Struct GpuTransferQueue 

pub struct GpuTransferQueue { /* private fields */ }

GPU Transfer Queue for async bounded transfers

Toyota Way: Heijunka (Load Balancing)

• Bounded queue prevents memory explosion (Poka-Yoke)
• Max 2 in-flight keeps PCIe bus busy without overwhelming GPU
• Async design prevents blocking Tokio reactor

References:

• Leis et al. (2014): Morsel-driven parallelism

Implementations

impl GpuTransferQueue

pub fn new() -> Self

Create new GPU transfer queue with bounded capacity

Returns

Queue with max 2 in-flight transfers

pub async fn enqueue(&self, batch: RecordBatch) -> Result<()>

Enqueue a record batch for GPU transfer

This will block if queue is full (2 batches in-flight)

Errors

Returns error if queue is closed

pub async fn dequeue(&mut self) -> Option<RecordBatch>

Dequeue a record batch from GPU transfer queue

Returns

Next batch if available, None if queue is empty and closed

pub fn sender(&self) -> Sender<RecordBatch>

Get sender for concurrent enqueueing

Trait Implementations

impl Default for GpuTransferQueue

Available on crate feature tokio only.

fn default() -> Self

Returns the “default value” for a type.