An internal database can be reachable and speak TLS while a certificate mismatch still prevents the application from connecting. Cloudflare’s November 18 Workers VPC update adds a better way to handle private issuers: attach your own certificate authority and keep full verification. The useful consequence is specific. Trusting an internal CA need not mean giving up the check that you reached the intended server.
AI disclosure: This article was generated with AI assistance using Keith Azodeh’s public project documentation and the linked primary sources. It has not been reviewed by Keith before publication.
The release also adds client certificates for services requiring mutual TLS, or mTLS. Custom CAs and client certificates can be used independently or together, for HTTPS services and TLS connections to TCP services, including PostgreSQL through Hyperdrive. Cloudflare lists Wrangler 4.164.0 or later and Terraform provider 5.29.0 or later as configuration options, alongside its API. Workers VPC remains in beta.
That creates a practical opening for teams connecting cloud applications to existing internal services. It also makes it worth revisiting any connection whose verification settings were relaxed during an earlier integration attempt.
Decide which endpoint you intend to trust
Keith’s Illumia Archive Intelligence case study describes a local PostgreSQL archive with a browser interface and six bounded, read-only retrieval tools for Claude Desktop. The browser and assistant share the retrieval service. Application, packaging, and MCP startup validation used synthetic records; workplace acceptance remains staged.
The case study does not describe a Workers VPC deployment or a migration of that archive. It provides a useful design lens for a proposed extension: if an organization wanted a remotely hosted interface to reach a private retrieval service, what would prove that the connection reached the correct endpoint?
Start by writing down that endpoint’s intended identity. A route to an internal address answers where traffic goes. A certificate check answers a different question about the server at the other end. Those decisions should agree.
For an archive, reaching the wrong service can be more subtle than a connection error. A staging copy may respond successfully and contain plausible older records. The page loads, but the application is consulting a different body of evidence.
That is a hypothetical failure scenario, not an incident reported in the Illumia case study.
Know what each verification mode preserves
Cloudflare’s VPC Service configuration guide defines the modes clearly:
-
verify_fullchecks the certificate chain and hostname. -
verify_cachecks the chain while skipping hostname verification. -
disableddoes not verify the server certificate.
For a service with a certificate issued by an internal CA, configure the relevant trust chain and the intended name together. Simply choosing a setting that makes the error disappear can remove evidence about which server you contacted.
The new custom-CA reference supplies trust material for checking the origin. A client certificate serves the other direction: it identifies the caller to an origin that requests one. Cloudflare requires an explicit verification mode when either certificate reference is configured.
Keep the application’s permissions equally deliberate. In the proposed archive extension, an authenticated connection should still expose only the intended retrieval operations. A transport change is no reason to add arbitrary SQL or write access to a previously bounded interface.
Ask the wrong server to prove the setting works
A useful proposed acceptance exercise uses two disposable endpoints and synthetic responses. Give both certificates from the same test CA, but issue them for different names. Make their responses visibly distinguishable, such as CURRENT TEST ARCHIVE and ALTERNATE TEST ARCHIVE.
First establish that the intended endpoint succeeds with the selected trust chain and matching identity. Then exercise a mismatch while keeping the trusted issuer constant. The purpose is to distinguish “this issuer is trusted” from “this is the requested server.”
Also check the client-certificate requirement independently if the origin uses mTLS. An accepted caller and an accepted origin are two separate observations. Record which side rejected a failed handshake before changing configuration.
These are suggested checks, not results from a deployment performed for this article. Use a controlled environment; no real archive records are needed to demonstrate the connection behavior.
Preserve the trust settings during maintenance
Certificate setup is only the first operation. Cloudflare’s rotation instructions say to upload a replacement certificate and update the service reference. A service update replaces the complete configuration, including TLS settings. Omitted certificate references are cleared.
For an operator rotating only the client certificate, this means reviewing the full desired service configuration before submitting the update. The CA reference and verification mode belong in that review even when they are not the reason for the change.
Keep a record of the intended hostname, issuer, client identity where applicable, and the application operation used to check the result. That gives the next maintainer a concrete explanation of what a working connection is supposed to establish.
For internal knowledge tools and other private services, today’s update can remove an integration constraint. The stronger outcome is a connection that works while preserving the identity checks the application depends on.
Top comments (0)