AI NewsInfrastructureAnnouncement

Google Cloud clones your Postgres, tests the change, then deletes the clone

Database Center can now test a recommended change on a copy of your Cloud SQL for Postgres. It then deletes the copy at the end.

AI News

Editorial2 min read

LinkedInX

Why it mattersA database admin can test a vendor recommendation without a change window. The number comes from a copy of your own database.

Google Cloud added a preview feature to Database Center on 18 November 2026. It tests a recommended change on a clone of your Cloud SQL for Postgres. It then deletes the clone at the end. The post was written by Isabella Lubin, a Group Product Manager, and Apoorv Shrivastava, a Product Manager. The preview covers three Database Center recommendations today: a bigger machine, Enterprise Plus edition, and the data cache.

What the workflow does

The team calls it "an agentic workflow that lets you run automated, isolated synthetic benchmarks on cloned instances". It builds a test plan in five steps, per the post.

First, it writes the plan from plain-English prompts. You give the number of client connections, the number of transactions, the query mode and how long the test runs. Second, it clones the primary instance. Third, it runs a synthetic workload on the clone to get a baseline. Fourth, it applies the change to the clone. Fifth, it runs the workload again and writes a report that compares the two runs.

The clone is deleted at the end of the run. The post says the workflow "helps you verify potential performance gains before touching your production environment". No test traffic touches the real database.

What it measures

The report covers four numbers from the clone. Transaction throughput. Average latency. CPU use. Memory use. Those four cover what a Database Center recommendation would move in most cases. A bigger machine lifts throughput and cuts CPU contention. Enterprise Plus changes what the engine does under load. The data cache changes read latency and I/O.

The preview supports three recommendations today. A machine upgrade measures vCPU and memory impact on throughput and compute contention. An edition upgrade compares Enterprise and Enterprise Plus. Enabling the data cache measures its effect on read latency and I/O.

What is not in it

The post names three limits. The preview is for Cloud SQL for Postgres only. It only runs on recommendations Database Center already surfaces. There is no number for how long a run takes, or what the clone costs on top.

The four metrics come from Database Center's own view of the clone. They are not the ones your own application cares about. A query a user actually runs is not in the workload.

So the question to ask is whether that synthetic workload looks like your own traffic. If your database spends most of its time on three slow queries, the clone may not run them. Average latency can go down on the report and still miss them. The team does not say how a user picks the workload. Watch the preview page for that answer. Read your own slow-query log before the test.

Source

Testing changes, fast: Introducing agentic assessments in Database Center, Google Cloud.

This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.

Share
LinkedInX
Start a project