Struct RetrievalFailureLogger 

pub struct RetrievalFailureLogger { /* private fields */ }

Async background writer that batches retrieval failure records to SQLite.

Construct with RetrievalFailureLogger::new and call RetrievalFailureLogger::log from the recall hot path. Records are sent via a bounded channel; if the channel is full the record is silently dropped (zero hot-path latency, per INV-1).

Fields are Option so that shutdown can take them without a move-out-of-Drop conflict, and so the Drop impl can abort any task not yet drained. tx is declared before handle to ensure the channel is closed before the handle is dropped, which allows the background task to exit cleanly when Drop fires.

Implementations

impl RetrievalFailureLogger

pub fn new( sqlite: SqliteStore, channel_capacity: usize, batch_size: usize, flush_interval: Duration, retention_days: u32, ) -> Self

Spawn the background writer task and return a logger handle.

batch_size records are flushed at once, or after flush_interval elapses, whichever comes first. Old records are purged every CLEANUP_FLUSH_INTERVAL batch flushes according to retention_days.

pub fn log(&self, record: RetrievalFailureRecord)

Queue a retrieval failure record for async persistence.

Both query_text (512 chars) and error_context (256 chars) are truncated before enqueueing to bound in-channel memory usage (INV-3). If the channel is full the record is dropped and a debug message is emitted (INV-1).

pub async fn shutdown(self)

Shut down the background writer, draining any queued records.

Closes the sender and waits for the background task to complete. Drop is best-effort only; call this method for a clean drain on process exit.

Trait Implementations

impl Drop for RetrievalFailureLogger

fn drop(&mut self)

Abort the background writer task on drop.

For a clean drain (flushing queued records) call RetrievalFailureLogger::shutdown explicitly before dropping. This impl ensures the task is not silently detached.