---
title: Resize a server in place
description: "Add CPU or RAM to the servers of a service one server at a time, then retune the PostgreSQL settings of the service to the new size."
related:
  - /operations/replace-a-server
---

Shut down one server at a time, change its CPU or RAM, and power it on. On an
HA service, the other members continue to serve. The resize includes one
switchover.

:::danger
Never have two of three etcd members off at the same time: etcd loses
quorum, and writes can stop. etcd runs on the PostgreSQL nodes (Standard HA)
or on the DCS servers (Advanced HA).
:::

While one member is down, the service **Overview** shows the **DCS quorum at
risk** banner. The console does not prevent a power-off. While the banner
shows, do not power off, restart or resize another node.

![DCS quorum at risk banner on the service Overview](/docs/screenshots/v2026.11.7-beta/resize-in-place/dcs-quorum-at-risk.light.webp)

## Before you start

- The service **Overview** shows the service as healthy, and no operation
  runs. The topology shows one **Primary**, and each replica with a lag near
  zero.
- The **Overview** shows no **DCS quorum at risk** banner. On Advanced HA,
  the cluster on **Infrastructure → DCS clusters** is healthy.
- You have a recent [backup](/docs/v2026.11.7-beta/backups).

If you cannot confirm a prerequisite, do not start.

## HA service: rolling resize

Resize the replicas first. Then switch over to a resized replica, and resize
the old primary last.

1. **Shut down a replica**

    Connect to a replica, and shut it down cleanly:

    ```bash
    sudo shutdown -h now
    ```

    If the node held the VIP, the VIP moves, and connections through it
    close once.

2. **Resize the server**

    Wait for the server to power off. Change its CPU or RAM, then power it
    on.

3. **Check that it rejoins**

    Wait until the topology shows the node as a healthy replica with a lag of
    zero, and the **DCS quorum at risk** banner is gone. If not, do not
    resize another node. Follow
    [If a node does not join the cluster again](#if-a-node-does-not-join-the-cluster-again).

4. **Resize the other replicas**

    Do steps 1 to 3 for each other replica, one at a time.

5. **Switch over**

    On the **Overview** topology, open the menu of a resized replica and
    select **Switchover** (see
    [Failover and switchover](/docs/v2026.11.7-beta/database-services/high-availability#failover-and-switchover)).
    Connections through the VIP close once.

6. **Check the new primary**

    Confirm that the topology shows the new primary, and the old primary as a
    healthy replica. If the switchover fails, or the old primary is not a
    healthy replica, do not shut it down. Contact support.

7. **Resize the old primary**

    Do steps 1 to 3 on the old primary.

8. **Switch back (optional)**

    Switch over to the node that you prefer, and check it as in step 6.

## Advanced HA: resize DCS servers

Resize one DCS server at a time. All services that use the DCS cluster
depend on its quorum.

1. **Check the DCS cluster**

    On **Infrastructure → DCS clusters**, confirm that all three members are
    healthy.

2. **Shut down the server**

    Connect to the DCS server, and shut it down with `sudo shutdown -h now`.

3. **Resize the server**

    Wait for it to power off. Change its CPU or RAM, then power it on.

4. **Check that it rejoins**

    Wait until the DCS cluster page shows the member and the cluster healthy.
    If it does not recover, do not shut down another server, and contact
    support.

## Single-node service

The resize stops the service.

1. **Stop the applications**

    Stop the client applications.

2. **Shut down the server**

    Shut down the server with `sudo shutdown -h now`.

3. **Resize the server**

    Wait for the server to power off. Change its CPU or RAM, then power it on.

4. **Start the applications**

    Wait until the console shows the service as healthy. Then start the
    applications. If the service does not recover, contact support.

## After the resize

The PostgreSQL settings do not change. To tune them to the new size, open
**Database service admin** and select **Retune** under **Operations**.
Retune uses the license size, or the server size if the server is smaller.

Retune restarts PostgreSQL: one switchover on an HA service, a short outage
on a single node. Check the result on **Operations**. See
[Changing the license of a deployed service](/docs/v2026.11.7-beta/licensing#changing-the-license-of-a-deployed-service).

## If a node does not join the cluster again

- **HA PostgreSQL node:** follow
  [Node power loss](/docs/v2026.11.7-beta/operations/node-power-loss), from step 4 of
  [Recovery](/docs/v2026.11.7-beta/operations/node-power-loss#recovery). A rebuild deletes the
  data directory of the member.
- **Single node or DCS server:** contact support.

If two etcd members are off, power on one of them. If the cluster does not
recover, see [Site outage and recovery](/docs/v2026.11.7-beta/operations/site-outage).
