Tool Rate Limiting — References

The protocol conventions for communicating a limit, and the operational literature on admission control and fairness.

Communicating a limit

• RFC 6585: Additional HTTP Status Codes — defines the too-many-requests status, the response a client is expected to produce when an allowance is exhausted.
• RFC 9110: HTTP Semantics — the Retry-After field and the service-unavailable semantics that distinguish a deliberate refusal from a failure, which is the distinction a caller's classifier depends on.
• RateLimit header fields for HTTP — an IETF HTTPAPI working group document specifying fields for a server to advertise remaining allowance and reset time, so a client can pace itself rather than discover the boundary by being refused. Cited by name; its publication status is not confirmed here.
• RFC 9457: Problem Details for HTTP APIs — a structured body for a refusal, letting a caller distinguish a quota exhaustion from a concurrency rejection without parsing prose.

Admission control and overload

• Google SRE Book — the handling-overload chapter on client-side throttling, criticality-based admission, and why shedding early is preferable to queueing; and the cascading-failures chapter on how limits interact with retry behaviour.
• Release It! Design and Deploy Production-Ready Software — Michael Nygard, Pragmatic Bookshelf. The bulkhead pattern, which is the concurrency-partitioning form of this capability. Cited by name.
• Congestion Avoidance and Control — Van Jacobson and Michael Karels, SIGCOMM 1988. The origin of additive-increase multiplicative-decrease control, which adaptive limiters against undocumented upstream limits are a direct application of. Cited by name.

No verified reference is recorded for per-agent-run call budgeting, or for fairness between tenants sharing a single third-party credential. Both are current practice problems without a published treatment.