HydraDB

Embedded hybrid SQL engine in C# (.NET 8, zero external dependencies in the core).

SQLite-style: in-process, one directory, no server. Postgres-style: MVCC snapshot isolation, multiple concurrent writers, first-updater-wins conflict detection.

Layout

src/HydraDB.Core/Storage   Pager (4 KiB pages), SlottedPage, WAL (CRC32 framed, torn-tail truncation)
src/HydraDB.Core/Txn      MvccStore: row versions (Xmin/Xmax), snapshots, vacuum horizon
src/HydraDB.Core/Log      Commit-record codec (one atomic WAL record per transaction)
src/HydraDB.Core/Sql      Lexer, recursive-descent parser, AST
src/HydraDB.Core/Exec     Evaluator, Executor, PK/FK enforcement
src/HydraDB.Core/Graph    Deterministic golden-angle graph layout over the live catalog
src/HydraDB.Cli           REPL
tests/HydraDB.Tests       Durability, isolation, conflict, constraint tests

Durability model

A transaction buffers its ops in memory. COMMIT writes exactly one WAL record ([len:i32][crc32:u32][payload]), fsyncs it, then publishes the commit sequence number. A record is replayed only if its length is sane and its CRC32 matches, so a torn tail is truncated at open instead of corrupting state. CHECKPOINT writes all committed-live rows into slotted pages, fsyncs the snapshot file, resets the WAL, and vacuums dead versions.

Visibility rule

A version v is visible to transaction T iff

(v.Xmin == T.Tid  or  committed(v.Xmin) and commitSeq(v.Xmin) <= T.SnapshotSeq)
and not (v.Xmax != 0 and (v.Xmax == T.Tid or committed(v.Xmax) and commitSeq(v.Xmax) <= T.SnapshotSeq))

A write to a version whose Xmax belongs to an active transaction, or to a transaction committed after T.SnapshotSeq, raises SerializationConflictException.

Graph layout

Nodes are tables, edges are declared foreign keys. Positions are computed, not simulated: node i sits at r = sqrt(i+1), theta = i * 2.399963229728653 (golden angle), which gives a stable, non-overlapping, repeatable layout with no physics loop. Heat = writes / max(writes).

Build

dotnet run --project src/HydraDB.Cli -- ./hydradata
dotnet test tests/HydraDB.Tests

Supported SQL

CREATE TABLE (INT / TEXT / BOOL, PRIMARY KEY, FOREIGN KEY ... REFERENCES), INSERT, SELECT (projection, WHERE, ORDER BY, LIMIT), UPDATE, DELETE, BEGIN / COMMIT / ROLLBACK.

Known limits

No secondary indexes (scans are linear), no joins, no aggregates, DDL is not rolled back on abort, single-process only.