---
title: Scaling
category: Projects
description: Resource limits and CPU-based autoscaling for your containers.
slug: scaling
---

The project's **Scaling** tab holds two cards: resource limits and the
autoscaling policy.

![Scaling tab](/screenshots/scaling.png)

## Resource limits

Set per-container bounds that are enforced on every deployment:

- **CPU limit (cores)** — for example `1.0` for one full core.
- **Memory limit (MB)** — for example `512`. A container that exceeds it
  is killed with an OOM error and restarted.

Autoscaling requires a CPU limit, because the engine scales on CPU
utilization.

## Autoscaling policy

Click **Enable autoscaling** and configure:

- **Min replicas** — the floor of instances always running (at least 1).
- **Max replicas** — the ceiling under load.
- **CPU threshold (%)** — when average CPU utilization across replicas
  exceeds this for the evaluation window, Dequel adds a replica; when
  load falls, it removes replicas down to the floor.
- **Cooldown** — quiet period between scaling actions so the system does
  not flap.

Disable the policy any time to return to a single replica. The engine
evaluates policies continuously in the background
(`ALERT_EVAL_INTERVAL_MS` controls the platform-wide cadence).

:::note[Stateless services scale best]
Horizontal replicas do not share local files. Keep uploads and caches on
[persistent volumes](/docs/volumes) or in [managed databases](/docs/databases)
so any replica can serve any request.
:::

## Next steps

- Read [Observability](/docs/observability) to watch CPU and memory.
- Read [Alerts](/docs/alerts) to get notified on resource pressure.