Struct EventReplay 

pub struct EventReplay {
    pub max_per_run: usize,
    pub max_runs: usize,
    /* private fields */
}

Bounded per-run event replay buffer. Closes a broadcast race: a client that calls POST /api/v1/scan and then opens the WebSocket would miss RunStarted (and possibly the first few RepoStarted/RepoFailed) frames because tokio’s broadcast::Sender does not replay history. events_ws reads back the snapshot here before joining the live stream so the LiveScanView always sees the run’s lifecycle from the start.

Events that lack a run_id (e.g. plain heartbeats) are not buffered because there is nothing for a subscriber to scope to.

Eviction is least-recently-touched: the side order deque tracks run ids with the most recently pushed-into run at the back. When a new run needs admission past max_runs, the front (oldest activity) is evicted.

Fields

max_per_run: usize

Hard cap on events stored per run. The LiveScanView acceptance set is small (one RunStarted + N RepoStarted/RepoFinished pairs + RunFinished). 128 frames covers ~60 repos before the head is dropped, which is more than the static-pass budget.

max_runs: usize

Cap on tracked runs. Past this we evict the least-recently- touched tracked run. 16 covers the realistic concurrent- LiveScanView count.

Implementations

impl EventReplay

pub fn new() -> Self

pub async fn push(&self, event: &AgentEvent)

Append an event to the per-run buffer. No-op for events that do not carry a run_id.

pub async fn snapshot(&self, run_id: &str) -> Vec<AgentEvent>

Snapshot every buffered event for run_id. Cheap clone.

pub async fn tracked_runs(&self) -> usize

Number of currently tracked runs. Used in tests; cheap.

Trait Implementations

impl Debug for EventReplay

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

Formats the value using the given formatter.

impl Default for EventReplay

fn default() -> Self

Returns the “default value” for a type.