Nobody puts "identity broker topology" on the first slide of a VCF 9 kickoff. It comes up around week three, in a side conversation, when someone from the security team asks which directory the platform signs in against and the room goes quiet.
Certificates arrive the same way, a little later and with an expiry date attached. Neither topic gets much attention at bring-up, and both land on somebody's desk months afterward.
One Broker, One Failure Domain
VCF 9 centralizes sign-in behind the VCF Identity Broker. It comes in two forms: embedded on the management domain's vCenter, which is the simple option, or an external appliance cluster, which is the one built for larger, multi-environment scale. Either way it sits in front of your external identity provider (supported options include Okta, Entra ID, AD FS, PingFederate, and generic SAML 2.0) and can draw users from AD or OpenLDAP.
The part that gets missed: in a standard fleet there is one broker, deployed in the management domain of the first VCF instance, and every later instance and workload domain gets its sign-in from it. If that broker is down, the whole fleet is affected, not one site. The exception is the design for multiple sites in a single region plus additional regions, which deploys a broker per VCF instance so an outage stays contained to that instance.
That's a failure-domain decision, and it deserves the scrutiny any other one gets. A single-site shop may be perfectly happy with one broker. A fleet spread across regions should ask whether "nobody can log in anywhere" is a scenario they've actually priced. Whether a broker can be moved between the embedded and external forms later is something I'd confirm in the documentation for the target version instead of assuming.
Habits from 5.x get in the way here, too. Enhanced Linked Mode is gone in VCF 9, replaced by a new vCenter linking model, so a sign-in design built on ELM assumptions starts from the wrong diagram.
The Account That Works When Nothing Else Does
In VCF Automation, identity is set up per tier, and the two named patterns are Enterprise (provider and tenant share one identity provider) and Service Provider (each uses its own). If a tier's provider is never configured, only that tier's original bootstrap admin can log in. That makes the bootstrap account your break-glass account by default, so someone should know where its credentials live before the IdP integration goes in, not after it breaks.
VMCA Is Not a General-Purpose CA
VMCA exists to automate certificates for vSphere components, and it's deliberately narrow. There's no CRL, so trust is handled through thumbprint prompts and not revocation. No wildcard certificates, since the trust model is subject-name based. No ACME. No passphrase-protected keys, because those would block automated cluster startup. And it won't proxy to an external PKI.
It runs in one of two modes: the default self-signed root, or subordinate CA mode, where your organization's PKI issues VMCA's signing certificate. Subordinate mode is the recommended pattern for regulated and large environments, and the reason is easy to see. With the default root, every browser and every script in the company is asked to trust something your PKI team has never heard of. With a subordinate, the chain ends somewhere they already trust.
Pick before the first certificate is issued. Changing the root later means touching everything that trusted the old one.
On the VCF side, certificate automation covers auto-renewal, VMCA or an external CA (Microsoft or OpenSSL), bulk operations, and public APIs for scripting the lifecycle. That's the part that works. The next list is the part that doesn't.
Automation covers what it covers. The expiry date doesn't care about the difference.
Certificates Nobody Will Remind You About
- NSX Global Manager. Lifecycle and certificates are manual, outside SDDC Manager's automation, and its certificate thumbprints have to be re-captured by hand after any Local Manager certificate rotation done through SDDC Manager. Miss it and Global-to-Local authentication fails quietly.
- Supervisor management interface. The default self-signed certificate should be replaced with a PEM-encoded, CA-signed certificate (SHA-2 or better), and that replacement happens outside SDDC Manager's certificate automation.
- VCF Operations cluster certificate. The SANs have to list every cluster node FQDN plus the external load balancer FQDN if there is one (collector nodes are excluded), and the certificate is reissued whenever a node is added.
- Cross-instance log forwarding. SSL forwarding between VCF Operations for Logs clusters fails silently with the default self-signed certificate until a CA-signed one is in the Java keystore of every receiving node.
Check each item against the page for your target build before it goes into a design.
Questions I Ask Before Anyone Requests a Certificate
Which CA is the root of trust for this environment, and does VMCA sit under it or beside it?
Where does the identity broker run, and have we decided that a fleet-wide sign-in outage is acceptable?
Which certificates are outside SDDC Manager's automation, and who owns each renewal date?
Who holds the bootstrap admin credentials, and when were they last tested?
Has the owner of the external identity provider agreed to a change window?
If you're about to sit down with the security team and want to walk in knowing which of these are actually yours to decide, send me a DM. It's usually a short conversation.