Struct QueryBudget 

pub struct QueryBudget {
    pub query_class: String,
    pub ram_bytes_limit: u64,
    pub ssd_random_reads_limit: u32,
    pub ssd_sequential_bytes_limit: u64,
    pub cpu_cycles_limit: u64,
    pub latency_target: Duration,
    pub recall_target: f32,
    pub recall_confidence: f32,
}

Per-query budget limits derived from SLA targets

Fields

query_class: String

Query class identifier (e.g., “low_latency”, “high_recall”, “balanced”)

ram_bytes_limit: u64

Maximum RAM bytes scanned for candidate generation This bounds memory bandwidth consumption

ssd_random_reads_limit: u32

Maximum SSD random reads in hot path (ideally 0) Random IO is the main source of p99 variance

ssd_sequential_bytes_limit: u64

Maximum SSD sequential bytes for rerank batching Sequential IO is more predictable than random

cpu_cycles_limit: u64

Maximum CPU cycles for routing and scan Converted from latency target: cycles = latency_ns * freq_ghz

latency_target: Duration

Target latency SLA (e.g., p99 ≤ 10ms)

recall_target: f32

Target recall floor (e.g., recall@10 ≥ 0.95)

recall_confidence: f32

Probability of meeting recall target (1 - δ)

Implementations

impl QueryBudget

pub fn new(query_class: &str) -> Self

Create a new budget for a query class

pub fn ram_bytes(self, limit: u64) -> Self

Set RAM bytes limit

pub fn ssd_random_reads(self, limit: u32) -> Self

Set SSD random reads limit

pub fn ssd_sequential_bytes(self, limit: u64) -> Self

Set SSD sequential bytes limit

pub fn cpu_cycles(self, limit: u64) -> Self

Set CPU cycles limit

pub fn latency_target(self, target: Duration) -> Self

Set latency target

pub fn recall_target(self, target: f32, confidence: f32) -> Self

Set recall target

pub fn from_sla( query_class: &str, latency_target: Duration, recall_target: f32, hardware: &HardwareProfile, ) -> Self

Create budget from SLA parameters

Converts latency SLA into resource budgets:

• Bytes ≤ BW_eff · T
• RandomIO ≤ ⌊T / L_io⌋

pub fn low_latency() -> Self

Predefined budget for low-latency queries (p99 ≤ 5ms)

pub fn balanced() -> Self

Predefined budget for balanced queries (p99 ≤ 20ms)

pub fn high_recall() -> Self

Predefined budget for high-recall queries (p99 ≤ 100ms)

Trait Implementations

impl Clone for QueryBudget

fn clone(&self) -> QueryBudget

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for QueryBudget

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.