Preview draft. Copy and visuals are pending review and may change before launch.

Practice

Authorization in serverless and ephemeral compute

A traditional server authenticates once and stays running for hours. A serverless function may exist for a fraction of a second — long enough to matter, too short to assume anything persists.

← Back to the library

Delegation, Automation, and Machine-to-Machine · Published 2026-09-05

No persistent process means no persistent assumption

A long-running server process can authenticate once at startup and hold that credential in memory for as long as the process runs, refreshing it occasionally in the background. A serverless function doesn't have that luxury — each invocation may run in a fresh execution context, with no guarantee that anything from a previous invocation, including a cached credential, is still around by the time the next one starts.

This pushes toward provisioning credentials that match the function's own short lifetime: issued for the invocation, scoped to exactly what that invocation needs, and not expected to outlive it. A credential provisioned once and baked into configuration to avoid re-issuing it on every invocation defeats the isolation ephemeral compute is otherwise well-suited to provide.

Fast issuance becomes a hard requirement, not a nice-to-have

Because credential issuance now happens on a path that runs on every single invocation rather than once at process startup, the speed and reliability of that issuance directly affects the latency and reliability of the function itself. A slow or occasionally-failing credential-issuance step, tolerable when it happens once per long-running process, becomes a much more visible problem when it happens on every single short-lived invocation.