Struct TransportContext 

pub struct TransportContext<'a> { /* private fields */ }

A StrategyContext over one Transport per worker.

This is the piece that was missing: StrategyExecutor was written and had nowhere to run, because nothing implemented the context it takes.

It deliberately does not send GetState/SetState over the wire, even though the worker now answers them: a Fit already returns its trained states in the plan result, so asking again would be a second round trip for something already in hand. Gradients are different — nothing else carries them — so those two do go to the worker.

Implementations

impl<'a> TransportContext<'a>

pub fn new( transports: Vec<Arc<dyn Transport>>, plan: &'a ExecutionPlan, catalog: &'a NodeCatalog, seed: Option<i64>, ) -> Self

One transport per worker, in the order the strategy will index them.

pub fn with_targets(self, identities: Vec<WorkerIdentity>) -> Self

Name the workers, so a partition pinned to an id or a tag can find one. Without this, worker_for refuses rather than guessing.

Trait Implementations

impl StrategyContext for TransportContext<'_>

fn num_workers(&self) -> usize

Number of available workers.

fn execute_on_worker( &self, worker_idx: usize, _plan: &Value, input: &Value, y: Option<&Value>, ) -> Result<HashMap<String, Value>>

Execute a plan on a specific worker (by index). Returns trained states.

fn get_state( &self, worker_idx: usize, node_ids: &[String], ) -> Result<HashMap<String, Value>>

Get trained states from a worker.

fn worker_for(&self, target: &RemoteTarget) -> Result<usize>

Which worker answers to target.

fn execute_partition( &self, worker_idx: usize, node_ids: &[String], input: &Value, y: Option<&Value>, ) -> Result<(Value, HashMap<String, Value>)>

Run part of the graph on a worker, returning the activation and the states it learned.

fn read_back_state( &self, worker_idx: usize, node_ids: &[String], ) -> Result<HashMap<String, Value>>

Read a worker’s state now, over the wire, rather than recalling what its last fit returned.

fn set_state( &self, worker_idx: usize, states: &HashMap<String, Value>, ) -> Result<()>

Set states on a worker (e.g. after aggregation).

fn get_gradients( &self, worker_idx: usize, node_ids: &[String], ) -> Result<HashMap<String, Value>>

Get gradients from a worker.

fn apply_gradients( &self, worker_idx: usize, gradients: &HashMap<String, Value>, ) -> Result<()>

Apply gradients on a worker.