Google’s AlloyDB is looking more and more like PostgreSQL

Google is adding new features to its fully managed database-as-a-service (DBaaS) AlloyDB in order to provide enterprises with an alternative to PostgreSQL, which has become the most popular choice when transitioning from legacy database management systems.

PostgreSQL is rising in popularity due to its open source nature, transactional and analytical capabilities, extended support for spatial data, broader SQL support, enhanced security and governance, and expanded support for programming languages.

Google released its PostgreSQL-compatible AlloyDB in May 2022 as part of a strategy to take on PostgreSQL and also the likes of Amazon Aurora and Microsoft Azure’s Database for PostgreSQL.

Now it’s adding inline filtering to AlloyDB, which Google expects enhance the performance of filtered vector search by allowing enterprises to run it directly on the database instead of post-processing on the application side.

“This kind of filtering helps ensure that these types of searches are fast, accurate, and efficient — automatically combining the best of vector indexes and traditional indexes on metadata columns to achieve better query performance,” the cloud services provider explained in a blog post.

Inline filtering can also be executed in PostgreSQL to achieve performance gains, especially with the use of WHERE clause within a SELECT statement inside a query or a database optimizer or by using any vector search extension such as pgvector, but larger datasets might result in complexities and decrease in performance, said Mukesh Ranjan, vice president at consulting firm Everest Group.

In contrast, he said, Google’s AlloyDB executes inline filtering as part of a single optimized plan rather than a separate filter pass, resulting in performance gains,

This approach is also expected to drive advantages, such as simpler queries, consistent data management, especially for AI and semantic search use cases, for developers, he added.

Google has also added new observability tooling to AlloyDB, including a new recall evaluator that helps in measurement of vector search quality.

“That means you no longer have to build your own measurement pipelines and processes for your applications to deliver good results,” the company wrote.

In addition, Google is also adding vector index distribution statistics to the available metrics, which it hopes will help enterprises with rapidly changing real-time data to achieve more stable and consistent performance.

AlloyDB’s new observability tooling, according to Ranjan, are comparable to PostgreSQL’s core capabilities but have been extended with Google’s managed services.

“You still get the fundamental Postgres observability (like pg_stat_statements) plus the advanced Google Cloud UI, deeper analytics, and potential machine learning-based tuning suggestions,” Everest’s Ranjan said.

AlloyDB appears to have kept its compatibility with PostgreSQL and many of its features (SQL syntax, concurrency, indexing, stored procedures) but adds a new architecture with the aim of achieving performance enhancements, he said.

“Essentially, Google’s approach has been: Take the Postgres framework, keep it compatible, then accelerate it with a new storage engine and Google’s infrastructure and management tools,” he said.