
Management |
Appliance nodes |
TCP / 50000 |
Authenticated Talos API used by installation, Operations and firewall changes. It remains reachable between cluster access leases. |
Management |
Appliance node IPs |
TCP / 6443 |
Authenticated Kubernetes access for operations, health and tools. External access is opened for the cluster access lease. |
Management Operations |
Appliance node registries |
TCP / 443 |
Authenticated artifact publication and reads. External access remains open for the cluster access lease, rather than opening separately for each operation. |
Management |
Configured release registries |
TCP / endpoint port |
Retrieve release or update bundles. HTTPS commonly uses 443; use the configured registry endpoint. For offline use, upload the bundle through the browser. |
Customer monitoring |
Appliance customer metrics listener |
TCP / 9443 |
Optional HTTPS metrics interface; closed when metrics access is disabled. |
Appliance |
Configured Syslog destination |
Configured TCP or UDP port |
Optional log forwarding. Initial settings use TCP 6514 with TLS; use the receiver’s agreed settings. |
Appliance |
Configured S3 backup endpoint |
TCP / endpoint port |
Optional publication of full backups. Use the endpoint’s configured scheme and port. |
Appliance |
Configured SFTP backup server |
TCP / configured port, default 22 |
Optional publication of full backups using SSH key authentication. |
The node-registry listener on TCP 443 is an artifact endpoint. Do not confuse it with the Management HTTPS listener or an ERS application endpoint.
Access Lifetime and Network Restrictions
A valid artifact bundle opens external Kubernetes and registry access for a six-hour lease. Expiry, explicit eviction and normal Management shutdown attempt to close those external paths while preserving cluster-internal traffic. After a crash or loss of node connectivity, Management may be unable to confirm that remote access is closed. Reconnect to reconcile access and follow the displayed warning.
The lease does not create a Management source-IP allowlist. While open, these authenticated endpoints can be reached from routable IPv4 sources permitted by your surrounding network. Restrict those sources through your own network controls. See Renew or end cluster access and Security and hardening.
Network Services and ERS Dependencies
Management and appliance nodes need the DNS and time services selected during installation. Management startup waits for its configured network and time synchronization. Allow the actual DNS resolver and NTP destination specified in the deployment plan.
ERS client interfaces, HSM connections, identity providers, and external certificate services depend on the configured product interfaces and vendor endpoints. Use the matching ERS product manual and HSM vendor requirements for those flows; record the selected ports in the deployment plan. See external monitoring and backup destinations for optional integrations.
Scope of this Table
This table covers Management access and customer integrations. It is not a complete node-to-node firewall allowlist. The appliance also needs internal cluster, storage, database, and workload communication. Use the deployment network requirements agreed for the release before adding filtering between appliance nodes. Do not expose cluster-internal service ports to client networks merely because a workload uses them.