LLKV Executor

Work in Progress

llkv-executor is the query execution engine for the LLKV toolkit.

Purpose

• Execute SELECT queries only over table data with projection, filtering, and ordering.
• Provide streaming query results via RecordBatch iterators.
• Integrate with llkv-runtime for transaction-aware query execution.

Executor vs Runtime

The executor (llkv-executor) and runtime (llkv-runtime) serve different purposes:

• Executor: Low-level query evaluation engine that only handles SELECT queries. It takes a SELECT plan and produces streaming Arrow RecordBatch results. The executor knows nothing about transactions, MVCC metadata, or other SQL operations (INSERT, UPDATE, DELETE, CREATE TABLE).

• Runtime: High-level orchestration layer that handles all SQL operations, manages transactions, injects MVCC metadata, and coordinates between storage and execution layers. The runtime invokes the executor for SELECT operations.

In short: Executor = SELECT-only query engine | Runtime = Full SQL coordinator

Design Notes

• The executor works with Arrow RecordBatch data and integrates with llkv-table scanning primitives.
• Query execution is designed for streaming results to avoid materializing entire result sets in memory.
• Invoked by llkv-runtime which provides the transaction context for row visibility filtering.
• Scalar subqueries run through the same evaluation pipeline as other scalar expressions, using planner-supplied correlated column placeholders to pull outer values during execution.

License

Licensed under the Apache-2.0 License.