PostgreSQL 17

Original Article ↗ Hacker News Discussion ↗

Overall reception

  • Strongly positive response; many call PostgreSQL 17 a “gift that keeps on giving” and an “amazing release.”
  • Users highlight the breadth of changes as unusually large, spanning performance, JSON, vacuum, backups, and QoL features.

Vacuum, storage, and performance

  • New vacuum memory structure (up to ~20x less memory) is welcomed as a substantial improvement.
  • Some argue vacuum remains a bottleneck in high-churn workloads (I/O and CPU, not just memory), sometimes requiring throttling writes.
  • Others note ongoing work (and third‑party projects like OrioleDB, earlier zheap) aiming at more fundamental storage changes.

JSON and SQL/JSON features

  • SQL/JSON additions (JSON_TABLE, JSON_QUERY options, json_scalar, json_serialize) are seen as major usability gains.
  • People already storing raw JSON responses and flattening via views feel validated; JSON_TABLE should simplify multi-level parsing compared to previous JSONB_TO_RECORDSET patterns.
  • There is interest in cross‑DB SQL/JSON compatibility (e.g., with SQLite).

FDW, ETL, and data ingestion

  • Foreign Data Wrappers (FDWs) are described as an “untapped gem” and used as simple ETL: define foreign tables, then materialize into local tables via schedulers (e.g., pg_cron).
  • Alternatives like DuckDB (including its Postgres connector) are praised for flexible CSV/JSON/Parquet ingestion and transformation.
  • Credential management for FDWs in cloud setups is discussed (secrets managers, platform “vaults”), but no universal pattern emerges.

Backups, replication, and failover

  • Incremental backups in pg_basebackup and pg_combinebackup are seen as much‑needed.
  • Logical replication slot failover support is called out as unblocking production use for some.

Upgrades and tooling

  • Multiple commenters wish for “seamless” major version upgrades (just changing container tags) without manual pg_upgrade and dual binaries.
  • Third‑party tools (e.g., pgautoupgrade, Bytebase) are discussed as partial answers, but some view this as something core Postgres should solve.

Comparisons and missing features

  • PostgreSQL is contrasted favorably with MySQL on performance evolution and with Oracle on cost; SQL Server is praised for BI stack and declarative DB projects.
  • Desired future features include: UUIDv7, incremental view maintenance in core, system‑versioned/bi‑temporal tables, Parquet import/export, and more vacuum/optimizer improvements.