Struct NodeCaps 

pub struct NodeCaps {
    pub rank: u32,
    pub role: NodeRole,
    pub devices: Vec<String>,
    pub platform: String,
}

What a node tells the coordinator about itself, so placement is driven by what the hardware can actually do rather than by assumption.

Fields

rank: u32

role: NodeRole

devices: Vec<String>

Backends live on this node, as Device::as_arg tokens (the wire stays stable even if the Device enum gains variants).

platform: String

Free-form platform tag ("ios", "android", "linux", …) for logs and placement heuristics.

Implementations

impl NodeCaps

pub fn accepts(&self, spec: &StageSpec) -> Result<(), String>

Whether this node can run spec at all.

pub fn can_train(&self) -> Result<(), String>

Whether this node can join a training run.

A coordinator must ask before shipping a TrainSpec. Shipping one to a rank that cannot train does not fail loudly: the spec lands on a tag that rank never reads, and then the coordinator blocks forever on the first gradient all-reduce, because a barrier needs every rank to arrive.

Trait Implementations

impl Clone for NodeCaps

fn clone(&self) -> NodeCaps

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for NodeCaps

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

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for NodeCaps

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for NodeCaps

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.