Skip to content

STORAGE_ENGINE.md

Latest commit

 

History

History
89 lines (56 loc) · 3.13 KB

STORAGE_ENGINE.md

File metadata and controls

89 lines (56 loc) · 3.13 KB

Storage engine

This document describes data storage: heap pages, tuple identifiers (TID), the buffer pool, and the on-disk layout.

Core concepts

Heap table

A table is stored as a sequence of fixed-size heap pages. Each inserted row is placed into a free slot on a page.

TID (Tuple Identifier)

TID is an address of a tuple within the heap storage:

  • pageId — page number
  • slotId — slot number within the page

In code: src/main/java/ru/open/cu/student/index/TID.java.

Heap pages

File: src/main/java/ru/open/cu/student/memory/page/HeapPage.java

The page size is 8KB. The structure follows a classic slot-based layout:

┌────────────────────────────────────────────────────┐
│ Header                                              │
├────────────────────────────────────────────────────┤
│ Slot array (растёт вниз)                            │
├────────────────────────────────────────────────────┤
│ Free space                                          │
├────────────────────────────────────────────────────┤
│ Tuples (растут вверх)                               │
└────────────────────────────────────────────────────┘

This layout supports variable-length tuples and allows compacting/reclaiming space.

Tuple serializer

File: src/main/java/ru/open/cu/student/memory/serializer/HeapTupleSerializer.java

Serialization/deserialization converts the logical tuple representation into bytes suitable for writing into heap pages.

PageFileManager

File: src/main/java/ru/open/cu/student/memory/manager/HeapPageFileManager.java

Responsibilities:

  • read/write fixed-size pages
  • manage table data files

Buffer pool

File: src/main/java/ru/open/cu/student/memory/buffer/DefaultBufferPoolManager.java

The buffer pool caches pages in memory to reduce disk I/O.

Key points:

  • eviction strategy: LRU (LRUReplacer) or Clock (ClockReplacer)
  • dirty pages are written on eviction and/or explicit flush
  • thread-safety via locks at the manager level

On-disk layout

The storage engine keeps its files under data/ (created automatically):

data/
├── table_definitions.dat
├── column_definitions.dat
├── index_definitions.dat
├── <tableId>.dat
└── indexes/
    ├── <hashIndexName>.idx
    └── <btreeIndexName>.bpt   (см. ограничения по персистентности)

Catalog metadata is maintained by CatalogManager and IndexManager.

Limitations (intentional)

  • No WAL/redo logging and no production-grade durability guarantees.
  • Concurrency is simplified; the server serializes queries.