PostgreSQL 17 in production: three features we liked

Postgres 17 shipped in September 2024. Six months after upgrading our larger projects, three features that entered practice.

1. JSON_TABLE

Postgres finally has standard SQL/JSON syntax to extract tables from JSON. For semi-structured data (logs, events, external API payloads), JSON_TABLE is the difference between "30 SQL lines" and "5".

SELECT * FROM JSON_TABLE(
  events_data,
  '$[*]'
  COLUMNS (
    user_id INT PATH '$.user.id',
    action TEXT PATH '$.action',
    ts TIMESTAMP PATH '$.timestamp'
  )
);

2. Much faster vacuum

Postgres 17 internally changes the vacuum process: a new memory layout for dead tuples cuts I/O and CPU. On our biggest dbs (50+ GB) nightly vacuum went from 38 minutes to 14. On write-heavy tables, noticeable.

3. Improved COPY

New ON_ERROR and LOG_VERBOSITY options let you import data with corrupted rows without stopping the whole operation. For periodic CSV imports (ERP → Postgres), valuable.

Not yet in production

• Logical replication with failover: useful on multi-region scenarios, not urgent for us.
• Incremental backup (built-in, replacing some pgBackRest pieces): test before adopting.

Migration

16 to 17 broke nothing on our 12 projects. For those still on Postgres 14 or 15 (looming EOL): move within the year. The performance and feature delta justifies the work.