---
title: "EZGH Optimized Images"
description: "Images with the EZGH guest agents already installed, what each agent does, the bot permissions that turn them on, and what they cost."
---

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

# EZGH Optimized Images

An EZGH Optimized image is a stock [image](/compute/images) with EZGH's guest agents already
installed and enabled. `linux/ezgh-ubuntu:24.04` is the same Ubuntu as `linux/ubuntu:24.04`, with
the same login user and nothing else changed.

| Image | Alias | Built from | Login user |
| --- | --- | --- | --- |
| Ubuntu 26.04 LTS (EZGH Optimized) | `linux/ezgh-ubuntu:26.04` | `linux/ubuntu:26.04` | `ubuntu` |
| Ubuntu 24.04 LTS (EZGH Optimized) | `linux/ezgh-ubuntu:24.04` | `linux/ubuntu:24.04` | `ubuntu` |
| Debian 13 (EZGH Optimized) | `linux/ezgh-debian:13` | `linux/debian:13` | `debian` |
| Debian 12 (EZGH Optimized) | `linux/ezgh-debian:12` | `linux/debian:12` | `debian` |
| Rocky Linux 10 (EZGH Optimized) | `linux/ezgh-rocky:10` | `linux/rocky:10` | `rocky` |
| Rocky Linux 9 (EZGH Optimized) | `linux/ezgh-rocky:9` | `linux/rocky:9` | `rocky` |

There's no optimized Alpine image: the agents need systemd. All are `amd64`.

Launch one by its alias, like any image:

```sh
ezgh compute instances launch web-1 --instance-type <type> --root-volume-type <volume type> \
  --image linux/ezgh-ubuntu:24.04 --ssh-key laptop --ipv4 --wait
```

The image carries no credentials. The agents start idle and do nothing until you give them access,
as below.

## What's on the image

| Component | What it does | Images |
| --- | --- | --- |
| `qemu-guest-agent` | For the platform's own use, such as shutting the instance down cleanly; nothing to configure | All |
| `ezgh-fleets-agent` | Registers the instance as a [Fleets](/fleets) node, for shell sessions and run command without SSH keys | All |
| `ezgh-obs-agent-linux` | Sends the guest's own metrics and journald logs to [Observability](/observability) | All |
| `ezgh-vpc-addresses` | Configures the instance's secondary private addresses from the metadata service | Ubuntu and Debian |

### Fleets agent

The agent is configured to take keys for the instance's [bot](/compute/instance-bots) from the
metadata service. Attach a bot with the Fleets node permissions, `fleets.nodes.register` and
`fleets.nodes.connect`, and the instance registers as a node, linked to the instance and named
after it. [Install the agent](/fleets/install-the-agent#create-a-node-bot) has the exact policy and
why a node bot should have nothing else. Until a bot is attached, the agent checks again every few
minutes, so attaching one later is enough; there's nothing to install or run on the instance.

### Observability agent

The agent is installed and enabled, set to send CPU, memory, disk, filesystem and network metrics
every 30 seconds and the system journal, to the project's `production` environment. It sends with
keys for the instance's [bot](/compute/instance-bots), so no key is stored on the disk.

Attach a bot with the `ObsIngestAccess` managed policy:

```sh
ezgh iam bots create instance-telemetry
ezgh iam policies attach ObsIngestAccess --bot instance-telemetry
ezgh compute instances update web-1 --bot bot_...
```

Until a bot is attached, the agent sends nothing and checks again every few minutes, so attaching
one later is enough. While the bot isn't allowed to send, the agent drops what it collects instead
of storing it. During a network outage it stores data on the disk, within limits, and sends it
afterwards. The instance needs a route to `obs-ingest.ezghcloud.com` over IPv6 or IPv4.

Instances launched before this was available keep the agent they have, set up with an
[ingest key](/observability/ingest-keys).

> **Caution**
>
> **What the agent sends is metered like any data you send to Observability**, and counts against
> [its limits](/observability/limits). This is unlike the platform's own
> [instance metrics](/compute/monitoring), which are free and need no agent.

To change what it collects, edit `/etc/ezgh-obs-agent-linux/config.yaml` and run
`sudo systemctl restart ezgh-obs-agent-linux`.

### VPC addresses hook

On Ubuntu and Debian images, the [`ezgh-vpc-addresses`](/private-networks/vpcs#configure-secondary-addresses-in-the-guest)
hook is already installed and enabled. It configures the
[secondary private addresses](/private-networks/vpcs#secondary-private-addresses) each private
interface owns, as reported by the metadata service. On an instance without a private interface it
does nothing. The instance's boot-time install still checks it at every boot and keeps it at the
version EZGH Compute pins; the copy on the image only saves the download.

Rocky images don't carry the hook. It isn't supported on Rocky, with or without an optimized image.

### Turn a component off

Each is an ordinary systemd unit. To turn one off on an instance:

```sh
sudo systemctl disable --now ezgh-fleets-agent
sudo systemctl disable --now ezgh-obs-agent-linux
sudo systemctl disable --now ezgh-vpc-addresses.timer ezgh-vpc-addresses.service
```

## How the images are updated

Each optimized image is rebuilt whenever the stock image it's built from moves to a new release, and
whenever what goes on it changes. Every new image is booted and checked before its alias moves to
it. As with any alias, a launch uses the image the alias names at that moment, and existing
instances aren't changed.

`ezgh compute images get linux/ezgh-ubuntu:24.04` shows the image the alias names now.
`ezgh compute images get <img_…> -o json` shows where that image came from in its `provenance`:

- `kind` is `built`, rather than the stock images' `upstream`.
- `sourceUrl` is the stock image's source, the distribution's cloud image.
- `upstreamSerial` is `<stock serial>-<build revision>`: the stock image it was built from, and the
  revision of what was installed on it.

Source: https://docs.ezghcloud.com/compute/optimized-images/index.mdx
