Read what a persisted deployment is actually serving and say whether it is
what the record desires.
Single-shot on purpose: convergence backoff exists to wait out a promote this
process just performed, and a host reading state performed none. Waiting here
would only spend the provider budget re-asking a question whose answer has
already settled.
Reports rather than enforces. A disagreement comes back as
matchesDesired: false with the attestation that proves it, because a host
rendering fleet state wants to show drift, not fail on it. Only a provider
that cannot be attested at all throws.
NOT for per-request status rendering. Each single-shot call spends two
provider reads on the plain-Worker backend or three on Workers for Platforms,
against a default 1,100-request-per-five-minute account-wide budget shared by
every provisioning operation. A status surface with any traffic at all must
cache these results rather than attest per request.
Read what a persisted deployment is actually serving and say whether it is what the record desires.
Single-shot on purpose: convergence backoff exists to wait out a promote this process just performed, and a host reading state performed none. Waiting here would only spend the provider budget re-asking a question whose answer has already settled.
Reports rather than enforces. A disagreement comes back as
matchesDesired: falsewith the attestation that proves it, because a host rendering fleet state wants to show drift, not fail on it. Only a provider that cannot be attested at all throws.NOT for per-request status rendering. Each single-shot call spends two provider reads on the plain-Worker backend or three on Workers for Platforms, against a default 1,100-request-per-five-minute account-wide budget shared by every provisioning operation. A status surface with any traffic at all must cache these results rather than attest per request.