TiKV PD endpoints need network isolation beyond dashboard login

TiKV is a distributed transactional key-value store. The Placement Driver, or PD, coordinates the cluster and exposes client endpoints used by TiKV and TiDB components. TiDB Dashboard may require authentication, but that does not necessarily make every PD client endpoint require the same identity. A network-visible PD service can therefore be a control-plane question even when the dashboard appears protected.

On November 20, 2026, ZoomEye returned 14 matches for title="TiKV" and zero for app="TiKV" with sub_type=all, page 1, and pagesize 1. A 14-match title set is a very small sample and requires manual validation. It does not prove that only 14 TiKV deployments are reachable, and the zero application result does not clear the product from the internet. It simply means this count-only observation is not a complete inventory.

Read the documented endpoint model

TiDB's security documentation warns that privileged PD interfaces, such as the member API on the PD client port, may not require identity authentication even when the dashboard does. The same documentation notes that TiKV and other components need access to that port, so it should not be blocked between cluster components. The correct boundary is the management network, not a blanket host firewall that breaks the cluster.

Review which addresses expose the PD client port, which expose the dashboard, and which expose metrics or health endpoints. Map the network zones that can reach each listener. A dashboard login should be treated as one control for one surface. It does not prove that the PD member API is protected.

Review TLS as a separate control

TiKV security documentation describes mutual TLS between TiKV nodes and PD, including per-service certificates and keys. TLS protects the transport path, but it does not replace network authorization. A valid cluster certificate proves a participant in the cluster trust domain. It does not decide whether an unrelated host should be able to reach a privileged administrative endpoint.

Verify certificate issuance, renewal, expiry, and private-key custody. Test that a client without a trusted certificate is rejected. Then test that the permitted cluster client cannot reach an administrative endpoint that should be restricted to a different management network.

Inventory cluster metadata carefully

Record the cluster name, owner, environment, service addresses, TLS mode, and the list of components that are allowed to contact PD. Review backup and restore paths as well, because PD metadata defines the cluster topology and placement state. A monitoring dashboard is useful for detection, but it is not proof that the underlying control interfaces have the right network boundary.

For the 14 title candidates, classify by host, certificate, response, and any visible TiKV or dashboard evidence. Do not probe PD state-changing endpoints without explicit authorization. The external observation is enough to decide whether the owner needs to review an exposed management plane.

Small counts deserve precise language. Fourteen title matches may identify a handful of legitimate public references or a few real deployments. The finding is the exposure of a high-authority service, not the count itself.

References