Struct AilakeFileReader 

pub struct AilakeFileReader { /* private fields */ }

Implementations

impl AilakeFileReader

pub fn new(bytes: Bytes, vector_column: &str, dim: u32) -> Self

pub fn ailk_offset(&self) -> AilakeResult<u64>

Returns the absolute byte offset of the primary AILK section.

Tries ailake.footer_offset from Parquet KV metadata first (files written by write() / write_multi()). Falls back to AilakeTrailer bootstrap for files produced by write_single_pass() / write_multi_single_pass().

pub fn ailk_offset_for_column(&self, column: &str) -> AilakeResult<u64>

Returns the absolute byte offset of the AILK section for a named vector column.

Resolution order:

1. ailake.{column}.footer_offset KV (extra columns in multi-column files)
2. ailake.footer_offset KV (primary column or single-column files)
3. AilakeTrailer scan (streaming / single-pass files without KV injection)

pub fn has_column_footer(&self, column: &str) -> bool

True iff this file has its own AILK section specifically for column — i.e. a ailake.{column}.footer_offset KV entry is present.

Unlike ailk_offset_for_column(), this does not fall back to the primary column’s footer or the trailer bootstrap: those fallbacks make ailk_offset_for_column(...).is_ok() return true for any AI-Lake file regardless of column, which is correct for “give me an offset to read” but wrong for “does this specific secondary column exist” — e.g. an idempotency check before adding a new vector column to files that don’t have it yet.

pub fn is_ailake_file(&self) -> bool

Returns true if the file contains an embedded AILK section.

pub fn read_header(&self) -> AilakeResult<AilakeHeader>

Parse the 64-byte AI-Lake header from the embedded AILK section.

pub fn read_header_for_column(&self, column: &str) -> AilakeResult<AilakeHeader>

Parse the 64-byte AI-Lake header for a named vector column.

Uses ailake.{column}.footer_offset for extra columns and falls back to ailake.footer_offset for the primary column (single-column files).

pub fn get_centroid(&self) -> AilakeResult<Centroid>

Read centroid + radius from the AILK section.

pub fn load_index(&self) -> AilakeResult<HnswIndex>

Load the HNSW index from the primary AILK section.

pub fn load_index_for_column(&self, column: &str) -> AilakeResult<HnswIndex>

Load the HNSW index for a specific vector column.

Works for both single-column files (falls back to primary AILK) and multi-column files written with AilakeFileWriter::write_multi.

pub fn load_any_index(&self) -> AilakeResult<AnyIndex>

Load primary index as AnyIndex, dispatching on header flags.

pub fn load_any_index_for_column(&self, column: &str) -> AilakeResult<AnyIndex>

Load index for a specific vector column as AnyIndex.

pub fn read_parquet(&self) -> AilakeResult<(RecordBatch, Vec<Vec<f32>>)>

Read the Parquet section (tabular data + decoded embeddings). The full file is valid Parquet; the AILK section is invisible to standard readers.

pub fn read_parquet_filtered( &self, filter: &ColumnFilter, ) -> AilakeResult<(RecordBatch, Vec<Vec<f32>>)>

Same as read_parquet, but pushes filter down to the Parquet read — see ParquetVectorReader::read_all_filtered for the row-group skip + exact RowFilter mechanics.

pub fn matching_row_ids( &self, filter: &ColumnFilter, ) -> AilakeResult<HashSet<u64>>

Returns the set of original (file-relative) row indices matching filter, without disturbing row identity — see ParquetVectorReader::matching_row_ids. Use this (not read_parquet_filtered) wherever the caller also indexes rows by an externally-computed position (HNSW row_id, deletion vector bitmap).

pub fn load_fts_blob(&self) -> AilakeResult<Option<Bytes>>

Load the raw FTS blob from the AILK_FTS section, if present.

Returns Ok(None) when the file has no FTS section (opt-in feature). The returned bytes can be passed directly to ailake_fts::FtsSearcher::from_blob.

pub fn verify_integrity(&self) -> AilakeResult<()>

Verify the positional invariant: Parquet record_count == index node_count.

Dispatches via load_any_index() (HNSW or IVF-PQ, per header.flags) — unlike load_index(), which always assumes HNSW and errors out on an IVF-PQ-indexed file.