Struct IndexConfig 

pub struct IndexConfig {

    pub num_threads: usize,
    pub num_indexing_threads: usize,
    pub num_compression_threads: usize,
    pub term_cache_blocks: usize,
    pub store_cache_blocks: usize,
    pub max_indexing_memory_bytes: usize,
    pub merge_policy: Box<dyn MergePolicy>,
    pub optimization: IndexOptimization,
    pub reload_interval_ms: u64,
    pub max_concurrent_merges: usize,
    pub background_merge_permits: Arc<Semaphore>,
    pub merge_bp_time_budget: Option<Duration>,
    pub bp_memory_budget_bytes: usize,
    pub background_reorder_permits: Arc<Semaphore>,
    pub background_reorder_pool: Option<Arc<ThreadPool>>,
}

Index configuration

Fields

num_threads: usize

Number of threads shared by CPU-intensive search work.

Indexes in the same process that request the same width reuse one Rayon pool. A value of zero is invalid and is rejected by Index::create and Index::open.

num_indexing_threads: usize

Number of parallel segment builders (documents distributed round-robin)

num_compression_threads: usize

Number of threads for parallel block compression within each segment

term_cache_blocks: usize

Block cache size for term dictionary per segment

store_cache_blocks: usize

Block cache size for document store per segment

max_indexing_memory_bytes: usize

Max memory (bytes) across all builders before auto-commit (global limit)

merge_policy: Box<dyn MergePolicy>

Merge policy for background segment merging

optimization: IndexOptimization

Index optimization mode (adaptive, size-optimized, performance-optimized)

reload_interval_ms: u64

Reload interval in milliseconds for IndexReader (how often to check for new segments)

max_concurrent_merges: usize

Maximum number of concurrent background merges per index (default: 4)

background_merge_permits: Arc<Semaphore>

Application-wide background merge gate shared by clones of this config. The per-index limit alone multiplied large merge working sets by the number of active indexes.

merge_bp_time_budget: Option<Duration>

Wall-clock budget for merge-time BP reorder per field (only applies when the index has reorder_on_merge). A truncated pass still writes a valid, better-ordered segment; it is marked bp_converged = false and the background optimizer deepens it later (warm-started). None = unbudgeted (BP runs to full depth inside the merge, which can hold a merge slot for 10-30+ minutes on 10M+ doc outputs).

bp_memory_budget_bytes: usize

Memory budget (bytes) for the BP forward index during reorder passes (merge-time and background). When a large segment’s forward index would exceed this, the highest-df dims are dropped from BP’s input (logged loudly) — clustering quality degrades gracefully. Production evidence: 18M-doc merges exceeded the former 2 GB default and dropped ~10% of eligible dims; hosts with less headroom may lower this.

background_reorder_permits: Arc<Semaphore>

Hard limit on simultaneous whole-segment BP rewrites. This is shared by all indexes opened from clones of this config and applies to optimizer, merge-time, and manual reorder passes. It is deliberately separate from the Rayon pool width: one pass can already use every background CPU thread and consume the full BP memory budget.

background_reorder_pool: Option<Arc<ThreadPool>>

Optional process/application-owned Rayon pool for BP work. Supplying one lets every index and the optimizer share the same worker threads; None lazily uses one process-wide cores/2 fallback pool.

Trait Implementations

impl Clone for IndexConfig

fn clone(&self) -> IndexConfig

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for IndexConfig

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

Formats the value using the given formatter.

impl Default for IndexConfig

fn default() -> Self

Returns the “default value” for a type.