Struct TensorInferenceScheduler 

pub struct TensorInferenceScheduler {
    pub jobs: HashMap<u64, InferenceJob>,
    pub next_job_id: u64,
    pub config: SchedulerConfig,
    pub stats: SchedulerStats,
}

Deadline-aware priority scheduler for TensorLogic inference jobs.

Uses a logical tick counter (driven by the caller) for deadline evaluation and ordering. All operations are synchronous and single-threaded; wrap with a mutex for multi-threaded use.

Fields

jobs: HashMap<u64, InferenceJob>

All known jobs, keyed by job_id.

next_job_id: u64

Monotonically increasing counter used to assign job IDs.

config: SchedulerConfig

Immutable configuration.

stats: SchedulerStats

Incrementally maintained statistics.

Implementations

impl TensorInferenceScheduler

pub fn new(config: SchedulerConfig) -> Self

Creates a new scheduler with the given configuration.

pub fn submit( &mut self, goal: &str, priority: u32, deadline_tick: Option<u64>, estimated_cost_ms: u64, tick: u64, ) -> Option<u64>

Submits a new inference job.

Returns Some(job_id) on success, or None if the pending queue is full.

pub fn tick(&mut self, current_tick: u64)

Advances the scheduler to current_tick.

1. Expires any Running or Pending jobs whose deadline has passed.
2. Promotes Pending jobs (highest priority first, ties broken by job_id ascending) into Running state until max_concurrent slots are filled.

pub fn complete(&mut self, job_id: u64, tick: u64) -> bool

Marks a Running job as Completed.

Returns false if the job is not found or is not currently Running.

pub fn cancel(&mut self, job_id: u64) -> bool

Cancels a Pending or Running job.

Returns false if the job is not found or is already in a terminal state.

pub fn queue_depth(&self) -> usize

Returns the number of jobs currently in Pending state.

pub fn running_jobs(&self) -> Vec<&InferenceJob>

Returns all currently Running jobs, sorted ascending by job_id.

pub fn job(&self, job_id: u64) -> Option<&InferenceJob>

Looks up a job by ID.

pub fn stats(&self) -> &SchedulerStats

Returns a reference to the current scheduler statistics.