Module engine 

The engine tying the write and read paths together.

Sub-ms writes: Table::put appends to the WAL without fsyncing, upserts the skip-list memtable, and updates the in-memory HOT index + secondary indexes. A batch-driven Table::commit does the group fsync and bumps the epoch. Table::flush commits, drains the memtable into an immutable sorted run, and rotates the WAL. Reads merge versions across the live memtable and all sorted runs (Table::get, Table::visible_rows).

Structs

ApproxResult

Point estimate with a normal-theory confidence interval from the reservoir sample (Phase 8.2). ci_low/ci_high bracket point at the requested z-score; the interval has zero width when the sample equals the whole table.

CachedAgg

A cached incremental aggregate (Phase 8.3): the mergeable state, the row-id watermark it covers (rows [0, watermark)), and the snapshot epoch.

ColumnStat

Exact aggregate of a column’s page stats into a min/max/null_count triple (Phase 7.1). Only meaningful when the owning table is insert-only, which Table::exact_column_stats gates on.

IncrementalAggResult

Outcome of Table::aggregate_incremental.

ReadGeneration

S1C-001: an immutable, atomically-published table read view — the engine-layer counterpart of database::TableReadGeneration. Readers pin an Arc<ReadGeneration>; writers publish a replacement with a single ArcSwap store (Table::publish_read_generation) after sealing their active deltas, so no write ever clones the complete table/index set merely because readers exist: every captured piece is either an Arc share of immutable frozen layers or a small metadata copy.

Table

An open MongrelDB table.

TableDeltas

The sealed in-memory deltas captured with a ReadGeneration: memtable, mutable-run tier, HOT primary-key index, and the reverse primary-key map. Each is a post-seal clone — frozen layers are Arc-shared with the writer, the active delta is empty — so capturing copies no row data, and pinning the view keeps exactly the frozen layers it captured alive.

Enums

AggState

A mergeable running aggregate state (Phase 8.3). Two states over disjoint row sets merge into the state over their union, so a cached analytical aggregate can be updated by merging in only the delta (newly inserted rows) instead of a full recompute.

ApproxAgg

A supported approximate aggregate over the reservoir sample (Phase 8.2).

IndexBuildPolicy

When a bulk load (bulk_load / bulk_load_columns / bulk_load_fast) builds the live in-memory indexes.

NativeAgg

A supported native aggregate (Phase 7.2).

NativeAggResult

The typed result of a NativeAgg over a column.

Constants

CACHE_DIR

KEYS_FILENAME

META_DIR

RCACHE_DIR

RUNS_DIR

SCHEMA_FILENAME

WAL_DIR