~postgresweekly | Issue 589 - Postgres in the time of monster hardware (7)

Postgres in the Time of Monster Hardware — You might think your M4 Max is pretty fast, but imagine having an AMD EPYC with 192 cores per socket and 10 terabytes of RAM that benchmarks at 160x faster than a Xeon-powered server from 15 years ago! Modern CPU power (not to mention faster storage) invites us to ask questions about how we scale database...

PostgreSQL 17.3, 16.7, 15.11, 14.16, and 13.19 Released — A raft of updates to all maintained lines of Postgres to tackle one security vulnerability and lots of smaller bugs. As minor updates, the upgrade process is simple and doesn’t require any dumping and reloading.

A PostgreSQL Compatibility Index to Compare Implementations — Postgres has found itself in an envious, though complex, situation of becoming a lingua franca among databases where databases that contain no Postgres code at all attempt to be “compatible” with it in various ways. But how compatible? The Postgres Compatibility Index is an attempt to test and monitor numerous aspects.

A Look at Virtual Generated Columns in Postgres 18 — Postgres 18 is gaining the ability to have ‘virtual generated columns’ where the expression for a virtual column is computed at read-time so they’re not stored on disk as all generated columns currently are.

Representing Graphs in Postgres — Postgres isn’t a graph database, but you can emulate the concepts involved. Alternatively, you could use an extension like Apache AGE if you need Cypher-like graph querying support.

Incremental Archival from Postgres to Parquet — Crunchy Data’s pg_parquet and pg_incremental extensions can be used to set up a reliable pipeline to export ranges of rows to the columnar Parquet format on S3.

Expanding pgai Vectorizer — pgai is Timescale’s suite of tools for making it easier to use AI capabilities from Postgres, including automatically creating and syncing vector embeddings for data. Now they’ve upped the ante with SQLAlchemy support and support for more embedding models via LiteLLM (more detail here).