open-nomad/website/content/docs/drivers/qemu.mdx
hc-github-team-nomad-core c9ef870f96
backport of commit a9eecb457cef34ee856681a51af94fe0f6db4b21 (#17947)
This pull request was automerged via backport-assistant
2023-07-14 04:23:26 -05:00

185 lines
6.1 KiB
Plaintext

---
layout: docs
page_title: 'Drivers: QEMU'
description: The QEMU task driver is used to run virtual machines using QEMU/KVM.
---
# QEMU Driver
Name: `qemu`
The `qemu` driver provides a generic virtual machine runner. QEMU can utilize
the KVM kernel module to utilize hardware virtualization features and provide
great performance. Currently the `qemu` driver can map a set of ports from the
host machine to the guest virtual machine, and provides configuration for
resource allocation.
The `qemu` driver can execute any regular `qemu` image (e.g. `qcow`, `img`,
`iso`), and is currently invoked with `qemu-system-x86_64`.
The driver requires the image to be accessible from the Nomad client via the
[`artifact` downloader](/nomad/docs/job-specification/artifact).
## Task Configuration
```hcl
task "webservice" {
driver = "qemu"
config {
image_path = "/path/to/my/linux.img"
accelerator = "kvm"
graceful_shutdown = true
args = ["-nodefaults", "-nodefconfig"]
}
}
```
The `qemu` driver supports the following configuration in the job spec:
- `image_path` - The path to the downloaded image. In most cases this will just
be the name of the image. However, if the supplied artifact is an archive that
contains the image in a subfolder, the path will need to be the relative path
(`subdir/from_archive/my.img`).
- `drive_interface` - (Optional) This option defines on which type of interface
the drive is connected. Available types are: `ide`, `scsi`, `sd`, `mtd`,
`floppy`, `pflash`, `virtio` and `none`. Default is `ide`.
- `accelerator` - (Optional) The type of accelerator to use in the invocation.
If the host machine has `qemu` installed with KVM support, users can specify
`kvm` for the `accelerator`. Default is `tcg`.
- `graceful_shutdown` `(bool: false)` - Using the [qemu
monitor](https://en.wikibooks.org/wiki/QEMU/Monitor), send an ACPI shutdown
signal to virtual machines rather than simply terminating them. This emulates
a physical power button press, and gives instances a chance to shut down
cleanly. If the VM is still running after `kill_timeout`, it will be
forcefully terminated. This feature uses a Unix socket that is placed within
the task directory and operating systems may impose a limit on how long these
paths can be. This feature is currently not supported on Windows.
- `guest_agent` `(bool: false)` - Enable support for the [QEMU Guest
Agent](https://wiki.qemu.org/Features/GuestAgent) for this virtual machine.
This will add the necessary virtual hardware and create a `qa.sock` file in
the task's working directory for interacting with the agent. The QEMU Guest
Agent must be running in the guest VM. This feature is currently not
supported on Windows.
- `port_map` - (Optional) A key-value map of port labels.
```hcl
config {
# Forward the host port with the label "db" to the guest VM's port 6539.
port_map {
db = 6539
}
}
```
- `args` - (Optional) A list of strings that is passed to QEMU as command line
options.
## Examples
A simple config block to run a `qemu` image:
```
task "virtual" {
driver = "qemu"
config {
image_path = "local/linux.img"
accelerator = "kvm"
args = ["-nodefaults", "-nodefconfig"]
}
# Specifying an artifact is required with the "qemu"
# driver. This is the # mechanism to ship the image to be run.
artifact {
source = "https://internal.file.server/linux.img"
options {
checksum = "md5:123445555555555"
}
}
```
## Capabilities
The `qemu` driver implements the following [capabilities](/nomad/docs/concepts/plugins/task-drivers#capabilities-capabilities-error).
| Feature | Implementation |
| -------------------- | -------------- |
| `nomad alloc signal` | false |
| `nomad alloc exec` | false |
| filesystem isolation | image |
| network isolation | none |
| volume mounting | none |
## Client Requirements
The `qemu` driver requires QEMU to be installed and in your system's `$PATH`.
The task must also specify at least one artifact to download, as this is the only
way to retrieve the image being run.
## Client Attributes
The `qemu` driver will set the following client attributes:
- `driver.qemu` - Set to `1` if QEMU is found on the host node. Nomad determines
this by executing `qemu-system-x86_64 -version` on the host and parsing the output
- `driver.qemu.version` - Version of `qemu-system-x86_64`, ex: `2.4.0`
Here is an example of using these properties in a job file:
```hcl
job "docs" {
# Only run this job where the qemu version is higher than 1.2.3.
constraint {
attribute = "${driver.qemu.version}"
operator = ">"
value = "1.2.3"
}
}
```
## Plugin Options
```hcl
plugin "qemu" {
config {
image_paths = ["/mnt/image/paths"]
args_allowlist = ["-drive", "-usbdevice"]
}
}
```
- `image_paths` (`[]string`: `[]`) - Specifies the host paths the QEMU
driver is allowed to load images from.
- `args_allowlist` (`[]string`: `[]`) - Specifies the command line
flags that the [`args`] option is permitted to pass to QEMU. If
unset, a job submitter can pass any command line flag into QEMU,
including flags that provide the VM with access to host devices such
as USB drives. Refer to the [QEMU documentation] for the available
flags.
## Resource Isolation
Nomad uses QEMU to provide full software virtualization for virtual machine
workloads. Nomad can use QEMU KVM's hardware-assisted virtualization to deliver
better performance.
Virtualization provides the highest level of isolation for workloads that
require additional security, and resource use is constrained by the QEMU
hypervisor rather than the host kernel. VM network traffic still flows through
the host's interface(s).
Note that the strong isolation provided by virtualization only applies
to the workload once the VM is started. Operators should use the
`args_allowlist` option to prevent job submitters from accessing
devices and resources they are not allowed to access.
[`args`]: /nomad/docs/drivers/qemu#args
[QEMU documentation]: https://www.qemu.org/docs/master/system/invocation.html