Struct NodeResources 

pub struct NodeResources {
    pub id: String,
    pub vram_gb: f32,
    pub system_memory_gb: f32,
    pub compute_capability: String,
    pub gpu_name: Option<String>,
    pub rpc_port: Option<u16>,
    pub rpc_build_id: Option<String>,
}

Node resources structure for discovery frames

Fields

id: String

Unique node identifier

vram_gb: f32

GPU VRAM in GB (f32 for precision)

system_memory_gb: f32

System memory in GB

compute_capability: String

CUDA compute capability string (e.g., “8.9”)

gpu_name: Option<String>

GPU name/model

rpc_port: Option<u16>

Port this node’s ggml-rpc-server is listening on, when it has opted in to contributing compute to the cluster (see ghost_link::rpc_cluster). None means this node either isn’t running one or the peer that sent this frame predates this field — both are indistinguishable and both correctly mean “don’t route distributed inference through this node.”

rpc_build_id: Option<String>

Short commit-hash fingerprint of this node’s local llama.cpp build (see ghost_link::native_engine::NativeEngineClient::get_llama_build_id), e.g. "da296d6". None means this node either couldn’t determine its own build id or the peer that sent this frame predates this field — both are indistinguishable, and both are treated as “unknown, can’t verify” rather than “mismatched” by rpc_cluster::discover_rpc_peers (mismatched-but-unknown must not be treated the same as a confirmed mismatch, or every not-yet-updated peer becomes unroutable overnight). Used to refuse routing distributed inference through a peer whose llama.cpp build doesn’t match the local one — version-mismatched ggml-rpc peers have been confirmed on real hardware to silently corrupt output for larger models while reporting healthy status the whole time (see docs/BENCHMARKS.md’s native two-machine entry).

Implementations

impl NodeResources

pub fn new( id: impl Into<String>, vram_gb: f32, system_memory_gb: f32, compute_capability: impl Into<String>, gpu_name: Option<String>, ) -> Self

Create new node resources

pub fn with_rpc_port(self, port: u16) -> Self

Builder-style setter so adding this field didn’t require touching every existing NodeResources::new(...) call site across the workspace’s tests and CLI commands.

pub fn with_rpc_build_id(self, build_id: impl Into<String>) -> Self

Builder-style setter, same rationale as with_rpc_port.

pub fn encode_payload_into( &self, buffer: &mut Vec<u8>, max_size: usize, ) -> Result<usize, &'static str>

Serialize node resources into an existing binary buffer.

Returns the length of the written payload on success, or an Error on validation failure.

pub fn encode_payload(&self, max_size: usize) -> Vec<u8>

Serialize node resources to fixed-width binary payload

Format: [id_len(1) + id + vram_f32_le + mem_f32_le + cc_len(1) + cc]

pub fn decode_payload(payload: &[u8]) -> Result<Self, String>

Deserialize node resources from binary payload

Trait Implementations

impl Clone for NodeResources

fn clone(&self) -> Self

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for NodeResources

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

Formats the value using the given formatter.

impl Default for NodeResources

fn default() -> Self

Returns the “default value” for a type.