Enum LowPriCommand 

pub enum LowPriCommand {
    WriteConceptsChunk {
        chunk: Vec<ConceptUpsert>,
        responder: Sender<Result<ChunkOutcome>>,
    },
    WriteAnalyticsChunk {
        chunk: Vec<Annotation>,
        responder: Sender<Result<ChunkOutcome>>,
    },
    UpsertEmbeddingChunk {
        model: ModelName,
        chunk: Vec<(String, Vec<f32>)>,
        responder: Sender<Result<ChunkOutcome>>,
    },
    BulkImportChunk {
        chunk: Vec<EdgeAssertion>,
        responder: Sender<Result<ChunkOutcome>>,
    },
    Archive {
        cutoff: String,
        archive_path: PathBuf,
        responder: Sender<Result<ArchiveReport>>,
    },
    Rehydrate {
        ids: Vec<String>,
        archive_path: PathBuf,
        responder: Sender<Result<RehydrateReport>>,
    },
    RebuildFts {
        responder: Sender<Result<()>>,
    },
    Analyze {
        incremental: bool,
        responder: Sender<Result<()>>,
    },
    ShadowRebuild {
        step: ShadowStep,
        responder: Sender<Result<ShadowOutcome>>,
    },
}

Commands sent to the Write Actor on the low-priority channel (background work).

Variants

WriteConceptsChunk

One chunk of concepts — a ledger write, logged and versioned.

Fields

chunk: Vec<ConceptUpsert>

responder: Sender<Result<ChunkOutcome>>

WriteAnalyticsChunk

One chunk of derived annotations — off-ledger, no log trigger (D-041).

The pair is named apart deliberately: this variant was WriteAnalyticsChunk beside a WriteAnnotationsChunk that carried concepts, which is the crossing D-075 undid.

Fields

chunk: Vec<Annotation>

responder: Sender<Result<ChunkOutcome>>

UpsertEmbeddingChunk

One chunk of vectors for one model (§5.9, D-048).

Low priority: embedding is bulk derived work and must never preempt an interactive assertion.

Fields

model: ModelName

chunk: Vec<(String, Vec<f32>)>

responder: Sender<Result<ChunkOutcome>>

BulkImportChunk

Fields

chunk: Vec<EdgeAssertion>

responder: Sender<Result<ChunkOutcome>>

Archive

Fields

cutoff: String

archive_path: PathBuf

responder: Sender<Result<ArchiveReport>>

Rehydrate

Move named concepts back out of the cold file (0.9.0, C3).

Low priority for the same reason Archive is: it is bulk physical movement with no latency bound, and it holds the write lock for its whole transaction.

Fields

ids: Vec<String>

archive_path: PathBuf

responder: Sender<Result<RehydrateReport>>

RebuildFts

Reconstruct the FTS index from concepts (§5.9, D-036, D-051).

Low priority: it is maintenance on a derivative table, and a search index that is a few seconds stale is a smaller cost than an interactive write that waits behind a full reindex.

Fields

responder: Sender<Result<()>>

Analyze

Refresh or top up the query planner’s statistics (0.12.4, D-149).

Low priority, and not a close call: statistics being a few seconds stale costs a plan that was already the plan a moment ago, where preempting an interactive assertion costs a caller their latency bound. It is a write — it writes sqlite_stat1 — so it takes the write lock like anything else, and PRAGMA analysis_limit in configure is what keeps the hold a function of the index count instead of the table size.

Fields

incremental: bool

true runs PRAGMA optimize, which re-analyses only what SQLite believes has gone stale; false runs ANALYZE unconditionally.

responder: Sender<Result<()>>

ShadowRebuild

One step of a chunked shadow rebuild (§5.8, T1.2, D-082).

Low priority, and one command per step rather than one per rebuild: the whole value of building beside the live table is that the actor returns here between chunks. See Database::rebuild_current_chunked.

Fields

step: ShadowStep

responder: Sender<Result<ShadowOutcome>>