Struct Request 

pub struct Request {

    pub request_id: String,
    pub priority: i32,
    pub arrival_time: f64,
    pub status: RequestStatus,
    pub num_prompt_tokens: u32,
    pub max_output_tokens: u32,
    pub target_output_tokens: u32,
    pub num_computed_tokens: u32,
    pub num_output_tokens: u32,
    pub num_tokens: u32,
    pub num_cached_tokens: u32,
    pub prompt_block_hashes: Vec<u64>,
    pub kv_blocks: Vec<BlockId>,
    pub num_preemptions: u32,
    pub first_token_time: Option<f64>,
    pub completion_time: Option<f64>,
    pub token_generation_times: Vec<f64>,
    pub preempted_time: f64,
    pub last_preempted_at: Option<f64>,
}

Request represents a single inference request in the simulation

Fields

request_id: String

Unique request ID

priority: i32

Client priority (lower = higher priority)

arrival_time: f64

Arrival time (simulated time)

status: RequestStatus

Request status

num_prompt_tokens: u32

Number of input tokens

max_output_tokens: u32

Maximum number of output tokens to generate

target_output_tokens: u32

Actual number of output tokens to generate (sampled, may be less than max) This simulates hitting an EOS token

num_computed_tokens: u32

Number of tokens computed so far

num_output_tokens: u32

Number of output tokens generated so far

num_tokens: u32

Total tokens (prompt + output)

num_cached_tokens: u32

Number of prefix-cached tokens (set by cache manager)

prompt_block_hashes: Vec<u64>

Synthetic block hashes for prefix caching modeling In synthetic mode: pre-generated hashes (some shared, some unique) In semantic mode: will be computed from actual token content

kv_blocks: Vec<BlockId>

KV cache blocks allocated to this request

num_preemptions: u32

Number of times this request has been preempted

first_token_time: Option<f64>

Time when first token was generated (TTFT tracking)

completion_time: Option<f64>

Time when request completed

token_generation_times: Vec<f64>

Per-token generation times

preempted_time: f64

Time spent preempted (not running)

last_preempted_at: Option<f64>

Last preemption start time

Implementations

impl Request

pub fn new_with_target( request_id: String, priority: i32, arrival_time: f64, num_prompt_tokens: u32, max_output_tokens: u32, target_output_tokens: u32, ) -> Self

Create a new request with a target output length

pub fn new( request_id: String, priority: i32, arrival_time: f64, num_prompt_tokens: u32, max_output_tokens: u32, ) -> Self

Create a new request (target = max)

pub fn get_prompt_block_hashes(&self) -> &[u64]

Get block hashes for the prompt These should be thought of as ‘incremental hashes’ - i.e. the hash of block n is the hash of all the tokens up to that block (not just that block alone). In synthetic mode: returns pre-generated hashes In semantic mode: will compute from actual token content

pub fn is_prefill(&self) -> bool

Check if this is in prefill phase

pub fn tokens_to_process(&self) -> u32

Get number of tokens needed to process

pub fn is_finished(&self) -> bool

Check if request is done

pub fn total_tokens(&self) -> u32

Get total tokens (prompt + max output)

pub fn remaining_tokens(&self) -> u32

Get remaining tokens to process

pub fn kv_cache_size(&self, model: &ModelConfig) -> u64

Calculate KV cache requirement for this request

pub fn record_generated_tokens( &mut self, num_new_tokens: u32, current_time: f64, )

Record that tokens were generated (update output token count and total)

pub fn mark_preempted(&mut self, current_time: f64)

Mark request as preempted

pub fn resume(&mut self, current_time: f64)

Resume a preempted request

Trait Implementations

impl Clone for Request

fn clone(&self) -> Request

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for Request

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.