This document documents the PerformanceProfile API introduced by the Performance controller.
This document is generated from code comments on the
PerformanceProfilestruct. When contributing a change to this document please do so by changing those code comments.
- CPU
- CPUSet
- Device
- HugePage
- HugePageSize
- HugePages
- CPUfrequency
- HardwareTuning
- NUMA
- Net
- PerformanceProfile
- PerformanceProfileList
- PerformanceProfileSpec
- PerformanceProfileStatus
- RealTimeKernel
- WorkloadHints
CPU defines a set of CPU related features.
| Field | Description | Scheme | Required |
|---|---|---|---|
| reserved | Reserved defines a set of CPUs that will not be used for any container workloads initiated by kubelet. | *CPUSet | true |
| isolated | Isolated defines a set of CPUs that will be used to give to application threads the most execution time possible, which means removing as many extraneous tasks off a CPU as possible. It is important to notice the CPU manager can choose any CPU to run the workload except the reserved CPUs. In order to guarantee that your workload will run on the isolated CPU:\n 1. The union of reserved CPUs and isolated CPUs should include all online CPUs\n 2. The isolated CPUs field should be the complementary to reserved CPUs field | *CPUSet | true |
| balanceIsolated | BalanceIsolated toggles whether or not the Isolated CPU set is eligible for load balancing work loads. When this option is set to "false", the Isolated CPU set will be static, meaning workloads have to explicitly assign each thread to a specific cpu in order to work across multiple CPUs. Setting this to "true" allows workloads to be balanced across CPUs. Setting this to "false" offers the most predictable performance for guaranteed workloads, but it offloads the complexity of cpu load balancing to the application. Defaults to "true" | *bool | false |
| offlined | Offline defines a set of CPUs that will be unused and set offline | *CPUSet | false |
CPUSet defines the set of CPUs(0-3,8-11).
CPUSet is of type string.
Device defines a way to represent a network device in several options: device name, vendor ID, model ID, PCI path and MAC address
| Field | Description | Scheme | Required |
|---|---|---|---|
| interfaceName | Network device name to be matched. It uses a syntax of shell-style wildcards which are either positive or negative. | *string | false |
| vendorID | Network device vendor ID represnted as a 16 bit Hexmadecimal number. | *string | false |
| deviceID | Network device ID (model) represnted as a 16 bit hexmadecimal number. | *string | false |
HugePage defines the number of allocated huge pages of the specific size.
| Field | Description | Scheme | Required |
|---|---|---|---|
| size | Size defines huge page size, maps to the 'hugepagesz' kernel boot parameter. | HugePageSize | false |
| count | Count defines amount of huge pages, maps to the 'hugepages' kernel boot parameter. | int32 | false |
| node | Node defines the NUMA node where hugepages will be allocated, if not specified, pages will be allocated equally between NUMA nodes | *int32 | false |
HugePageSize defines size of huge pages, can be 2M or 1G.
HugePageSize is of type string.
CPUfrequency defines cpu frequencies for isolated and reserved cpus
CPUfrequency is a type of int.
HugePages defines a set of huge pages that we want to allocate at boot.
| Field | Description | Scheme | Required |
|---|---|---|---|
| defaultHugepagesSize | DefaultHugePagesSize defines huge pages default size under kernel boot parameters. | *HugePageSize | false |
| pages | Pages defines huge pages that we want to allocate at boot time. | []HugePage | false |
HardwareTuning defines cpu frequencies for isolated and reserved cpus.
| Field | Description | Scheme | Required |
|---|---|---|---|
| isolatedCpuFreq | IsolatedCpuFreq defines the maximum cpu frequency for isolated CPUs. | *CPUfrequency | true |
| reservedCpuFreq | ReservedCpuFreq defines the maximum cpu frequency for reserved CPUs. | *CPUfrequency | true |
NUMA defines parameters related to topology awareness and affinity.
| Field | Description | Scheme | Required |
|---|---|---|---|
| topologyPolicy | Name of the policy applied when TopologyManager is enabled Operator defaults to "best-effort" | *string | false |
Net defines a set of network related features
| Field | Description | Scheme | Required |
|---|---|---|---|
| userLevelNetworking | UserLevelNetworking when enabled - sets either all or specified network devices queue size to the amount of reserved CPUs. Defaults to "false". | *bool | false |
| devices | Devices contains a list of network device representations that will be set with a netqueue count equal to CPU.Reserved . If no devices are specified then the default is all devices. | []Device | false |
PerformanceProfile is the Schema for the performanceprofiles API
| Field | Description | Scheme | Required |
|---|---|---|---|
| metadata | metav1.ObjectMeta | false | |
| spec | PerformanceProfileSpec | false | |
| status | PerformanceProfileStatus | false |
PerformanceProfileList contains a list of PerformanceProfile
| Field | Description | Scheme | Required |
|---|---|---|---|
| metadata | metav1.ListMeta | false | |
| items | []PerformanceProfile | true |
PerformanceProfileSpec defines the desired state of PerformanceProfile.
| Field | Description | Scheme | Required |
|---|---|---|---|
| cpu | CPU defines a set of CPU related parameters. | *CPU | true |
| hugepages | HugePages defines a set of huge pages related parameters. It is possible to set huge pages with multiple size values at the same time. For example, hugepages can be set with 1G and 2M, both values will be set on the node by the Performance Profile Controller. It is important to notice that setting hugepages default size to 1G will remove all 2M related folders from the node and it will be impossible to configure 2M hugepages under the node. | *HugePages | false |
| hardwareTuning | HardwareTuning defines cpu frequencies for isolated and reserved cpus. It is an optional parameter and requires vendor recommendation to find suitable frequencies. The intention is to set higher frequencies for reserved cpus where platform application is running while setting isolated cpu frequencies to match vendor recommendation. | HardwareTuning | false |
| machineConfigLabel | MachineConfigLabel defines the label to add to the MachineConfigs the operator creates. It has to be used in the MachineConfigSelector of the MachineConfigPool which targets this performance profile. Defaults to "machineconfiguration.openshift.io/role=<same role as in NodeSelector label key>" | map[string]string | false |
| machineConfigPoolSelector | MachineConfigPoolSelector defines the MachineConfigPool label to use in the MachineConfigPoolSelector of resources like KubeletConfigs created by the operator. Defaults to "machineconfiguration.openshift.io/role=<same role as in NodeSelector label key>" | map[string]string | false |
| nodeSelector | NodeSelector defines the Node label to use in the NodeSelectors of resources like Tuned created by the operator. It most likely should, but does not have to match the node label in the NodeSelector of the MachineConfigPool which targets this performance profile. In the case when machineConfigLabels or machineConfigPoolSelector are not set, we are expecting a certain NodeSelector format <domain>/<role>: "" in order to be able to calculate the default values for the former mentioned fields. | map[string]string | true |
| realTimeKernel | RealTimeKernel defines a set of real time kernel related parameters. RT kernel won't be installed when not set. | *RealTimeKernel | false |
| additionalKernelArgs | Additional kernel arguments. | []string | false |
| numa | NUMA defines options related to topology aware affinities | *NUMA | false |
| net | Net defines a set of network related features | *Net | false |
| globallyDisableIrqLoadBalancing | GloballyDisableIrqLoadBalancing toggles whether IRQ load balancing will be disabled for the Isolated CPU set. When the option is set to "true" it disables IRQs load balancing for the Isolated CPU set. Setting the option to "false" allows the IRQs to be balanced across all CPUs, however the IRQs load balancing can be disabled per pod CPUs when using irq-load-balancing.crio.io/cpu-quota.crio.io annotations. Defaults to "false" | *bool | false |
| workloadHints | WorkloadHints defines hints for different types of workloads. It will allow defining exact set of tuned and kernel arguments that should be applied on top of the node. | *WorkloadHints | false |
PerformanceProfileStatus defines the observed state of PerformanceProfile.
| Field | Description | Scheme | Required |
|---|---|---|---|
| conditions | Conditions represents the latest available observations of current state. | []conditionsv1.Condition | false |
| tuned | Tuned points to the Tuned custom resource object that contains the tuning values generated by this operator. | *string | false |
| runtimeClass | RuntimeClass contains the name of the RuntimeClass resource created by the operator. | *string | false |
RealTimeKernel defines the set of parameters relevant for the real time kernel.
| Field | Description | Scheme | Required |
|---|---|---|---|
| enabled | Enabled defines if the real time kernel packages should be installed. Defaults to "false" | *bool | false |
WorkloadHints defines the set of upper level flags for different type of workloads.
| Field | Description | Scheme | Required |
|---|---|---|---|
| highPowerConsumption | HighPowerConsumption defines if the node should be configured in high power consumption mode. The flag will affect the power consumption but will improve the CPUs latency. Defaults to false. | *bool | false |
| realTime | RealTime defines if the node should be configured for the real time workload. Defaults to true. | *bool | false |
| perPodPowerManagement | PerPodPowerManagement defines if the node should be configured in per pod power management. PerPodPowerManagement and HighPowerConsumption hints can not be enabled together. Defaults to false. | *bool | false |