Class SchedulerMetrics

examples/server/server.py:4056–4128

@dataclass
class SchedulerMetrics:
    started_at: float = field(default_factory=time.time)
    requests_submitted_total: int = 0
    requests_admitted_total: int = 0
    requests_completed_total: int = 0
    requests_cancelled_total: int = 0
    requests_failed_total: int = 0
    prompt_tokens_total: int = 0
    prompt_seconds_total: float = 0.0
    tokens_predicted_total: int = 0
    tokens_predicted_seconds_total: float = 0.0
    scheduler_step_seconds_total: float = 0.0
    process_batch_seconds_total: float = 0.0
    sample_seconds_total: float = 0.0
    draft_seconds_total: float = 0.0
    draft_process_seconds_total: float = 0.0
    draft_generate_seconds_total: float = 0.0
    draft_sampled_batch_seconds_total: float = 0.0
    draft_process_calls_total: int = 0
    draft_generate_calls_total: int = 0
    draft_sampled_batch_calls_total: int = 0
    draft_batches_verified_total: int = 0
    draft_target_tokens_verified_total: int = 0
    draft_target_tokens_wasted_total: int = 0
    draft_tokens_reused_as_pending_total: int = 0
    n_decode_total: int = 0
    scheduler_step_calls_total: int = 0
    process_batch_calls_total: int = 0
    sample_calls_total: int = 0
    n_tokens_max: int = 0
    n_busy_slots_total: int = 0
    checkpoint_hits_total: int = 0
    checkpoint_saves_total: int = 0
    checkpoint_evictions_total: int = 0
    sequence_cache_hits_total: int = 0
    sequence_cache_save_requests_total: int = 0
    sequence_cache_lookup_failures_total: int = 0
    sequence_cache_load_failures_total: int = 0
    sequence_cache_save_failures_total: int = 0
    sequence_cache_tokens_loaded_total: int = 0

    def observe_decode(
        self,
        items: Sequence[Any],
        elapsed_seconds: float,
    ) -> None:
        if not items:
            return
        total_tokens = sum(
            item.batch_token_count
            for item in items
        )
        if total_tokens <= 0:
            return
        prompt_tokens = sum(
            item.batch_token_count
            for item in items
            if getattr(item, "kind", None) == "prefill"