Posts

  • pg_vexec one month on: ahead of ClickHouse on Parquet, close behind on MergeTree

    A month after I ported Greenplum to PostgreSQL 19, pg_vexec beats ClickHouse on Parquet and nearly catches up on MergeTree in ClickBench, and a rough draft of a GPU version is faster still. I've published the telemetry from the runs: charts, metrics, DBMS logs and numbers from deep inside the engine.
  • Which language should an AI agent use to search its memory: SQL or SPARQL?

    I gave an agent the same knowledge in two forms, 34 PostgreSQL tables and a graph over them through R2RML, and asked three open-weight models 15 questions through SQL, SPARQL or both. Plus pgvector against Lance for semantic search with filters, and the SQL injection DeepSeek came up with to answer my question.
  • A knowledge base without a second database: Oxigraph's SPARQL in PostgreSQL 19

    pg_knowledge_base runs Oxigraph's SPARQL engine inside PostgreSQL 19. RDF triples live in ordinary tables, R2RML mappings bring existing tables into the same SPARQL queries, and Oxigraph's HTTP endpoints are still there for any SPARQL client, while transactions, replicas and backups come from PostgreSQL.
  • My PostgreSQL is faster than yours and has almost caught up with ClickHouse

    ClickHouse's MergeTree table engine plugged into PostgreSQL 19 as a table access method, without a fork or a single core patch, and read by the pg_vexec vectorized executor, plus iheap, a table access method that updates rows in place. On my laptop, ClickBench on 10 million rows runs only 3 times slower than in ClickHouse on the same machine.
  • Vectorized Query Execution in PostgreSQL 19: Arrow, ORCA and Flight SQL in a PostgreSQL Extension

    pg_vexec, a family of extensions for PostgreSQL 19: a vectorized planner and executor, an Arrow Flight SQL endpoint, and kernel packs for pgvector and PostGIS. PostgreSQL's planner and ORCA choose vector nodes by cost, the server needs no core patches, and in the Cloudberry port, batches travel between segments as Arrow IPC frames.
  • Greenplum without the fork: Apache Cloudberry as PostgreSQL 19 Extensions

    A personal experiment: Apache Cloudberry, the most up-to-date open-source version of Greenplum, turned into a set of extensions for PostgreSQL 19. Plans are cut into slices, rows travel between segments, and transactions use distributed snapshots and two-phase commit, while PostgreSQL itself needed only 22 small patches that add new extension points.

subscribe via RSS