Persistent data path
Storage Engine & B+Tree
Local rows, indexes, pages, caching, and recovery boundaries in storage format v3.
4 KB slotted pages
asadb_page_manager.pl ↗ defines the fixed 4 KiB ASPG page format: a 32-byte header, page type, ID, record and slot counts, free-space bounds, sibling pointers, and checksum. Slots grow forward while payload grows backward from the page end.
Records and page I/O
asadb_record_manager.pl ↗ manages typed rows, heap pages, RIDs, scans, updates/deletes, index invalidation, snapshots, append undo, mutation backup, and recovery. asadb_pager.pl ↗ performs fixed-offset disk I/O, allocation, scanning, flushing, and invalidation.
Persistent B+Tree
asadb_btree.pl ↗ supplies compatibility-tree helpers and the persistent B+Tree. Page zero holds index metadata; internal and linked leaf pages support equality/range lookups and ordered RID traversal. The bulk builder uses 2,048-entry external sorted runs so a complete index list does not remain in the Prolog heap.
- Equality lookup descends from the root.
- Range and ordered scans follow linked leaves.
- Indexed
ORDER BY ... LIMITcan stop at its result window. - The bounded buffer pool uses Clock or LRU replacement and never evicts pinned pages.
Recovery boundary: AsaDB uses append undo, mutation backups, transaction snapshots, and atomic replacement. It is not ARIES and does not currently provide MVCC.