Summary

Andy Pavlo’s annual database retrospective for 2025, covering PostgreSQL’s continued market dominance (Databricks paid 250M for CrunchyData), the explosion of MCP server adoption across every DBMS, the MongoDB vs. FerretDB lawsuit, new competing columnar file formats, and a comprehensive acquisition roundup.

Andy Pavlo 的 2025 年資料庫年度回顧,涵蓋 PostgreSQL 持續主導市場(Databricks 以 10 億美元收購 Neon)、所有 DBMS 爭相支援 MCP 協議、MongoDB 控告 FerretDB 的訴訟,以及大量企業收購消息。

Key Points

  • PostgreSQL dominates: every major cloud vendor now has a PostgreSQL offering; three new sharding projects (Multigres, Neki, PgDog) launched
  • MCP for databases: 2025 was the year every DBMS added MCP server support — called “the middleman who keeps the bricks counted”
  • MongoDB sued FerretDB for trademark/patent infringement; Microsoft’s DocumentDB donation to Linux Foundation complicates the situation
  • 5 new columnar file formats launched competing with Parquet; 94% of Parquet files in the wild still use v1 features from 2013
  • Redis Ltd. switched its license back after the 2024 rugpull (Pavlo called this the prior year)
  • Major acquisitions: DataStax→IBM (1B), CrunchyData→Snowflake (8B)

Insights

The MCP adoption pattern in 2025 mirrors the 2023 “every DBMS adds a vector index” trend — a Cambrian explosion driven by a single standard announcement (OpenAI adopting MCP in March 2025 after Anthropic’s Nov 2024 announcement). Pavlo’s concern about unfettered agent access to databases via MCP is prescient: enterprise DBMSs (IBM Guardium, Oracle Database Firewall) have automated query anomaly detection that open-source systems lack, creating a security asymmetry. The distributed PostgreSQL race (Multigres/Neki/PgDog) echoes failed 2010s efforts like Postgres-XC and Postgres-XL — the question is whether any of the new attempts have learned from those failures.

Connections

Raw Excerpt

MCP is the middleman who keeps the bricks counted and the cream straight, so the database and LLMs trust each other enough to do business.