PLOMID is a single-node database with its own SQL engine. It owns its parser, planner surface, executor, MVCC-based transactions, storage, WAL, and PostgreSQL-protocol frontend. It does not embed another database engine. For the repository layout see Repo structure.
Mental model#
flowchart LR
Client["psql / DBeaver / any PG client"] -->|PG wire v3| Server["plomid-server"]
Server --> Exec["SQL execution"]
Exec --> Txn["Transactions + MVCC"]
Exec --> Idx["Indexes"]
Txn --> Store["Storage + WAL"]Diagram source · mermaidcopy included
flowchart LR
Client["psql / DBeaver / any PG client"] -->|PG wire v3| Server["plomid-server"]
Server --> Exec["SQL execution"]
Exec --> Txn["Transactions + MVCC"]
Exec --> Idx["Indexes"]
Txn --> Store["Storage + WAL"]A request moves: parse → execute → transact → index/storage → persist (WAL + checkpoint). Details: Request lifecycle.
What PLOMID is / is not (v0.1.0)#
| PLOMID is | PLOMID is not |
|---|---|
| SQL database with relational + JSON + temporal data | PostgreSQL (only speaks its wire protocol) |
| Persistent engine with WAL + checkpoints + recovery | Distributed / multi-region system |
| Single-writer transactional engine with MVCC snapshots | Graph, blob, or distributed store |
Local-first server + native plomid CLI |
Managed cloud service |
v0.1.0 has no cost-based planner — EXPLAIN describes scan/join/set shapes rather than an index-choice trace. Measure with EXPLAIN ANALYZE (see Performance).
Data model in one paragraph#
Databases contain schemas contain tables. Columns are PostgreSQL-typed (int4, text, numeric, json/jsonb, timestamptz, … — see Types). Rows are versioned by MVCC. Indexes (B-tree persistent, ART in-memory) accelerate equality, range, ordering, and uniqueness. JSON columns mix document and relational access in one query. Time columns use ordinary B-tree ordering plus zone-map/BRIN pruning on columnar data.