Workload identity beyond one operator.

Workload identity separates the identity of an application from its network location. Putting that pattern into operation also requires decisions about who issues credentials, which authorities are trusted and what happens when infrastructure fails.

Browse documentation
On this page

Applications as principals

An address helps locate a service. A cryptographic identity allows a peer to authenticate a named principal before deciding whether to accept its request. This distinction matters when applications move between machines, run several replicas or communicate through intermediaries.

Google’s December 2017 ALTS whitepaper describes an implementation of this pattern. ALTS authenticates identities associated with services and other entities, with an identity model that accommodates replication and rescheduling. It supplies an authenticated peer identity that applications can use in authorization decisions.

This is a documented design for Google’s environment at that time. It provides a useful precedent without establishing that every large infrastructure provider uses the same architecture or makes the same assumptions.

Operating the credential lifecycle

The ALTS account also describes certificate issuance, rotation, revocation and session resumption. Those mechanisms make credential management an infrastructure responsibility. The handshake is one part of a system that must distribute keys, propagate trust changes and recover from failures.

SPIFFE’s concepts give another concrete expression of workload identity: an identifier within a trust domain, a verifiable identity document, and an API through which workloads obtain credentials and trust bundles. Its X.509 credentials are short-lived and rotated automatically.

The operational consequence is a dependency to design and test. Issuance must identify the right workload, private keys need protection, and renewal must happen before expiry. Restarting or stopping a process does not itself invalidate a copied certificate and private key. Credential validity and process lifetime are different conditions.

Federation across independent authorities

SPIFFE defines a trust domain around an independent identity infrastructure and its trust roots. This gives separate operators a way to retain distinct authorities while describing the identities they issue.

Consider a sensor service exchanging information with a partner’s command application. Their operators may control different hosts, issuance policies and maintenance schedules. Connecting them requires agreement about which foreign identities can be authenticated and which requests those identities may make.

Federation makes those trust relationships explicit. The deployment must establish how trust material is obtained, updated and withdrawn. Accepting an authority’s identity documents is separate from authorizing access to data. Each operator still needs policy for its own services and a way to respond when a partner’s trust material changes.

Designing for failure conditions

An outage can prevent renewal or delay a trust update. Cached credentials and bundles may remain usable within their validity and local policy, but they cannot reveal a remote change that has not arrived. Deployment plans therefore need explicit expiry behavior, recovery procedures and acceptance criteria for disconnected operation.

SPIFFE assumes isolation sufficient to protect one workload’s credentials from another; it does not define that isolation mechanism. A compromised host can undermine this assumption. An authenticated identity consequently does not establish that the host remains uncompromised or that an application’s response is correct.

The useful architectural inheritance is a clear separation of responsibilities: authenticate application principals, authorize their actions, operate the credential lifecycle and state the conditions under which those controls remain dependable.