How does this work?, v0.1.0 docs, PLOMID

How does a SELECT work?#

Parse → execute_select (single-table fast path or general join engine) → snapshot scan → evaluate_expression/evaluate_predicate per row → project → aggregate/group/window → order/limit → rows. COUNT(*) without WHERE shortcuts to key count. Time predicates normalized for zone-map pruning.

How does an INSERT work?#

Parse VALUES|SELECT|DEFAULT → materialize source (join engine for SELECT) → execute_insert_rows → enforce constraints/defaults → WAL Data(Put) + buffer → commit fsync → apply pages → install MVCC versions → Inserted(n) or RETURNING rows. ON CONFLICT checks unique gates/B-tree first.

How do UPDATE / DELETE work?#

Locate visible rows under snapshot → evaluate SET/WHERE → stage new versions/tombstones → same commit path. UPDATE…FROM/DELETE…USING materialize the joined relation first (autocommit only for UPDATE…FROM).

How does JSON access work?#

->/->> navigate decoded JsonbValue; #>/@> lower to path/contains functions; JSON_TABLE projects row/column specs via ExpressionEval; subscripts compile to JsonSubscript/ArrayIndex.

How does an index get used?#

Executor encodes SQL key → bytes → B-tree lookup/range_scan or ART lookup; RowIds resolved under MVCC visibility; unique writes reserve transient gates but B-tree remains authority.

How does a transaction work?#

MVCC + Behavior contracts: Begin(snapshot+WAL Begin) → buffer(Data) → Commit(ts+fsync+publish) or Abort.

diagram

flowchart LR
    SQL["write statements"] --> WB["WAL batch\nBegin + Data + Commit+ts"]
    WB --> FS["group fsync\ndurable watermark"]
    FS --> PG["apply pages"]
    PG --> PB["publish versions\n→ visible"]

Diagram source · mermaidcopy included

mermaidsource

flowchart LR
    SQL["write statements"] --> WB["WAL batch\nBegin + Data + Commit+ts"]
    WB --> FS["group fsync\ndurable watermark"]
    FS --> PG["apply pages"]
    PG --> PB["publish versions\n→ visible"]

How does data reach disk?#

Buffer pool write-through on evict → sync + fsync; WAL fsync always first; checkpoint publishes generation + reclaims segments.

How does recovery work?#

WAL / checkpoints / recovery: validate → select checkpoint → replay after boundary → reconstruct → Ready.

How does this work?