Database Migration ServiceのGeminiを使ってPostgreSQLへの移行を加速するAccelerate PostgreSQL migrations using Gemini in Database Migration Service

Google Cloud has integrated its Gemini models into Database Migration Service (DMS) to tackle what has long been the most stubborn obstacle in moving off commercial databases: converting proprietary stored procedures, triggers, and functions into code that runs on PostgreSQL. For teams weighing a shift from Oracle or SQL Server to open source PostgreSQL or the fully managed AlloyDB for PostgreSQL, the feature targets the exact phase where modernization projects tend to stall.

The significance becomes clear once you look at how these migrations usually unfold. The opening stages often move quickly. Schemas convert, tables populate, and automated pipelines can transfer terabytes of data within hours, leaving a project looking ahead of schedule. The real bottleneck sits deeper in the source system, where hundreds of stored procedures, complex triggers, and custom functions hold years of accumulated business logic. These routines commonly govern transaction validation, order processing, and custom reporting, and they are written in vendor-specific dialects such as Oracle's PL/SQL or Microsoft's T-SQL.

Those dialects do not translate cleanly into PostgreSQL's own procedural language, PL/pgSQL. Historically, rewriting them has meant slow, manual work by engineers who understand both the source and target environments, a process that is error-prone and can stretch a timeline by months. Gemini is now applied to this translation step inside DMS, automatically producing PostgreSQL-compatible versions of the proprietary routines. The stated goal is to remove the single largest source of delay and risk in this class of migration, letting the code conversion keep pace with the data movement that already runs quickly.

This move fits a broader pattern in which Google is embedding Gemini across its cloud portfolio, from coding assistants to data analytics, rather than confining generative AI to standalone chat products. It also reflects intensifying competition in database migration tooling. AWS offers its own schema conversion utilities and has folded generative assistance into services like Amazon Q, while migration vendors and consultancies have built businesses around the manual code-conversion work that AI now aims to shorten.

For context on the destination, AlloyDB is Google's PostgreSQL-compatible database, introduced in 2022 and positioned as a higher-performance managed option for workloads leaving proprietary systems. PostgreSQL itself has grown into one of the most widely adopted open source databases, valued for its extensibility and licensing freedom, which makes it a frequent target for organizations trying to reduce commercial database licensing costs. The combination of a familiar open standard and a managed service is part of what makes these migrations attractive, provided the legacy business logic can be carried across.

As with other generative AI applied to code, the output is likely to require human review before it reaches production. Translated procedures can carry subtle behavioral differences, and business-critical logic around transactions and reporting typically demands testing to confirm the converted code behaves identically to the original. Gemini appears to accelerate the first draft of that translation rather than eliminate validation entirely, so teams should still budget for verification, performance tuning, and edge-case handling. Differences in data types, error handling, and transaction semantics between the source dialect and PL/pgSQL are common areas where converted routines need adjustment.

Even with those caveats, the integration addresses a well-documented pain point. Data migration has become largely a solved, automated problem, while logic conversion has remained a manual bottleneck that could derail otherw