Struct VpaEngine 

pub struct VpaEngine { /* private fields */ }

Per-container vertical-pod-autoscaler engine.

Keeps a small ring buffer of recent usage samples per container, derives a CPU rate (millicores) from successive cumulative-CPU readings, and emits a percentile-based recommendation clamped to the service’s VerticalScaleSpec bounds. This is the agent-local engine used by AutoscaleController; it deliberately lives here (rather than in zlayer-scheduler) so the whole vertical-apply path is self-contained in the controller.

Implementations

impl VpaEngine

pub fn new() -> Self

Construct an empty engine.

pub fn observe(&mut self, container: &str, stats: &RawContainerStats) -> f64

Record one raw stats sample for container and return the busiest-replica CPU rate (millicores) this sample implies, so callers can drive scale-to-zero idle accounting off the same observation.

CPU rate is derived from the delta between this sample’s cumulative cpu_usage_usec and the previous one over the elapsed wallclock time: millis = (Δusec / Δsec) / 1000. The first sample for a container has no predecessor, so it contributes a memory reading but a 0.0 CPU rate.

pub fn recommend( &self, container: &str, spec: &VerticalScaleSpec, ) -> Option<VpaRecommendation>

Compute a recommendation for container from its observed history, clamped to spec’s bounds. Returns None until enough samples exist to produce a stable percentile (at least one CPU rate observation).

pub fn forget(&mut self, container: &str)

Drop all history for a container (e.g. after it is removed by a scale-down or rolling restart) so a recreated replica starts fresh.

Trait Implementations

impl Debug for VpaEngine

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

Formats the value using the given formatter.

impl Default for VpaEngine

fn default() -> VpaEngine

Returns the “default value” for a type.