Struct IndexWriter 

pub struct IndexWriter { /* private fields */ }

Writes documents to a Luci index.

Usage:

1. Create with IndexWriter::new(storage, schema, analyzers)
2. Call put(doc) for each document
3. Call commit() to make documents searchable

Implementations

impl IndexWriter

pub fn new( storage: impl Storage + 'static, schema: Mapping, analyzers: AnalyzerRegistry, ) -> Self

Create a new IndexWriter.

pub fn set_build_threads(&mut self, threads: BuildThreads)

Override the thread budget for the commit-time HNSW connect phase. Production leaves the Ambient default (rayon global pool); deterministic Rust tests/profile harnesses set Fixed(1) to get a bit-identical graph independent of the ambient pool size. Not exposed to the Python API. See [[optimization-concurrent-hnsw-insert]].

pub fn set_analysis_json(&mut self, json: Option<Value>)

Set the analysis settings JSON to persist alongside the mapping.

pub fn load_deletions(&mut self, deletions: DeletionMap)

Load persisted deletions (called on Index::open).

pub fn load_global_hnsw(&mut self, global_hnsw: GlobalHnsw)

Replace the in-memory global HNSW with one deserialized from persisted bytes (called on Index::open).

pub fn global_hnsw(&self) -> &GlobalHnsw

Access the global HNSW for serialization on commit and for taking a search-side snapshot via to_bytes.

pub fn mark_deleted(&mut self, segment_id: SegmentId, doc_id: DocId)

Mark a document as deleted. Takes effect on next search (no commit needed).

pub fn deletions(&self) -> &DeletionMap

Get the current deletion map (for search-time filtering).

pub fn set_memory_budget(&mut self, budget: usize)

Set the memory budget for auto-flush.

pub fn set_write_timeout(&mut self, timeout: Duration)

Set the timeout for acquiring the cross-process write lock.

Default: 5 seconds. If another process holds the write lock, retries until the timeout, then returns WriterLocked.

pub fn add(&mut self, doc: Value) -> Result<()>

Add a JSON document to the index.

The document is validated against the schema, text fields are analyzed, and the result is buffered. Call commit() to make it searchable.

pub fn commit(&mut self) -> Result<()>

Flush the buffer and commit all pending segments to storage.

Persists the current field mappings alongside segment metadata. After commit returns, all documents added via add() are searchable. If the merge policy triggers, a synchronous merge is executed before returning.

pub fn force_merge(&mut self, max_segments: usize) -> Result<()>

Force-merge all segments down to at most max_segments.

Repeatedly merges until the segment count is at or below the target. This is expensive and should only be called after bulk indexing is complete, not during normal operation.

pub fn buffered_doc_count(&self) -> u32

Number of documents in the current (unflushed) buffer.

pub fn discard_buffer(&mut self)

Discard the in-memory segment buffer without flushing to storage.

Used for transaction rollback. Resets the builder to an empty state.