> ## Documentation Index
> Fetch the complete documentation index at: https://docs.e6data.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Suspend and resume

> How clusters suspend to save cost and resume automatically when a query arrives - auto-suspend, manual suspend, auto-resume, and what's preserved.

Suspension is the primary cost-control lever for clusters. A suspended cluster has zero replicas, consumes no compute, and costs nothing. The first query after suspension resumes it automatically.

## Auto-suspend

When no queries have run for the configured idle period, the cluster suspends automatically.

| Parameter | Range | Default |
| - | - | - |
| Enabled | on / off | on |
| Idle timeout | 5–1440 minutes | 30 minutes |

Keep auto-suspend on for all non-production clusters with a short idle timeout. For always-on clusters, increase the timeout or disable it.

## Manual suspend

Suspend a cluster immediately from **Cluster → Suspend**. The operation is idempotent - suspending an already-suspended cluster succeeds silently.

## Auto-resume

When a query arrives at a suspended cluster, e6data resumes it automatically:

* **HTTP / gRPC:** the request is held until the cluster is ready (up to a 300 s timeout).
* **PostgreSQL wire:** the client receives SQLSTATE `57P03` ("cluster is being resumed, please retry"). Standard drivers (`libpq`, `psql`, recent `pgjdbc`) retry automatically.

The first query after resume waits for the cluster to come back; larger clusters (XL+) may take longer.

| Parameter | Range | Default |
| - | - | - |
| Enabled | on / off | on |
| Timeout | 30–600 s | 300 s |

You can also resume manually from **Cluster → Resume**.

## Resume failure

If a cluster fails to come back, e6data automatically **re-suspends** it and marks it "Suspended after failed resume." Check the error in the Console, fix the underlying issue, and resume again.

## What's preserved across suspend/resume

* Replica counts are stored before suspension and restored on resume.
* If **autoscaling** is enabled, the cluster resumes at the **min executor count** (not the pre-suspension count), avoiding resuming an over-scaled cluster.
* If no history is available, the cluster falls back to a safe default (1 executor).

For pausing **all** compute in a workspace at once (rather than per cluster), see [Workspace settings](/platform/workspaces/settings).

## See also

* [Cluster sizing and scaling](/query/clusters/sizing-and-scaling) - scale-to-zero and autoscaling.
* [Create and manage clusters](/query/clusters/create-and-manage)


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.