Type Alias SchedulerConfig 

pub type SchedulerConfig = SchedulerConfig;

Scheduler configuration

Aliased Type

pub struct SchedulerConfig {

    pub policy: SchedulingPolicy,
    pub max_waiting_requests: usize,
    pub max_running_requests: usize,
    pub enable_preemption: bool,
    pub enable_load_balancing: bool,
    pub fair_share_weights: HashMap<String, f32>,
    pub enable_sla_enforcement: bool,
    pub prompt_token_estimate: bool,
    pub prefill_first_until_active: Option<usize>,
    pub prefill_step_chunk: Option<usize>,
    pub active_decode_prefill_chunk: Option<usize>,
    pub scheduler_none_prof: bool,
    pub sequence_fit_policy: SequenceFitPolicy,
}

Fields

policy: SchedulingPolicy

Scheduling policy

max_waiting_requests: usize

Maximum waiting queue size

max_running_requests: usize

Maximum running requests

enable_preemption: bool

Enable request preemption

enable_load_balancing: bool

Enable load balancing

fair_share_weights: HashMap<String, f32>

Fair share weights per client

enable_sla_enforcement: bool

SLA enforcement enabled

prompt_token_estimate: bool

Use prompt-token metadata for initial continuous-batch admission estimates.

prefill_first_until_active: Option<usize>

Prefer new prefills over early decodes until this many requests are active.

prefill_step_chunk: Option<usize>

Optional hard cap for per-request prefill chunks. None spends the live per-step token budget and lets capacity feedback narrow or regrow each request independently.

active_decode_prefill_chunk: Option<usize>

Cap prefill admission chunks only while decode requests are already active.

scheduler_none_prof: bool

Emit diagnostic scheduler None/SOME decisions.

sequence_fit_policy: SequenceFitPolicy

Non-reserving sequence fit gate used before prefill admission.