Struct MergeEntry 

pub struct MergeEntry {
    pub run_index: usize,
    pub key: DecoratedKey,
    pub clustering_key: Option<ClusteringKey>,
    pub timestamp: i64,
    pub row_data: RowData,
    pub complex_deletions: Vec<ComplexDeletion>,
    pub range_deletion: Option<RangeTombstone>,
    pub row_deletion: Option<(i64, i32)>,
    pub partition_deletion: Option<(i64, i32)>,
}

Entry in the merge stream

Represents a single row from one of the input SSTables. This is the fundamental unit that flows through the merge heap.

Fields

run_index: usize

Which SSTable this came from (0 = newest)

key: DecoratedKey

Partition key with token

clustering_key: Option<ClusteringKey>

Clustering key (None for tables without clustering)

timestamp: i64

Timestamp in microseconds since Unix epoch

row_data: RowData

Row data (live cells or tombstone)

complex_deletions: Vec<ComplexDeletion>

Complex (collection / UDT) deletion markers for the multi-cell columns of this (pk, ck) (issue #886 substrate).

Carried through the merge so the per-cell-path collection/UDT followup (#844) and shadow-before-purge (#887) can preserve collection/UDT deletion timestamps. reconcile_cluster unions and preserves these across a cluster, but reconciliation does not yet consult them and the writer does not yet apply them — defaults to empty, so output is byte-unchanged. Population (per-element reader emit) lands in #899.

range_deletion: Option<RangeTombstone>

Range-deletion marker covering a span of clustering keys (issue #886 substrate).

A first-class slot so range tombstones can flow through the merge stream instead of being skipped by the parser; applying them to shadow covered cells is the follow-up #846. Carried (and timestamp-max-preserved) through reconcile_cluster but not yet consulted by reconciliation or the writer, so output is byte-unchanged. None when this entry carries no range deletion.

row_deletion: Option<(i64, i32)>

Row-level deletion that COEXISTS with this entry’s surviving live cells (issue #932).

Some((deletion_time_micros, local_deletion_time_secs)) when a row tombstone whose timestamp is OLDER than the surviving cells must be kept alongside RowData::Live. RowData::Live has no row-deletion slot, so — mirroring the carry-only complex_deletions / range_deletion fields — the coexisting deletion rides here instead of forcing a RowData change across ~100 RowData::Live sites. reconcile_cluster folds it into the winning row deletion and re-attaches it to the emitted live entry; merge_entry_to_mutation emits it as Mutation::row_tombstone. Without this, a partial compaction that keeps newer cells would DROP the row deletion and let older cells of other columns (in SSTables not part of the compaction) resurrect. None for a row with no coexisting deletion.

partition_deletion: Option<(i64, i32)>

Partition-level deletion (markedForDeleteAt µs, localDeletionTime s) carried by a synthetic partition-tombstone carrier entry (issue #1072).

Some(..) ONLY on the carrier entry the reader emits for a partition header that has a tombstone (empty RowData::Live, clustering_key: None). The merge extracts the MAX markedForDeleteAt across sources, applies it as the OUTERMOST per-cell <= shadow floor for the whole partition, and re-emits the surviving tombstone via merge_entry_to_mutation (→ Mutation::partition_tombstone). Without this a newer partition tombstone in one SSTable failed to shadow older live rows in another, resurrecting the deleted partition. None for every non-carrier entry.

Implementations

impl MergeEntry

pub fn new( run_index: usize, key: DecoratedKey, clustering_key: Option<ClusteringKey>, timestamp: i64, row_data: RowData, ) -> Self

Create a new merge entry.

The carry-only #886 substrate fields (complex_deletions, range_deletion) default to empty/None; attach them with with_complex_deletions / with_range_deletion once the reader emit surfaces them (#899).

pub fn with_complex_deletions( self, complex_deletions: Vec<ComplexDeletion>, ) -> Self

Attach complex-deletion markers (issue #886 substrate; carry-only).

pub fn with_row_deletion(self, deletion_time: i64, ldt: i32) -> Self

Attach a coexisting row-level deletion (issue #932).

deletion_time is markedForDeleteAt (microseconds); ldt is the localDeletionTime (GC-clock seconds, 0 = not surfaced). Used when a row’s surviving cells are newer than a row tombstone that must still be preserved so it keeps shadowing older cells in non-compacted SSTables.

pub fn with_range_deletion(self, range_deletion: RangeTombstone) -> Self

Attach a range-deletion marker (issue #886 substrate; carry-only).

pub fn with_partition_deletion(self, partition_deletion: (i64, i32)) -> Self

Attach a partition-level deletion (issue #1072).

deletion_time is markedForDeleteAt (microseconds); ldt is the localDeletionTime (GC-clock seconds). Marks this entry as the synthetic partition-tombstone carrier so the merge applies the partition floor and re-emits the surviving tombstone.

pub fn is_metadata_only_no_op(&self) -> bool

True when this entry has NO writer-emittable content at all — an empty RowData::Live { cells: vec![] } with no row tombstone and no carried deletion metadata. Routing such an entry to the writer would create a PHANTOM live empty (pure-PK) row at timestamp 0 (DataWriter::merge_row_group treats a no-op mutation as a primary-key insert), so merge filters it.

Epic #899 Phase C: a COMPLEX DELETION is consumed by the writer (a real CellOperation::ComplexDeletion marker), so a complex-deletion-only entry is NOT a no-op.

Issue #933: a RANGE DELETION is now ALSO consumed by the compaction writer — merge_entry_to_mutation threads it onto the mutation’s range_tombstones, which the writer interleaves as on-disk bound markers AND uses to shadow covered same-partition rows. A range-deletion-only carrier is therefore NO LONGER a no-op and MUST reach the writer (dropping it would resurrect the covered rows it shadowed — roborev #959 High #2). In practice reconcile_cluster never emits a truly-empty entry (one with empty cells, no row tombstone, and no carried metadata reconciles to None), so this guard is now effectively unreachable but kept as a defensive phantom-row filter.

pub fn is_partition_delete_carrier(&self) -> bool

True when this entry is a synthetic partition-tombstone carrier (issue #1072): an empty live row whose only payload is partition_deletion. Produced by [Self::build_merge_entry] from a reader PartitionDelete. Such an entry produces NO output row (it yields a Mutation::partition_tombstone, not a clustering row), so the merge must not count it toward output row counts — but it MUST reach the writer.

Trait Implementations

impl Clone for MergeEntry

Available on crate feature write-support only.

Manual Clone (was #[derive(Clone)]) so the #1664 double-clone regression guard can count MergeEntry clones via a #[cfg(test)]-gated recorder. The clone is field-wise identical to the former derived clone (all 9 fields); in a production (non-test) build the record() call vanishes entirely, so this is a plain field-wise clone with ZERO added cost.

fn clone(&self) -> Self

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for MergeEntry

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

Formats the value using the given formatter.

impl Eq for MergeEntry

impl Ord for MergeEntry

Available on crate feature write-support only.

Ord implementation for min-heap routing ONLY (not LWW winner selection).

This orders entries so the heap yields them grouped by partition and clustering key. The actual equal-timestamp Delete-vs-Live winner is chosen in merge_partition_rows (timestamp → liveness → run_index), NOT here.

Order by:

1. Token (ascending)
2. Key bytes (ascending, for hash collisions)
3. Clustering key (ascending, schema-aware)
4. Run index (ascending) - stable routing tiebreak only

fn cmp(&self, other: &Self) -> Ordering

This method returns an Ordering between self and other.

fn max(self, other: Self) -> Self

where Self: Sized,

Compares and returns the maximum of two values.

fn min(self, other: Self) -> Self

where Self: Sized,

Compares and returns the minimum of two values.

fn clamp(self, min: Self, max: Self) -> Self

where Self: Sized,

Restrict a value to a certain interval.

impl PartialEq for MergeEntry

fn eq(&self, other: &MergeEntry) -> bool

Equality operator ==.

fn ne(&self, other: &Rhs) -> bool

Inequality operator !=.

impl PartialOrd for MergeEntry

Available on crate feature write-support only.

fn partial_cmp(&self, other: &Self) -> Option<Ordering>

This method returns an ordering between self and other values if one exists.

fn lt(&self, other: &Rhs) -> bool

Tests less than (for self and other) and is used by the < operator.

fn le(&self, other: &Rhs) -> bool

Tests less than or equal to (for self and other) and is used by the <= operator.

fn gt(&self, other: &Rhs) -> bool

Tests greater than (for self and other) and is used by the > operator.

fn ge(&self, other: &Rhs) -> bool

Tests greater than or equal to (for self and other) and is used by the >= operator.

impl StructuralPartialEq for MergeEntry