Struct WalRecorder 

pub struct WalRecorder { /* private fields */ }

Adapter that lets a Wal act as a MutationRecorder on lora_store::InMemoryGraph::set_mutation_recorder.

Implementations

impl WalRecorder

pub fn new(wal: Arc<Wal>) -> Self

pub fn wal(&self) -> &Arc<Wal>

Underlying log handle. Exposed so admin paths (Database::checkpoint_to, truncate_up_to) can hit the WAL directly without going through the recorder’s transaction state machine.

pub fn arm(&self) -> Result<(), WalError>

Mark the recorder as inside a query critical section. No WAL I/O happens here — Wal::begin is deferred until the first mutation event fires. A pure read query that never produces a MutationEvent therefore costs the WAL nothing: no record allocation, no buffer drain, no fsync.

Errors with WalError::Poisoned if a prior failure has poisoned the recorder, or if the host is double-arming (arm already in effect).

pub fn commit(&self) -> Result<WroteCommit, WalError>

Append a TxCommit for the active transaction (if any) and clear the armed/active state.

Returns:

• WroteCommit::Yes when a lazy TxBegin had been issued and a matching TxCommit was now appended. The host should flush() next under SyncMode::PerCommit.
• WroteCommit::No when no mutations fired during the query and no records were written. The host should skip flush().

pub fn abort(&self) -> Result<bool, WalError>

Append a TxAbort for the active transaction (if any) and clear the armed/active state. Returns Ok(true) when an abort record was actually written, Ok(false) when the query never got far enough to issue a TxBegin.

pub fn flush(&self) -> Result<(), WalError>

Flush the WAL — write the pending buffer to the OS and (under SyncMode::PerCommit) fsync.

pub fn force_fsync(&self) -> Result<(), WalError>

Force the underlying WAL to write, fsync, and advance its durable fence regardless of the configured sync mode. Admin paths use this when they need a durability point immediately.

pub fn checkpoint_marker(&self, snapshot_lsn: Lsn) -> Result<Lsn, WalError>

Append a Checkpoint marker. Used by the checkpoint admin path after a successful snapshot rename — the marker doubles as the log-side fence the next replay will trust.

pub fn truncate_up_to(&self, fence_lsn: Lsn) -> Result<(), WalError>

Drop sealed segments at or below fence_lsn. Forwards to Wal::truncate_up_to.

pub fn is_poisoned(&self) -> bool

True iff the recorder has already failed an append, or the background flusher has latched a failure. Cheap to poll under the engine mutex.

pub fn poison(&self, reason: impl Into<String>)

Quarantine the recorder after the host detects that the live in-memory graph may no longer match durable state. Once poisoned, future query arms fail until the database is restarted from a snapshot + WAL.

Trait Implementations

impl MutationRecorder for WalRecorder

fn record(&self, event: &MutationEvent)

fn poisoned(&self) -> Option<String>

Sticky failure flag for durability-shaped recorders.