Struct DiskSpool 

pub struct DiskSpool<R, S> { /* private fields */ }

Durable write-ahead spool tier. See the module docs for the on-disk protocol and crash-recovery semantics.

Construction is infallible and does no I/O; the spool directory is created and leftover segments from a previous run are replayed downstream on the first ingest or flush call.

File I/O uses synchronous std::fs calls: appends are a few kilobytes plus an fsync, which is acceptable to block the runtime for at the collection rates this crate targets.

Implementations

impl<R, S> DiskSpool<R, S>

pub fn new(dir: impl Into<PathBuf>, policy: FlushPolicy, inner: S) -> Self

Create a spool rooted at dir. The pipeline name used when reconstructing WindowMeta for replayed segments is the last component of dir.

pub fn with_pipeline_name(self, name: impl Into<String>) -> Self

Override the pipeline name derived from the directory.

Trait Implementations

impl<R, S> Sink<R> for DiskSpool<R, S>

where R: Serialize + DeserializeOwned + Send + 'static, S: Sink<R>,

type Error = SpoolError<<S as Sink<R>>::Error>

Concrete error type (a thiserror enum, not a boxed error).

async fn ingest( &mut self, _meta: &WindowMeta, records: Vec<R>, ) -> Result<(), Self::Error>

Hand records to this layer. A buffering layer may hold them; a terminal sink ships them immediately.

async fn flush(&mut self) -> Result<(), Self::Error>

Force-drain this layer and everything downstream (shutdown, final flush, startup recovery).