[1/2] Databases in 2025: A Year in Review by Andy Pavlo (Category Databases)
I studied the annual survey Database market from Andy Pavlo, a cool teacher and researcher of the database world from Carnegie Melon University. Below is a brief sammari highlights
PostgreSQL rules the ball PostgreSQL’s dominance continues, and it is around Postgres that much of the energy and innovation in a DBMS is concentrated. In 2025 This was confirmed by high-profile deals: Databricks acquires Neon’s PostgreSQL cloud service (~$1 billion)
- Snowflake – Crunchy Data ($250 million) Microsoft launches PostgreSQL-DBaaS (HorizonDB)
Postgres scale projects appeared horizontally (sharding): Supabase hires Vitess co-author for Multigres project PlanetScale Announces Neki – Postgres-on-steroid System for Heavy Loads Pavlo welcomes this wave of "charted Postgres," though it recalls that attempts to make a scalable Postgres have been made before.
MCP for LLM in every home If 2023It was a year of vector indexes in every engine. 2025Year of support for Anthropic MCP (Model Context Protocol) In all databases. After OpenAI support for this protocol, almost every vendor released an MCP server to its base - from OLAP systems. (ClickHouse, Snowflake et al.) nosql (MongoDB, Neo4j, Redis)and PostgreSQL-DBaaS providers have written their implementations. According to Andy Pavlo, the emergence of such a standard of interaction between AI and databases is a blessing, but it is necessary to limit and monitor these implementations. (For example, only read-only queries, result limits and timeouts)Sooner or later, the facade will happen.
MongoDB vs. FerretDB – Judgment Day? For the first time in the history of the DBMS industry, one company sued another for imitating its API. MongoDB, Inc. accuses startup FerretDB (A proxy that translates MongoDB queries into PostgreSQL) infringement of copyrights, patents and licenses. Andy Pavlo compares the lawsuit to Oracle vs Google’s Java API process, where the court sided with free implementation. The outcome of the case is unclear, but there is a piquant nuance: FerretDB was originally called "MangoDB". It is interesting that in parallel with Microsoft 2025 Open source of its own Mongo-compatible DocumentDB database (Transferring the project to the Linux Foundation) That is, in fact, doing what Mongo accuses FerretDB of doing.
The War of Data Formats Almost. 15 For years, the Parquet format reigned in column vaults, but in 2025 Five new open source formats challenged him. (A few more have appeared in 2024-). Among them are projects from academics (FastLanes, F3, AnyBlox)start-ups (Vortex by SpiralDB) giant (Microsoft Amudai was quietly shut down by the end of the year.). The Parquet community stepped up and set about upgrading the standard. Andy Pavlo notes that the main problem with Parquet is not in the format itself, but in the fragmentation of the implementation: too many parsers / writers in different languages supporting different subset of features. As a result, 94% of files in the wild are still created with v1 features 2013 years. No one will rewrite petabytes of old parquet files, and it is impossible to be sure that any system will read the latest file – suddenly it will not have enough support for specific options. New formats try to solve the compatibility problem in different ways. For example, the F3 format (Developed with the participation of Pavlo) embeds its own decoders into each file: both in the form of native modules and in the form of WASM code lying directly in the file. The idea is that the database can read a column of even a new, unfamiliar format by executing an attached WASM decoder. Others (AnyBlox) Create a single WASM for the entire file. It is not yet clear whose approach will win - Parquet is ubiquitous and the inertia is enormous. The next battle will be to support GPUs in formats, Pavlo predicts, but one thing is clear: the battle for efficient data storage has entered a new phase.
In continuation I'll tell you about the other items.
#Database #Architecture #Management #DistributedSystems #Software #Engineering #Future