Running an OPC UA GDS at fleet scale: intermediate CAs, CRLs and the certificate database
A Global Discovery Server managing one cell with fifty PLCs is a solved problem: run the container, enrol the devices, let rotation do its job. We've written about what a GDS is and how Push reaches devices you can't dial. This piece is about what changes when the GDS stops being a lab fixture and becomes the PKI for an industrial fleet — multiple plants, multiple vendors, one security team, and auditors who ask precise questions.
Four things change, and none of them is "more of the same."
1. Your GDS stops being the root of trust
In a single-cell deployment, the GDS's built-in CA is the PKI: self-signed root, done. At fleet scale that stops being acceptable — corporate security owns a root CA (often behind an HSM, often run by IT), and nothing on the OT side gets to mint its own root.
The standards-clean answer is an intermediate CA: the GDS operates a subordinate CA whose certificate is signed by the corporate root. The chain of trust runs device cert → GDS intermediate CA → corporate root, which gives you the best of both worlds — OT keeps operational autonomy (issue, renew, revoke without opening a ticket), while IT keeps cryptographic sovereignty (the root never leaves its vault, and revoking the intermediate kills the whole branch if needed).
Mechanically, this is a three-step external-signing workflow: the GDS generates a CSR for its CA key, the corporate root signs it, the signed certificate (with the issuer chain) is installed back. The private key never moves. The same workflow covers proactive renewal of the intermediate itself — renewing the CA certificate while keeping its key, so every certificate it previously issued remains valid. If your tooling makes those three steps idempotent and restart-safe, enrolling a new plant is an afternoon; if it doesn't, it's a change-control saga.
2. Chains — not certificates — are what you distribute
The moment an intermediate CA exists, every leaf certificate is only verifiable with its chain. OPC UA is explicit about this: Part 6 §6.2.6 defines how a certificate chain is conveyed — leaf first, then issuers in order.
This is, in our experience, the single most common fleet-scale failure: a server that stores and presents only its leaf certificate. Everything works in the lab (where the client happens to trust the intermediate directly), then fails at the worst moment — a SCADA reconnecting after the intermediate rotated, a new client from another vendor doing strict chain building. The symptoms look random: BadCertificateChainIncomplete, BadCertificateUntrusted on one client but not another.
The rule is boring and absolute: persist the full chain, serve the full chain — in CreateSession responses, in GDS UpdateCertificate pushes, in trust lists. And the trust-list layout matters just as much: the root belongs in the trusted store; intermediates belong in the issuer store, present for chain building but not trust anchors themselves. Promote intermediates to trust anchors and you've silently widened who can sign their way into your plant.
3. Revocation becomes a distribution problem
With five devices, revocation is theoretical. With five hundred, a decommissioned machine, a compromised vendor laptop, or an employee departure makes it operational — and a revocation that doesn't reach every verifier is a revocation that didn't happen.
That means CRLs are now part of your distribution plane, with the same discipline as certificates themselves: every CA in the hierarchy publishes one (yes, the root too — auditors check), the GDS pushes updated CRLs with trust-list updates, and CRL freshness is monitored like any other fleet-health metric. Stale CRLs produce the nastiest failure mode in PKI: not an error, but a silent acceptance of something you thought you had killed. Short CRL lifetimes force the discipline; they're only operationally bearable when distribution is automatic — which is precisely the GDS's job.
4. The audit stops being archaeology
Here's the question that separates a lab PKI from a fleet PKI. An auditor asks: "Show me every certificate issued for plant B in Q3, which are still valid, and who revoked serial 4F2A… and when."
If answering that means grepping directories of PEM files across sites, you don't have a PKI — you have a pile of files. Fleet scale demands a certificate database: every issuance, renewal and revocation recorded with serial, subject, validity, status and the operator who triggered it; queryable by any of those; exportable as the report the auditor actually wants. IEC 62443 and NIS2 assessments both land on this point — centralised certificate governance with evidence. The data model is not exotic; what matters is that it's maintained by the issuing path itself, not reconstructed afterwards.
The fleet-scale checklist
| Concern | Single cell | Fleet |
|---|---|---|
| Root of trust | GDS built-in CA | Corporate root + GDS intermediate CA |
| What's distributed | Leaf certificates | Full chains (Part 6 §6.2.6) + trust lists |
| Trust-list layout | Everything trusted | Root = anchor; intermediates = issuer store only |
| Revocation | Theoretical | CRLs published per CA, pushed, freshness-monitored |
| Records | Files on disk | Queryable certificate database + audit journal |
| Renewal of the CA itself | "We'll deal with it" | Proactive, key-preserving, documented |
Where Sterfive stands
All of the above is implemented and shipping. Sterfive GDS runs PULL and PUSH on one instance with a pluggable CA (built-in, Step-CA, Vault, ADCS, EJBCA); the underlying node-opcua-pki engine added production-grade subordinate-CA lifecycle management this year — the idempotent 3-step external-signing workflow, automatic full-chain output per Part 6 §6.2.6, key-preserving proactive renewal — plus the certificate database (query issued certificates by serial, status, subject) and buffer-based issue/revoke APIs. It deploys from one docker-compose.yml, admin console included, and runs air-gapped.
We maintain the open-source OPC UA stack underneath it all, which is exactly why the fleet-scale corner cases above read like a war-story list: we've been called into most of them.
Key takeaways
- At fleet scale, the GDS runs an intermediate CA under your corporate root — OT autonomy, IT sovereignty.
- Distribute chains, not leaves; keep intermediates out of the trust-anchor store.
- Revocation is only real when CRL distribution and freshness are automated and monitored.
- A certificate database with an audit journal is what turns compliance from archaeology into a query.
References
- OPC Foundation, OPC UA Part 6: Mappings, §6.2.6 Certificate chains. opcfoundation.org
- OPC Foundation, OPC UA Part 12: Discovery and Global Services — Global Discovery Server, Pull & Push management. opcfoundation.org
Sterfive builds OPC UA GDS and maintains the open-source node-opcua stack it runs on. If you're taking a GDS from one cell to a fleet — intermediate CA, CRLs, audits and all — talk to our engineers.