Struct DiskPersister 

pub struct DiskPersister { /* private fields */ }

Implementations

impl DiskPersister

pub fn new(path: impl AsRef<Path>) -> Result<Self>

Open (or create) the JSONL log at path. The parent directory is created if missing.

pub fn path(&self) -> &Path

pub fn replay(&self, ctx: &CellServerCtx) -> Result<usize>

Replay the JSONL log into ctx’s store. Call once after CellServer::builder().build() and before server.run() so the registry is hot before any client connects. Returns the number of entities restored.

pub fn start_watcher( self: &Arc<Self>, ctx: CellServerCtx, ) -> Result<DiskWatcher>

Spawn a filesystem watcher that tails the event log for external appends, truncations, and replacements. Caller must keep the returned DiskWatcher alive for the lifetime of the daemon — it owns the underlying notify debouncer thread.

Self-write suppression: every persist/replay updates applied_offset. The watcher only reads bytes past that mark, so the daemon’s own appends are skipped (file size doesn’t grow beyond the new offset before the watcher tick fires).

Truncation / replacement: detected by current_len < applied_offset or by an inode change. In either case the persister reopens its file handle (so subsequent appends land in the new file, not an unlinked one) and re-runs the dedupe-from-top loader. Pre-existing entities not in the new file are not DEL’d — hot-reload is a recovery affordance, not a full state-replacement primitive.

Trait Implementations

impl Persister for DiskPersister

fn persist(&self, event: MEvent) -> Result<(), PersistError>

Persist a single event.

fn health(&self) -> Arc<PersistHealth>

Health counters for monitoring persist throughput and errors.

fn startup_healthcheck(&self) -> Result<(), String>

Startup healthcheck hook.