Struct ToolLane 

pub struct ToolLane { /* private fields */ }

The tool lane: the capacity that bounds how many batches execute at once, and the plumbing a batch needs to report its outcome.

Implementations

impl ToolLane

pub fn new( runtime: Handle, results: UnboundedSender<ToolOutcome>, wake: Arc<Notify>, concurrency: usize, stats: Arc<ToolLaneStats>, ) -> Arc<Self>

Build a lane that runs concurrency batches at a time (clamped to at least one, matching ToolLaneStats::new).

pub fn serve( self: &Arc<Self>, jobs: UnboundedReceiver<ToolJob>, ) -> JoinHandle<()>

Start serving jobs. The returned handle completes once the job channel closes (the world is shutting down) and every batch it started has finished, so awaiting it drains the lane.

pub fn relieve(&self, extra: usize) -> usize

Add extra permits, widening the lane.

The relief valve under a lane that has stopped draining: handing out more capacity lets the queued batches run without cancelling anything. Returns how many were added. No longer “for good”: once the jam is over, Self::narrow hands the extra capacity back, so a single historical wedge does not raise the daemon’s peak concurrency (and with it, peak memory) for the rest of its life.

pub fn narrow(&self, upto: usize) -> usize

Take up to upto idle permits back out of the lane, returning how many were reclaimed.

forget_permits only removes permits that are currently available, so this can never stall a batch that is running or block waiting for one to finish - a busy lane just gives back fewer (possibly zero) permits, and the caller tries again on a later healthy cycle.