# Amazon RDS Proxy

> RDS Proxy bills per vCPU-hour of the database instance it fronts, with a 2-vCPU minimum per target, for every hour the proxy exists. ZopNight discovers proxies on the 6-hour cycle and flags hygiene problems, chiefly proxies still attached to idle or decommissioned databases, where the surcharge buys nothing.

Source: https://zop.dev/integrations/aws/rds-proxy
Updated: 2026-08-19

---

RDS Proxy is a managed connection pooler that sits between applications and RDS databases. It bills per vCPU-hour of the target database capacity, so proxies attached to idle or over-provisioned databases add a steady surcharge.

## A percentage of the database, by the hour

RDS Proxy's meter is unusual: it bills per vCPU-hour of the database instance it fronts, with a 2-vCPU minimum per target, regardless of how many connections actually flow through it. The proxy is priced as a fraction of the database's size, not of its own work. Front a large instance and the proxy charge scales with it, even if the connection pool idles. The meter runs for every hour the proxy exists.

## When the surcharge is worth it

Connection pooling earns its cost where connection churn is the problem: serverless and Lambda-heavy applications that open and drop connections faster than a database enjoys, or failover scenarios where the proxy's connection preservation shortens outages. A steady long-connection application gets little from it, since the proxy just adds its vCPU-hours to the bill.

## The stale-proxy pattern

Because the proxy is a separate resource, it survives changes around it. Proxies still attached to databases that no longer serve traffic. Proxies fronting over-provisioned instances, where the vCPU-based meter silently inherits the oversizing; right-size the database and the proxy charge falls with it. And proxies created during a Lambda experiment that ended, pooling connections nobody opens.

## Hygiene-driven coverage

Proxies are discovered on the 6-hour cycle, with per-proxy cost from Cost Explorer or CUR 2.0. Recommendations are hygiene-focused: proxies pointing at idle databases are the headline finding, since every vCPU-hour they bill buys pooling for connections that do not exist. There is no stop state; an unneeded proxy is simply deleted.

## Proxies in the console

RDS console, then Proxies in the left navigation. Each proxy row names its target database. Cross-reference that target's connection metrics, and a proxy on a flatlined database identifies itself.