Struct CollectionConfig 

#[non_exhaustive]
pub struct CollectionConfig {

    pub name: String,
    pub dimension: usize,
    pub metric: DistanceMetric,
    pub point_count: usize,
    pub schema_version: u32,
    pub storage_mode: StorageMode,
    pub metadata_only: bool,
    pub graph_schema: Option<GraphSchema>,
    pub embedding_dimension: Option<usize>,
    pub pq_rescore_oversampling: Option<u32>,
    pub hnsw_params: Option<HnswParams>,
    pub deferred_indexing: Option<DeferredIndexerConfig>,
    pub async_index_builder: Option<AsyncIndexBuilderConfig>,
    pub auto_reindex_config: Option<AutoReindexConfig>,
    pub streaming_config: Option<StreamingConfig>,
    pub indexed_fields: BTreeSet<String>,
}

Metadata for a collection.

#[non_exhaustive]: new fields are added over time (schema-versioned and serde-defaulted), so external crates must obtain a CollectionConfig via the VectorCollection::create* constructors / Collection::config() rather than a struct literal — this keeps future field additions non-breaking.

Fields (Non-exhaustive)

Non-exhaustive structs could have additional fields added in future. Therefore, non-exhaustive structs cannot be constructed in external crates using the traditional Struct { .. } syntax; cannot be matched against without a wildcard ..; and struct update syntax will not work.

name: String

Name of the collection.

dimension: usize

Vector dimension (0 for metadata-only or graph-without-embeddings collections).

metric: DistanceMetric

Distance metric.

point_count: usize

Number of points in the collection.

schema_version: u32

On-disk schema version for forward-compatibility detection.

When a newer VelesDB version writes a config.json with a higher schema version, older versions will refuse to open the collection rather than silently corrupting data.

Backward compatible: old config.json files without this field deserialize to 1 (the initial version).

storage_mode: StorageMode

Storage mode for vectors (Full, SQ8, Binary).

metadata_only: bool

Whether this is a metadata-only collection.

graph_schema: Option<GraphSchema>

Graph schema — Some iff this is a graph collection. Persisted to config.json; None for vector and metadata collections.

embedding_dimension: Option<usize>

Embedding dimension for graph node vectors (None = no embeddings). Only meaningful when graph_schema is Some.

pq_rescore_oversampling: Option<u32>

PQ rescore oversampling factor. Some(4) by default.

The search pipeline fetches max(k * factor, k + 32) candidates from HNSW and rescores them with full-precision ADC.

• None: disables rescore entirely (expert-only; risks silent recall collapse).
• Some(0): treated as disabled (equivalent to None) — the oversampling factor of 0 produces a candidates count of 0, which falls back to raw HNSW results.
• Some(n) where n > 0: enables rescore with n-fold oversampling.

hnsw_params: Option<HnswParams>

Custom HNSW index parameters (M, ef_construction, etc.).

When Some, these parameters are used to rebuild the HNSW index on collection reopen if no persisted index exists yet (native_meta.bin absent). When None, the default HnswParams::auto(dimension) is used.

Backward compatible: old config.json files without this field deserialize to None.

deferred_indexing: Option<DeferredIndexerConfig>

Deferred indexing configuration (US-366).

When Some and enabled, inserts are buffered in memory and batch-merged into the HNSW index when the buffer reaches merge_threshold. This decouples write latency from index cost.

Backward compatible: old config.json files without this field deserialize to None (disabled).

async_index_builder: Option<AsyncIndexBuilderConfig>

Async index builder configuration (Issue #488 — Bulk Insert V2).

When Some, enables the AsyncIndexBuilder for deferred HNSW insertion during bulk import. Buffered vectors are flushed to the HNSW index via HnswIndex::insert_batch_parallel.

Backward compatible: old config.json files without this field deserialize to None (disabled).

auto_reindex_config: Option<AutoReindexConfig>

Auto-reindex configuration (schema v2 — W2).

When Some, the AutoReindexManager is restored automatically on Collection::open so the policy survives a process restart instead of requiring a manual re-attach.

Backward compatible: v1 config.json files without this field deserialize to None (no manager attached).

streaming_config: Option<StreamingConfig>

Streaming ingestion configuration (schema v2 — STREAM-7).

Describes the persisted shape (channel/batch sizing, flush timing) of the streaming pipeline. The live StreamIngester is still created on demand via Collection::enable_streaming; persisting the config lets a future open-time hook re-enable streaming without a fresh API call.

Backward compatible: v1 config.json files without this field deserialize to None (streaming not configured).

indexed_fields: BTreeSet<String>

Names of payload fields carrying a secondary metadata index (CREATE INDEX (<field>)) — the persisted authority for which indexes exist (EPIC-081 phase 3d).

create_index adds a field here and drop_secondary_index removes it, each persisted via save_config. On Collection::open every listed field is rebuilt from the recovered payloads (backfill), so an index survives a process restart instead of silently vanishing — without which the ordered-index ORDER BY fast path, the bitmap pre-filter, EXPLAIN IndexLookup, and the index advisor would all change behaviour after a restart (results stay correct via the exhaustive fallback).

A BTreeSet so the on-disk ordering is deterministic. Backward compatible: configs written before this field deserialize to an empty set (no indexes restored), and an empty set is not serialized.

Downgrade caveat: a pre-3d binary opening this config ignores the field (no deny_unknown_fields), but the next save_config it performs re-serializes without it, dropping the authority — a subsequent newer binary then will not restore those indexes until CREATE INDEX is re-issued. Bounded and fully recoverable (results stay correct via the exhaustive fallback); no schema-version bump guards it, by design.

Trait Implementations

impl Clone for CollectionConfig

fn clone(&self) -> CollectionConfig

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for CollectionConfig

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for CollectionConfig

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for CollectionConfig

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.