Struct EqualityDeleteFilter 

pub struct EqualityDeleteFilter { /* private fields */ }

In-memory equality delete filter built from one or more delete files.

Each entry is one source delete file’s own sequence number plus its column_name → set of string-normalised values to delete. A data row is deleted by a given entry if, for every column in that entry’s predicate, the row’s value is a member of that column’s set (AND, not OR) — and the entry’s sequence number is strictly greater than the data file’s own (Iceberg spec: a delete only applies to files committed strictly before it). Kept as a Vec of per-file predicates rather than one merged map — unlike a plain column→values union, the per-file sequence number can’t be collapsed away: two delete files touching the same column need their masks evaluated against different data-file sequence numbers.

Implementations

impl EqualityDeleteFilter

pub async fn from_files( store: &Arc<dyn Store>, files: &[EqualityDeleteFile], ) -> AilakeResult<Self>

Build filter from a list of equality delete file references.

For each file, downloads the Avro payload from store, extracts (column_name, value) pairs, and keeps the file’s own EqualityDeleteFile::sequence_number alongside — needed later so should_delete_row/apply can skip predicates that don’t apply to the data file currently being scanned.

pub fn empty() -> Self

pub fn is_empty(&self) -> bool

pub fn should_delete_row( &self, batch: &RecordBatch, row_idx: usize, data_file_sequence_number: i64, ) -> bool

Check whether a single row (by its physical index in batch) matches any applicable delete predicate.

Returns true if the row should be logically deleted. data_file_sequence_number is the sequence number of the data file batch was read from (DataFileEntry::sequence_number) — a delete predicate only applies when its own sequence number is strictly greater than this, per Iceberg spec. This is what lets an insert and a delete of the same key committed in the same snapshot (equal sequence numbers) leave the insert visible, instead of the delete masking the row it was committed alongside.

pub fn apply( &self, batch: RecordBatch, data_file_sequence_number: i64, ) -> AilakeResult<RecordBatch>

Apply the filter to batch, returning a new batch with matching rows removed.

data_file_sequence_number — see should_delete_row doc.