---
title: External DCS
description: "Run etcd on a separate DCS cluster for Advanced HA services, check the health of the cluster, redeploy it, and replace a member server."
search:
  keywords: ["etcd", "DCS", "consensus"]
---

Each High-Availability service keeps its leader in etcd, the DCS (DCS: Distributed configuration store. etcd keeps the leader of each HA service.).
**Standard HA** runs etcd on its 3 PostgreSQL servers. **Advanced HA** uses a
DCS cluster: 3 servers that run only etcd. More than one service can use one
DCS cluster.

## Risks of colocated etcd

On **Standard HA**, etcd shares the disk and the CPU with PostgreSQL. A heavy
load or a full disk can cause a failover with no hardware fault (see
[etcd disk full](/docs/operations/dcs-disk-full)).

When the console manages 2 or more etcd clusters, the wizard recommends
**Advanced HA**. Each Standard HA service and each DCS cluster counts as one.

:::caution
You cannot move a service to a DCS cluster after the deploy. Select
**Advanced HA** at create time.
:::

## Deploy on an external DCS

1. **Enroll three servers**

    Use a different failure domain for each server (see
    [Placement rules](/docs/reference-architectures#placement-rules)), and an SSD
    for `/var/lib/etcd`.

2. **Create the DCS cluster**

    Go to **Infrastructure → DCS clusters** and select **Create DCS
    cluster**. Enter a name and select the three servers. **Run checks**
    warns below 1 vCPU, 1 GB memory or 2 GB disk, and when port `2379` or
    `2380` is in use. Warnings do not block the create.

3. **Deploy the service**

    In the create wizard, select **High-Availability**, then **Advanced HA**.
    In the **etcd configuration** step, select the **DCS cluster**. You can
    also create the DCS cluster from this step.

You cannot delete a DCS cluster while a service uses it. A delete removes
etcd and its data from the servers. The servers stay enrolled.

## Check cluster health

On **Infrastructure → DCS clusters**, select the cluster name. Select **Run
health check** to check again.

![The DCS cluster detail page with one card for each etcd member](/docs/screenshots/external-dcs/dcs-cluster-detail.light.webp)

| Status | Meaning |
| --- | --- |
| **Healthy** | All members are healthy. |
| **Degraded** | At least one member is down. Quorum holds. |
| **Quorum lost** | Fewer than 2 of 3 members are healthy. |
| **Quorum not verified** | The console could not check a member. The member shows **Not verified**. |
| **Console cannot reach the mesh** | The console reaches no server. Check the network of the console first. |

**Quorum at risk** shows when one more member failure would lose quorum.
Bring the down member back before you stop another member. Without quorum,
the services cannot fail over.

## Redeploy the DCS cluster

A console upgrade can change the etcd settings. To apply them, select
**Redeploy** on the cluster page. You need the admin role.

Each member with changed settings restarts, one at a time. If a member does
not come back healthy, the redeploy stops: fix it, then select **Redeploy**
again. A member that is down, or a service that is not healthy, blocks the
redeploy.

## Replace a member server

1. **Release the server**

    Open the menu of the member and select **Replace server**. The console
    deletes the etcd data of the server. The server stays enrolled.

2. **Fill the empty slot**

    On the empty slot, select **Fill with a server**, and select a free
    server or a new one. The rebuild starts.

See [Replace a server](/docs/operations/replace-a-server#dcs-member).
