incus/doc/reference/instance_options.md

(instance-options)=

Instance options

Instance options are configuration options that are directly related to the instance.

See {ref}instances-configure-options for instructions on how to set the instance options.

The key/value configuration is namespaced. The following options are available:

• {ref}instance-options-misc
• {ref}instance-options-boot
• cloud-init configuration
• {ref}instance-options-limits
• {ref}instance-options-migration
• {ref}instance-options-nvidia
• {ref}instance-options-oci
• {ref}instance-options-raw
• {ref}instance-options-security
• {ref}instance-options-snapshots
• {ref}instance-options-volatile

Note that while a type is defined for each option, all values are stored as strings and should be exported over the REST API as strings (which makes it possible to support any extra values without breaking backward compatibility).

(instance-options-misc)=

Miscellaneous options

In addition to the configuration options listed in the following sections, these instance options are supported:

% Include content from ../config_options.txt

    :start-after: <!-- config group instance-miscellaneous start -->
    :end-before: <!-- config group instance-miscellaneous end -->

:type: "string"
:liveupdate: "yes (exec)"
:shortdesc: "Environment variables for the instance"

You can export key/value environment variables to the instance.
These are then set for [`incus exec`](incus_exec.md).

(instance-options-boot)=

Boot-related options

The following instance options control the boot-related behavior of the instance:

% Include content from ../config_options.txt

    :start-after: <!-- config group instance-boot start -->
    :end-before: <!-- config group instance-boot end -->

(instance-options-cloud-init)=

cloud-init configuration

The following instance options control the cloud-init configuration of the instance:

% Include content from ../config_options.txt

    :start-after: <!-- config group instance-cloud-init start -->
    :end-before: <!-- config group instance-cloud-init end -->

Support for these options depends on the image that is used and is not guaranteed.

If you specify both cloud-init.user-data and cloud-init.vendor-data, the content of both options is merged. Therefore, make sure that the cloud-init configuration you specify in those options does not contain the same keys.

(instance-options-limits)=

Resource limits

The following instance options specify resource limits for the instance:

% Include content from ../config_options.txt

    :start-after: <!-- config group instance-resource-limits start -->
    :end-before: <!-- config group instance-resource-limits end -->

:type: "string"
:liveupdate: "no"
:condition: "container"
:shortdesc: "Kernel resources per instance"

You can set kernel limits on an instance, for example, you can limit the number of open files.
See {ref}`instance-options-limits-kernel` for more information.

Memory limits in virtual machines

Incus supports both increasing and decreasing the memory allocation of virtual machines.

Increasing the memory is done through memory hot plug, effectively adding virtual memory sticks to the VM. There is a limit of 16 virtual slots for this, limiting the number of memory increases that can be done without rebooting the VM.

Decreasing memory is not done through hot remove as that has a high risk of causing guest issues. Instead the memory balloon device is used, causing memory pressure inside the guest and causing memory to be released.

This is a pretty slow process, so it is common for a memory reduction to fail to meet the requested value. When that happens, re-applying the lower value will trigger another attempt.

As each attempt will cause the effective memory available to the guest to be reduced, it should eventually succeed and lead to the guest having the desired memory limit applied.

CPU limits

You have different options to limit CPU usage:

• Set limits.cpu to restrict which CPUs the instance can see and use. See {ref}instance-options-limits-cpu for how to set this option.
• Set limits.cpu.allowance to restrict the load an instance can put on the available CPUs. This option is available only for containers. See {ref}instance-options-limits-cpu-container for how to set this option.

It is possible to set both options at the same time to restrict both which CPUs are visible to the instance and the allowed usage of those instances. However, if you use limits.cpu.allowance with a time limit, you should avoid using limits.cpu in addition, because that puts a lot of constraints on the scheduler and might lead to less efficient allocations.

The CPU limits are implemented through a mix of the cpuset and cpu cgroup controllers.

(instance-options-limits-cpu)=

CPU pinning

limits.cpu results in CPU pinning through the cpuset controller. You can specify either which CPUs or how many CPUs are visible and available to the instance:

• To specify which CPUs to use, set limits.cpu to either a set of CPUs (for example, 1,2,3) or a CPU range (for example, 0-3).

  To pin to a single CPU, use the range syntax (for example, 1-1) to differentiate it from a number of CPUs.

• If you specify a number (for example, 4) of CPUs, Incus will do dynamic load-balancing of all instances that aren't pinned to specific CPUs, trying to spread the load on the machine. Instances are re-balanced every time an instance starts or stops, as well as whenever a CPU is added to the system.

CPU limits for virtual machines

Incus supports live-updating the `limits.cpu` option.
However, for virtual machines, this only means that the respective CPUs are hotplugged.
Depending on the guest operating system, you might need to either restart the instance or complete some manual actions to bring the new CPUs online.

Incus virtual machines default to having just one vCPU allocated, which shows up as matching the host CPU vendor and type, but has a single core and no threads.

When limits.cpu is set to a single integer, Incus allocates multiple vCPUs and exposes them to the guest as full cores. Those vCPUs are not pinned to specific physical cores on the host. The number of vCPUs can be updated while the VM is running.

When limits.cpu is set to a range or comma-separated list of CPU IDs (as provided by incus info --resources), the vCPUs are pinned to those physical cores. In this scenario, Incus checks whether the CPU configuration lines up with a realistic hardware topology and if it does, it replicates that topology in the guest. When doing CPU pinning, it is not possible to change the configuration while the VM is running.

For example, if the pinning configuration includes eight threads, with each pair of thread coming from the same core and an even number of cores spread across two CPUs, the guest will show two CPUs, each with two cores and each core with two threads. The NUMA layout is similarly replicated and in this scenario, the guest would most likely end up with two NUMA nodes, one for each CPU socket.

In such an environment with multiple NUMA nodes, the memory is similarly divided across NUMA nodes and be pinned accordingly on the host and then exposed to the guest.

All this allows for very high performance operations in the guest as the guest scheduler can properly reason about sockets, cores and threads as well as consider NUMA topology when sharing memory or moving processes across NUMA nodes.

(instance-options-limits-cpu-container)=

Allowance and priority (container only)

limits.cpu.allowance drives either the CFS scheduler quotas when passed a time constraint, or the generic CPU shares mechanism when passed a percentage value:

• The time constraint (for example, 20ms/50ms) is a hard limit. For example, if you want to allow the container to use a maximum of one CPU, set limits.cpu.allowance to a value like 100ms/100ms. The value is relative to one CPU worth of time, so to restrict to two CPUs worth of time, use something like 100ms/50ms or 200ms/100ms.
• When using a percentage value, the limit is a soft limit that is applied only when under load. It is used to calculate the scheduler priority for the instance, relative to any other instance that is using the same CPU or CPUs. For example, to limit the CPU usage of the container to one CPU when under load, set limits.cpu.allowance to 100%.

limits.cpu.priority is another factor that is used to compute the scheduler priority score when a number of instances sharing a set of CPUs have the same percentage of CPU assigned to them.

(instance-options-limits-hugepages)=

Huge page limits

Incus allows to limit the number of huge pages available to a container through the limits.hugepage.[size] key.

Architectures often expose multiple huge-page sizes. The available huge-page sizes depend on the architecture.

Setting limits for huge pages is especially useful when Incus is configured to intercept the mount syscall for the hugetlbfs file system in unprivileged containers. When Incus intercepts a hugetlbfs mount syscall, it mounts the hugetlbfs file system for a container with correct uid and gid values as mount options. This makes it possible to use huge pages from unprivileged containers. However, it is recommended to limit the number of huge pages available to the container through limits.hugepages.[size] to stop the container from being able to exhaust the huge pages available to the host.

Limiting huge pages is done through the hugetlb cgroup controller, which means that the host system must expose the hugetlb controller in the legacy or unified cgroup hierarchy for these limits to apply.

(instance-options-limits-kernel)=

Kernel resource limits

For container instances, Incus exposes a generic namespaced key limits.kernel.* that can be used to set resource limits.

It is generic in the sense that Incus does not perform any validation on the resource that is specified following the limits.kernel.* prefix. Incus cannot know about all the possible resources that a given kernel supports. Instead, Incus simply passes down the corresponding resource key after the limits.kernel.* prefix and its value to the kernel. The kernel does the appropriate validation. This allows users to specify any supported limit on their system.

Some common limits are:

% Include content from ../config_options.txt

    :start-after: <!-- config group kernel-limits start -->
    :end-before: <!-- config group kernel-limits end -->

A full list of all available limits can be found in the manpages for the getrlimit(2)/setrlimit(2) system calls.

To specify a limit within the limits.kernel.* namespace, use the resource name in lowercase without the RLIMIT_ prefix. For example, RLIMIT_NOFILE should be specified as nofile.

A limit is specified as two colon-separated values that are either numeric or the word unlimited (for example, limits.kernel.nofile=1000:2000). A single value can be used as a shortcut to set both soft and hard limit to the same value (for example, limits.kernel.nofile=3000).

A resource with no explicitly configured limit will inherit its limit from the process that starts up the container. Note that this inheritance is not enforced by Incus but by the kernel.

(instance-options-migration)=

Migration options

The following instance options control the behavior if the instance is {ref}moved from one Incus server to another <move-instances>:

% Include content from ../config_options.txt

    :start-after: <!-- config group instance-migration start -->
    :end-before: <!-- config group instance-migration end -->

(instance-options-nvidia)=

NVIDIA and CUDA configuration

The following instance options specify the NVIDIA and CUDA configuration of the instance:

% Include content from ../config_options.txt

    :start-after: <!-- config group instance-nvidia start -->
    :end-before: <!-- config group instance-nvidia end -->

(instance-options-oci)=

OCI configuration

The following instance options specify the OCI configuration of the instance:

% Include content from ../config_options.txt

    :start-after: <!-- config group instance-oci start -->
    :end-before: <!-- config group instance-oci end -->

(instance-options-raw)=

Raw instance configuration overrides

The following instance options allow direct interaction with the backend features that Incus itself uses:

% Include content from ../config_options.txt

    :start-after: <!-- config group instance-raw start -->
    :end-before: <!-- config group instance-raw end -->

Setting these `raw.*` keys might break Incus in non-obvious ways.
Therefore, you should avoid setting any of these keys.

(instance-options-qemu)=

Override QEMU configuration

For VM instances, Incus configures QEMU through a configuration file that is passed to QEMU with the -readconfig command-line option. This configuration file is generated for each instance before boot. It can be found at /run/incus/<instance_name>/qemu.conf.

The default configuration works fine for Incus' most common use case: modern UEFI guests with VirtIO devices. In some situations, however, you might need to override the generated configuration. For example:

• To run an old guest OS that doesn't support UEFI.
• To specify custom virtual devices when VirtIO is not supported by the guest OS.
• To add devices that are not supported by Incus before the machines boots.
• To remove devices that conflict with the guest OS.

To override the configuration, set the raw.qemu.conf option. It supports a format similar to qemu.conf, with some additions. Since it is a multi-line configuration option, you can use it to modify multiple sections or keys.

• To replace a section or key in the generated configuration file, add a section with a different value.

  For example, use the following section to override the default virtio-gpu-pci GPU driver:

  raw.qemu.conf: |-
      [device "qemu_gpu"]
      driver = "qxl-vga"
  

• To remove a section, specify a section without any keys. For example:

  raw.qemu.conf: |-
      [device "qemu_gpu"]
  

• To remove a key, specify an empty string as the value. For example:

  raw.qemu.conf: |-
      [device "qemu_gpu"]
      driver = ""
  

• To add a new section, specify a section name that is not present in the configuration file.

The configuration file format used by QEMU allows multiple sections with the same name. Here's a piece of the configuration generated by Incus:

[global]
driver = "ICH9-LPC"
property = "disable_s3"
value = "1"

[global]
driver = "ICH9-LPC"
property = "disable_s4"
value = "0"

The first global section disabled S3(Suspend to RAM), the second global section enabled S4(suspend to disk). In order to disable S4, the second global section index needs to be specified:

raw.qemu.conf: |-
    [global][1]
    value = "1"

Section indexes start at 0 (which is the default value when not specified), so the above example would generate the following configuration:

[global]
driver = "ICH9-LPC"
property = "disable_s3"
value = "1"

[global]
driver = "ICH9-LPC"
property = "disable_s4"
value = "1"

Override QEMU runtime objects

While raw.qemu and raw.qemu.conf can be used to alter the arguments and configuration file that's passed to QEMU, a lot of devices are now added through QMP instead.

This is used by Incus for any device which may need to be re-configured at runtime, effectively anything that can be hot-plugged.

Those devices cannot be overridden through the configuration or the command line, but instead additional configuration keys are available to run QMP commands directly.

Fixed commands can be provided through the raw.qemu.early, raw.qemu.pre-start and raw.qemu.post-start configuration keys. Those take a JSON encoded list of QMP commands to run.

The hooks correspond to:

• early, run prior to any device having been added by Incus through QMP, after QEMU has started
• pre-start, run following Incus having added all its devices but prior to the VM being started
• post-start, run immediately following the VM starting up

Advanced use

For anyone needing dynamic QMP interactions, for example to retrieve the current value of some objects before modifying or generating new objects, it's also possible to attach to those same hooks using a scriptlet.

This is done through raw.qemu.scriptlet. The scriptlet must define the qemu_hook(instance, stage) function. The instance arguments is an object representing the VM, whose attributes are those of the api.Instance struct. The stage argument is the name of the hook (config, early, pre-start or post-start), with config being run before starting QEMU, and the other hooks defined above.

The following commands are exposed to that scriptlet:

• log_info will log an INFO message
• log_warn will log a WARNING message
• log_error will log an ERROR message
• run_qmp will run an arbitrary QMP command (JSON) and return its output
• run_command will run the specified command with an optional list of arguments and return its output
• get_qemu_cmdline will return the list of command-line arguments passed to QEMU
• set_qemu_cmdline will set them
• get_qemu_conf will return the QEMU configuration file as a dictionary
• set_qemu_conf will set it from a dictionary

Additionally the following alias commands (internally use run_command) are also available to simplify scripts:

• blockdev_add
• blockdev_del
• chardev_add
• chardev_change
• chardev_remove
• device_add
• device_del
• netdev_add
• netdev_del
• object_add
• object_del
• qom_get
• qom_list
• qom_set

The functions allowing to change QEMU configuration can only be run during the config hook. In parallel, the functions running QMP commands cannot be run during the config hook.

(instance-options-security)=

Security policies

The following instance options control the {ref}security policies of the instance:

% Include content from ../config_options.txt

    :start-after: <!-- config group instance-security start -->
    :end-before: <!-- config group instance-security end -->

(instance-options-snapshots)=

Snapshot scheduling and configuration

The following instance options control the creation and expiry of {ref}instance snapshots <instances-snapshots>:

% Include content from ../config_options.txt

    :start-after: <!-- config group instance-snapshots start -->
    :end-before: <!-- config group instance-snapshots end -->

(instance-options-snapshots-names)=

Automatic snapshot names

{{snapshot_pattern_detail}}

(instance-options-volatile)=

Volatile internal data

The following volatile keys are currently used internally by Incus to store internal data specific to an instance:

% Include content from ../config_options.txt

    :start-after: <!-- config group instance-volatile start -->
    :end-before: <!-- config group instance-volatile end --

Volatile keys cannot be set by the user.