Struct IngestLedger 

pub struct IngestLedger { /* private fields */ }

Content-hash idempotency ledger for zeph knowledge ingest.

Records (source_uri, content_hash) pairs of inputs that have been successfully embedded so that unchanged files are skipped on the next run (spec-067 INV-5).

This is a re-read / cost guard only — it does NOT reconcile LLM extraction drift across model versions (INV-5). An unchanged (source_uri, content_hash) pair skips re-embedding; a changed hash for the same URI is treated as a new input and produces a new ledger row while leaving any previous Qdrant chunks in place (stale-point cleanup is Phase 2).

Implementations

impl IngestLedger

pub fn new(pool: DbPool) -> Self

Creates a new IngestLedger wrapping the given database pool.

The pool should come from SemanticMemory::sqlite().pool().clone() so the ledger shares the agent’s existing connection pool rather than opening a new file.

pub fn content_hash(bytes: &[u8]) -> String

Computes the BLAKE3 hex digest of bytes.

This is the canonical content-hash format used as the content_hash column value. Callers must use this function (rather than a different hash algorithm or encoding) to ensure ledger keys are consistent across invocations.

Examples

use zeph_memory::graph::ingest::IngestLedger;

let hash = IngestLedger::content_hash(b"hello world");
assert_eq!(hash.len(), 64, "BLAKE3 hex digest is always 64 characters");

pub async fn is_ingested( &self, source_uri: &str, content_hash: &str, ) -> Result<bool, MemoryError>

Returns true if the (source_uri, content_hash) pair is already recorded in the ledger.

When this returns true, the caller should skip re-embedding the input — its content has not changed since it was last ingested (spec-067 FR-012).

Errors

Returns MemoryError::Sqlx on database failure.

pub async fn mark_ingested( &self, source_uri: &str, content_hash: &str, batch_id: &str, entities: i64, edges: i64, ) -> Result<(), MemoryError>

Records a successful ingest of (source_uri, content_hash).

Idempotent: if the pair already exists the row is updated in place with the new batch_id, ingested_at, entities, and edges values (ON CONFLICT … DO UPDATE).

entities and edges should be 0 for the notes sink (Phase 1). Non-zero values are reserved for Phase 2 graph extraction.

Errors

Returns MemoryError::Sqlx on database failure.

pub async fn batch_exists(&self, batch_id: &str) -> Result<bool, MemoryError>

Returns true if any row in the ledger has the given import_batch_id.

Errors

Returns MemoryError::Sqlx on database failure.

pub async fn resolve_batch_id( &self, prefix: &str, ) -> Result<BatchIdResolution, MemoryError>

Resolves a (possibly abbreviated) batch_id prefix to a full import_batch_id.

Mirrors git’s unambiguous short-hash resolution: an exact match on a full id wins immediately; otherwise a prefix that uniquely identifies one batch resolves to it, a prefix shared by several batches is reported as BatchIdResolution::Ambiguous, and a prefix matching no batch is reported as BatchIdResolution::NotFound. This lets zeph knowledge rollback --batch-id accept the 8-character prefix printed by zeph knowledge status (#5399).

An empty (or whitespace-only) prefix always resolves to BatchIdResolution::NotFound, even when the ledger is non-empty — every id trivially starts_with(""), so without this guard a blank prefix (e.g. an unset --batch-id "$VAR" shell substitution) would silently resolve to the sole batch in a single-batch ledger instead of being rejected.

Errors

Returns MemoryError::Sqlx on database failure.

pub async fn delete_batch(&self, batch_id: &str) -> Result<u64, MemoryError>

Deletes all ledger rows for the given import_batch_id.

Returns the number of rows removed.

Errors

Returns MemoryError::Sqlx on database failure.

pub async fn delete_batch_in_tx( &self, batch_id: &str, tx: &mut DbTransaction<'_>, ) -> Result<u64, MemoryError>

Deletes all ledger rows for batch_id within a caller-provided transaction.

Executes the same DELETE as Self::delete_batch but uses tx so the caller can combine it with other writes in a single atomic unit. Returns the number of rows removed.

Errors

Returns MemoryError::Sqlx on database failure.

pub async fn summary(&self) -> Result<Vec<LedgerEntry>, MemoryError>

Returns all ledger rows ordered by ingested_at descending, newest first.

This is the data source for zeph knowledge status.

Errors

Returns MemoryError::Sqlx on database failure.

Trait Implementations

impl Clone for IngestLedger

fn clone(&self) -> IngestLedger

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.