async-bulkhead-llm - v3.17.0
    Preparing search index...

    Type Alias LLMAdmissionLimits

    Complete, versioned snapshot of every admission limit that can change at runtime. Updates are intentionally not partial: callers must provide the whole snapshot so the bulkhead can validate and apply it atomically.

    tokenBudget must be present when the bulkhead was constructed with a token budget and absent otherwise. Runtime enable/disable is deliberately unsupported because estimator configuration is construction-time state.

    type LLMAdmissionLimits = {
        admissionClasses?: Readonly<Record<string, LLMAdmissionClassLimits>>;
        maxConcurrent: number;
        maxQueue: number;
        revision: number;
        tokenBudget?: { budget: number; highPriorityReserve: number };
    }
    Index
    admissionClasses?: Readonly<Record<string, LLMAdmissionClassLimits>>

    Complete limits for the fixed construction-time admission-class table. Required on updates when classes are configured; forbidden otherwise. Keys cannot be added or removed at runtime.

    maxConcurrent: number

    0 is a fail-fast kill switch for new admissions.

    maxQueue: number

    Existing waiters are preserved when this shrinks.

    revision: number

    Monotonically increasing non-negative safe integer.

    tokenBudget?: { budget: number; highPriorityReserve: number }