Capacity — Growth Projections
Linear regression on storage, connections, query rate. Forecasts when you'll hit the next limit.
The Capacity page projects three things based on your last 30 days of data: storage growth, connection ceiling, and query rate.
Storage
Per-database growth rate over 30 days, projected linearly forward. Tells you "at the current rate, you'll hit 1TB in 47 days." Includes index and TOAST growth as separate components, so you can see whether you're growing in heap (data), indexes (over-indexing), or TOAST (large rows getting toasted).
Connection ceiling
Peak connection count over 30 days against max_connections. Projects when you'll start regularly hitting the cap. The action is usually pooler tuning rather than raising max_connections directly — adding a PgBouncer or raising its default_pool_size is cheaper than scaling Postgres.
QPS / TPS
Query rate trend. Useful for arguing for compute upgrades before the curve hits CPU saturation, which in practice is around 70% sustained CPU for read-heavy workloads. (Write-heavy clusters saturate earlier because of WAL.) The "argue for the upgrade" framing matters: if you're going to need a bigger instance in three months, you'd rather know that with three months' lead time than three days'.
How the projections are made
30-day daily-bucket history from the analytical store, run through linear regression with R² as a confidence indicator. R² below 0.7 means the data is too noisy for a confident projection — usually a workload that's bursty rather than growing. We surface the R² so you can tell which projections to trust. Outliers tied to deploys or migrations get excluded if you've marked Annotations.
Recommendations
For each projection that crosses a meaningful threshold, the page shows days until threshold, the recommended action (vacuum a bloated table, add a pooler, vertical scale to the next instance class), and a cost estimate where applicable — for RDS that's the instance class delta, for self-hosted it's "talk to your infra team."