Struct ClusterState 

pub struct ClusterState {
    pub gpus: Vec<GpuState>,
    pub model_registry: HashMap<ModelId, ModelSpec>,
}

The complete state of the GPU cluster at a point in time. This is the input to all routing decisions.

Fields

gpus: Vec<GpuState>

model_registry: HashMap<ModelId, ModelSpec>

Implementations

impl ClusterState

pub fn new( gpus: Vec<GpuState>, model_registry: HashMap<ModelId, ModelSpec>, ) -> Self

pub fn find_loaded_model(&self, model_id: &ModelId) -> Option<(GpuId, u16)>

Find which GPU (if any) currently has the given model loaded.

pub fn get_model_spec(&self, model_id: &ModelId) -> Option<&ModelSpec>

Get the spec for a model from the registry.

pub fn healthy_gpus_by_available_memory(&self) -> Vec<&GpuState>

Get all healthy GPUs, sorted by available memory (most available first).

pub fn gpus_with_space_for( &self, vram_required: ByteSize, headroom: ByteSize, ) -> Vec<&GpuState>

Get all GPUs that could fit a model of the given size (with headroom).

Trait Implementations

impl Clone for ClusterState

fn clone(&self) -> ClusterState

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for ClusterState

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

Formats the value using the given formatter.