DeviceSubRequest
io.k8s.api.resource.v1beta2.DeviceSubRequest
source ↗
DeviceSubRequest describes a request for device provided in the claim.spec.devices.requests[].firstAvailable array. Each is typically a request for a single resource like a device, but can also ask for several identical devices.
DeviceSubRequest is similar to ExactDeviceRequest, but doesn't expose the AdminAccess field as that one is only supported when requesting a specific device.
allocationModestringAllocationMode and its related fields define how devices are allocated to satisfy this subrequest. Supported values are:- ExactCount: This request is for a specific number of devices.
- All: This subrequest is for all of the matching devices in a pool.
capacityCapacityRequirementsCapacity define resource requirements against each capacity.If this field is unset and the device supports multiple allocations, the default value will be applied to each capacity according to requestPolicy. For the capacity that has no requestPolicy, default is the full capacity value.Applies to each device allocation. If Count > 1, the request fails if there aren't enough devices that meet the requirements. If AllocationMode is set to All, the request fails if there are devices that otherwise match the request, and have this capacity, with a value >= the requested amount, but which cannot be allocated to this request.
requestsmap[string]QuantityRequests represent individual device resource requests for distinct resources, all of which must be provided by the device.This value is used as an additional filtering condition against the available capacity on the device. This is semantically equivalent to a CEL selector with `device.capacity[<domain>].<name>.compareTo(quantity(<request quantity>)) >= 0`. For example, device.capacity['test-driver.cdi.k8s.io'].counters.compareTo(quantity('2')) >= 0.When a requestPolicy is defined, the requested amount is adjusted upward to the nearest valid value based on the policy. If the requested amount cannot be adjusted to a valid value—because it exceeds what the requestPolicy allows— the device is considered ineligible for allocation.For any capacity that is not explicitly requested: - If no requestPolicy is set, the default consumed capacity is equal to the full device capacity (i.e., the whole device is claimed).- If a requestPolicy is set, the default consumed capacity is determined according to that policy.
countinteger (int64)Count is used only when the count mode is "ExactCount". Must be greater than zero. If AllocationMode is ExactCount and this field is not specified, the default is one.derivedAttributes[]DeviceDerivedAttributeDerivedAttributes defines a set of virtual attributes computed via CEL expressions for each candidate device. These virtual attributes can be referenced in `.devices.constraints` to align and match different devices (e.g., co-allocating a GPU and a NIC on the same NUMA node) even if their drivers publish different attributes. Derived attributes are not available via `device.attributes` in the CEL environment when evaluating selector expressions.Derived attributes allow you to extract, transform, or normalize topology information (such as extracting a NUMA index from a complex topology string or renaming a vendor-specific attribute) into a common virtual attribute name at scheduling time. The scheduler then evaluates these virtual attributes exactly like static attributes when matching constraints.Every derived attribute defined in this list must be referenced by at least one MatchAttribute or DistinctAttribute constraint in the `.devices.constraints` list.The maximum number of derived attributes is 32.This is an alpha field and requires enabling the DRADerivedAttributes feature gate.
expressionstring requiredExpression is a CEL expression evaluated against each candidate device. The expression must evaluate to a primitive scalar (string, integer, boolean, or semver) or a list of these scalars ([]string, []int64, []bool, []semver) to act as a virtual grouping key. Any other return type is an error and causes CEL evaluation for the device to fail.The expression's input is an object named "device", which carries the same properties as in a CELDeviceSelector.When pod scheduling encounters CEL runtime errors (such as looking up an attribute that isn't defined) for some devices, it will abort allocation and fail scheduling for the Pod. Surfacing evaluation errors immediately prevents silent topology matching failures that are extremely hard to detect. A robust expression should, for example, check for the existence of attributes before referencing them to avoid runtime evaluation errors.The expression gets evaluated after a device has passed the other selector expressions for the request in which this expression is used. This allows writing expressions that are tailored towards the specific devices being requested (for example, by assuming the device is from a certain vendor and skipping those checks).The length of the expression must be smaller or equal to 10 Ki. The cost of evaluating it is also limited based on the estimated number of logical steps; the combined cost of all derived attributes in a claim is capped by a shared CEL cost budget.namestring requiredName is the identifier for this derived attribute, used in constraints.It must be a DNS subdomain followed by a slash ("/") followed by a C identifier (e.g. "example.com/numaNode" or "derived/numaNode").If the chosen name matches an existing physical attribute from a driver, the derived attribute's expression will shadow the physical attribute, and its evaluated value will be used in constraints instead. When the goal is to define a derived attribute that is only used within the ResourceClaim and not meant to shadow an existing attribute, use a domain prefix that no DRA driver should be using (e.g. "derived/myAttribute").It is not valid to define a derived attribute that isn't used in at least one constraint.deviceClassNamestring requiredDeviceClassName references a specific DeviceClass, which can define additional configuration and selectors to be inherited by this subrequest.A class is required. Which classes are available depends on the cluster.Administrators may use this to restrict which devices may get requested by only installing classes with selectors for permitted devices. If users are free to request anything without restrictions, then administrators can create an empty DeviceClass for users to reference.namestring requiredName can be used to reference this subrequest in the list of constraints or the list of configurations for the claim. References must use the format <main request>/<subrequest>.Must be a DNS label.selectors[]DeviceSelectorSelectors define criteria which must be satisfied by a specific device in order for that device to be considered for this subrequest. All selectors must be satisfied for a device to be considered.
celCELDeviceSelectorCEL contains a CEL expression for selecting a device.
expressionstring requiredExpression is a CEL expression which evaluates a single device. It must evaluate to true when the device under consideration satisfies the desired criteria, and false when it does not. Any other result is an error and causes allocation of devices to abort.The expression's input is an object named "device", which carries the following properties:- driver (string): the name of the driver which defines this device.
- attributes (map[string]object): the device's attributes, grouped by prefix
- capacity (map[string]object): the device's capacities, grouped by prefix.
- allowMultipleAllocations (bool): the allowMultipleAllocations property of the device
tolerations[]DeviceTolerationIf specified, the request's tolerations.Tolerations for NoSchedule are required to allocate a device which has a taint with that effect. The same applies to NoExecute.In addition, should any of the allocated devices get tainted with NoExecute after allocation and that effect is not tolerated, then all pods consuming the ResourceClaim get deleted to evict them. The scheduler will not let new pods reserve the claim while it has these tainted devices. Once all pods are evicted, the claim will get deallocated.The maximum number of tolerations is 16.This is a beta field and requires enabling the DRADeviceTaints feature gate.
effectstringEffect indicates the taint effect to match. Empty means match all taint effects. When specified, allowed values are NoSchedule and NoExecute.keystringKey is the taint key that the toleration applies to. Empty means match all taint keys. If the key is empty, operator must be Exists; this combination means to match all values and all keys. Must be a label name.operatorstringOperator represents a key's relationship to the value. Valid operators are Exists and Equal. Defaults to Equal. Exists is equivalent to wildcard for value, so that a ResourceClaim can tolerate all taints of a particular category.tolerationSecondsinteger (int64)TolerationSeconds represents the period of time the toleration (which must be of effect NoExecute, otherwise this field is ignored) tolerates the taint. By default, it is not set, which means tolerate the taint forever (do not evict). Zero and negative values will be treated as 0 (evict immediately) by the system. If larger than zero, the time when the pod needs to be evicted is calculated as <time when taint was adedd> + <toleration seconds>.valuestringValue is the taint value the toleration matches to. If the operator is Exists, the value must be empty, otherwise just a regular string. Must be a label value.