Struct DaemonConfig 

pub struct DaemonConfig {

    pub claim_batch_size: usize,
    pub default_model_concurrency: usize,
    pub model_concurrency_limits: Arc<DashMap<String, usize>>,
    pub model_escalations: Arc<DashMap<String, ModelEscalationConfig>>,
    pub claim_interval_ms: u64,
    pub max_retries: Option<u32>,
    pub stop_before_deadline_ms: Option<i64>,
    pub backoff_ms: u64,
    pub backoff_factor: u64,
    pub max_backoff_ms: u64,
    pub timeout_ms: u64,
    pub status_log_interval_ms: Option<u64>,
    pub heartbeat_interval_ms: u64,
    pub should_retry: ShouldRetryFn,
    pub claim_timeout_ms: u64,
    pub processing_timeout_ms: u64,
    pub stale_daemon_threshold_ms: u64,
    pub unclaim_batch_size: usize,
    pub cancellation_poll_interval_ms: u64,
    pub batch_metadata_fields: Vec<String>,
    pub purge_interval_ms: u64,
    pub purge_batch_size: i64,
    pub purge_throttle_ms: u64,
}

Configuration for the daemon.

Fields

claim_batch_size: usize

Maximum number of requests to claim in each iteration

default_model_concurrency: usize

Default concurrency limit per model

model_concurrency_limits: Arc<DashMap<String, usize>>

Per-model concurrency overrides (shared, can be updated dynamically)

model_escalations: Arc<DashMap<String, ModelEscalationConfig>>

Per-model escalation configurations for SLA-based model switching Maps model name -> escalation config (e.g., “gpt-4” -> “o1-preview”) When a request is escalated, it’s routed to the escalation_model by the control layer

claim_interval_ms: u64

How long to sleep between claim iterations

max_retries: Option<u32>

Maximum number of retry attempts before giving up.

stop_before_deadline_ms: Option<i64>

Stop retrying (including escalations) this many milliseconds before the batch expires.

• Negative (e.g., -300000 = -5 min): Retry for a buffer window AFTER SLA deadline (recommended)
• Zero: Stop exactly at SLA deadline
• Positive: Stop BEFORE SLA deadline (terminates retryable errors within the SLA deadline, avoid!)
• None: No deadline awareness

Default: 0 (stop retrying/escalating on SLA deadline)

backoff_ms: u64

Base backoff duration in milliseconds (will be exponentially increased)

backoff_factor: u64

Factor by which the backoff_ms is increased with each retry

max_backoff_ms: u64

Maximum backoff time in milliseconds

timeout_ms: u64

Timeout for each individual request attempt in milliseconds

status_log_interval_ms: Option<u64>

Interval for logging daemon status (requests in flight) in milliseconds Set to None to disable periodic status logging

heartbeat_interval_ms: u64

Interval for sending heartbeats to update daemon status in database (milliseconds)

should_retry: ShouldRetryFn

Predicate function to determine if a response should be retried. Defaults to retrying 5xx, 429, 408, and 404 status codes.

claim_timeout_ms: u64

Maximum time a request can stay in “claimed” state before being unclaimed and returned to pending (milliseconds). This handles daemon crashes.

processing_timeout_ms: u64

Maximum time a request can stay in “processing” state before being unclaimed and returned to pending (milliseconds). This handles daemon crashes during execution.

stale_daemon_threshold_ms: u64

Time after a daemon’s last heartbeat before its requests are considered orphaned and returned to pending (milliseconds). Should be significantly larger than heartbeat_interval_ms to avoid reclaiming from live daemons that are merely slow. Also reclaims from daemons explicitly marked dead. Default: 30,000 (30 seconds, 6× the default heartbeat interval).

unclaim_batch_size: usize

Maximum number of stale requests to unclaim in a single poll cycle. Limits database load when many requests become stale simultaneously (e.g., daemon crash).

cancellation_poll_interval_ms: u64

Interval for polling batches to perform finalization and check for cancellations (milliseconds).

This polling loop serves two purposes:

1. Batch Finalization: Fetches active batches and triggers lazy finalization (computing completion timestamps when all requests reach terminal states)
2. Cancellation Detection: Checks if any active batches have been cancelled and aborts their in-flight requests

Default: 5000ms (5 seconds)

batch_metadata_fields: Vec<String>

Batch table column names to include as request headers. These values are sent as x-fusillade-batch-{column} headers with each request. Example: [“id”, “created_by”, “endpoint”] produces headers like:

• x-fusillade-batch-id
• x-fusillade-batch-created-by
• x-fusillade-batch-endpoint

purge_interval_ms: u64

Interval for running the orphaned row purge task (milliseconds). Deletes orphaned request_templates and requests whose parent file/batch has been soft-deleted, for right-to-erasure compliance. Set to 0 to disable purging. Default: 3,600,000 (1 hour).

purge_batch_size: i64

Maximum number of orphaned rows to delete per purge iteration. Each iteration deletes up to this many requests and this many request_templates. Default: 1000.

purge_throttle_ms: u64

Throttle delay between consecutive purge batches within a single drain cycle (milliseconds). Prevents sustained high DB load when many orphans exist. Default: 100.

Trait Implementations

impl Clone for DaemonConfig

fn clone(&self) -> DaemonConfig

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Default for DaemonConfig

fn default() -> Self

Returns the “default value” for a type.

impl<'de> Deserialize<'de> for DaemonConfig

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl From<&DaemonConfig> for RetryConfig

fn from(config: &DaemonConfig) -> Self

Converts to this type from the input type.

impl Serialize for DaemonConfig

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.