Struct TagInvertedIndex 

pub struct TagInvertedIndex { /* private fields */ }

Thread-safe inverted tag index.

Maps "tag_key=tag_value" → set of segment IDs that contain that tag value. This enables O(1) lookup to determine which segments could possibly match a tag filter, avoiding full segment scans.

A reverse index (segment_id → set of tag keys) enables O(k) segment removal where k is the number of tag values for that segment, instead of scanning all entries.

Thread Safety

A single RwLock<InvertedState> is used intentionally — remove_segment must atomically update all three maps (index, reverse, key_values) to prevent partial state on crash/panic. parking_lot::RwLock allows concurrent readers; writers only hold the lock during short segment catalog mutations (add/remove), which are infrequent relative to the query read-path.

Implementations

impl TagInvertedIndex

pub fn new() -> Self

Create an empty inverted tag index.

pub fn with_max_cardinality(max_tag_cardinality: usize) -> Self

Create an empty inverted tag index with a custom cardinality limit.

Limits the number of distinct values tracked per tag key.

pub fn add_segment(&self, segment_id: SegmentId, tag_values: &[(&str, &str)])

Register a segment’s tag values in the index.

For each tag column in the segment’s catalog entry, the distinct values (if available via column stats or direct metadata) are mapped to the segment ID.

pub fn remove_segment(&self, segment_id: SegmentId)

Remove a segment from the index in O(k) time where k is the number of tag values for that segment.

pub fn segments_for_tag(&self, key: &str, value: &str) -> Vec<SegmentId>

Find segments that contain a specific tag value.

pub fn pruner<'a>(&'a self, filters: &[(&str, &str)]) -> TagPruner<'a>

Borrow the index for one query’s worth of pruning decisions.

The lock is taken once and held for the whole candidate sweep, so a pass over N segments costs one acquisition rather than N.

Panics

Never: the returned TagPruner only reads.

pub fn replace_segments( &self, old_segment_ids: &[SegmentId], new_segment_id: SegmentId, new_tag_values: &[(&str, &str)], )

Atomically replace multiple old segments with a single new segment.

Used during compaction to avoid a window where the old segments have been removed but the new one hasn’t been added yet, which would cause concurrent tag-filtered queries to miss data.

pub fn clear(&self)

Clear the entire index (e.g., for a full rebuild).

pub fn entry_count(&self) -> usize

Returns the total number of indexed tag-value pairs.

pub fn values_for_key(&self, key: &str) -> Vec<String>

Returns all distinct values for a given tag key.

O(1) lookup via the secondary key_values index instead of scanning all entries with strip_prefix.

pub fn all_keys(&self) -> Vec<String>

Returns all distinct tag keys present in the index.

O(1) lookup via the secondary key_values index instead of scanning all entries.

Trait Implementations

impl Debug for TagInvertedIndex

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

Formats the value using the given formatter.

impl Default for TagInvertedIndex

fn default() -> Self

Returns the “default value” for a type.