Scenario

EKS specifies the maximum resources allocated to a pod by annotation specification or automatic Request and Limit calculation. You can select either method.

Directions

Annotation specification

EKS can add template annotation in the YAML file of a workload to explicitly specify pod resource specifications. For more information, see Annotation Description.

Automatic Request and Limit calculation

EKS can calculate the Request and Limit parameters set for a workload to determine the resources required for running pods. The calculation method varies depending on the pod resource type. For more information on how to automatically calculate specified resource specifications based on the Request and Limit parameters, see CPU specification calculation methods for pods and GPU specification calculation methods for pods.

• If template annotation is specified for a workload, the annotation configuration prevails and the Request and Limit parameters are not calculated.
• For more information on Request and Limit resource allocation, see the supported CPU and GPU specifications in Resource Specifications. If the set values vary greatly from the supported specifications, the allocated resources may exceed expectations, resulting in resource waste.
• Regardless of the set Request and Limit values, the final calculation result will be matched against Resource Specifications, and resources allocated to a pod will never exceed the allowed specifications.
• If Request and Limit are not set for a container in a pod, the Request and Limit values of the container are regarded as 0.
• If Request and Limit are not set for all containers in a pod, the default pod specifications are 1 CPU core and 2-GiB memory.
• Initcontainer and Container are calculated based on the following methods, and the larger value will be used.

CPU specifications calculation methods for pods

1. Calculate the total CPU and memory values of a pod.
   The total values are the greater total Request value of all containers in a pod and the greater maximum Limit value of containers in a pod, respectively.
2. Match pod resource specifications based on the following table.

   Total CPU and Memory Values	Pod Resource Selection Rules
   The total CPU and memory values are both 0.	The pod specifications are 1 CPU core and 2-GiB memory.
   Either the total CPU value or the total memory value is 0.	Match the minimum value based on the non-0 total value. For example, if the total CPU value is 0 cores, and the total memory value is 8 GiB, match the minimum CPU value in the allowed specifications with 8-GiB memory. The selected pod specifications are 1 CPU core and 8-GiB memory.
   Neither the total CPU value nor the total memory value is 0.	Match resource specifications in Resource Specifications. First, select a higher specification (specification A) with a CPU value that is equal to or similar to the total CPU value. Then, select a higher specification with a memory value that is similar to the total memory value. If the total memory value is less than the minimum memory value in the memory range of specification A, select the minimum memory value in the memory range of specification A. If the total memory value is greater than the maximum memory value in the memory range of specification A, select a higher specification (specification B) with a memory value similar to the total memory value and change the total CPU value to that of specification B. If the total memory value falls within the memory range of specification A, select the nearest greater dual-value.
   Either the total CPU value or the memory value exceeds the maximum specification.	An error occurs, and resource matching fails.

You can better understand the CPU specification calculation methods for pods from the following examples.

• Example 1

  resources:
    limits:
      cpu: "1"
      memory: 2Gi
    requests:
      cpu: "1"
      memory: 2Gi

• Result*: the selected pod specification is 1 CPU core and 2-GiB memory.
• Example 2

  ## container1
  resources:
    limits:
      cpu: "4"
      memory: 4Gi
    requests:
      cpu: "2"
      memory: 4Gi
  ## container2
  resources:
    limits:
      cpu: "1"
      memory: 2Gi
    requests:
      cpu: "1"
      memory: 2Gi

  • Total CPU value: max((2+1),max(4,1)) = 4 cores
  • Total memory value: max((4+2),max(4,2)) = 6 GiB
  • Result*: TKS does not support the pod specification of 4 CPU cores and 6-GiB memory, and 6 GiB is less than the minimum memory value in the specifications with 4 CPU cores. Therefore, adjust the minimum memory value in the specifications with 4 CPU cores. The selected pod specification is 4 CPU cores and 8-GiB memory.
    

    GPU specification calculation methods for pods

    • Typically, GPUs have the same nvidia.com/gpu parameter value as vGPUs, and the value must be an integer.
    • vGPU can be regarded as an independent GPU type. For example, 1/4*V100 indicates that one fourth of the computing power of a V100 GPU card is virtualized to a complete card. During resource allocation, one GPU card is requested, that is, the nvidia.com/GPU value is 1.

1. Calculate the total GPU value of a pod.
   The total GPU value is the total Request value of all containers in a pod.
2. Match pod resource specifications based on the following table.

   Total CPU, Memory, and GPU Values	Pod Resource Matching Rules
   The total values must comply with specification requirements, for example, 1, 2, 4, and 8.	First, select a higher specification (specification A) with a GPU value that is equal to or similar to the total GPU value. Then, calculate the CPU and memory values based on CPU specification calculation methods for pods to obtain the CPU specification (specification B). If the CPU and memory values of specification A are greater than or equal to those of specification B, select the GPU value of specification A. If the CPU and memory values of specification A are less than those of specification B, select a higher GPU specification (specification C) with the CPU and memory values similar to those of specification B. In this method, the allocated number of GPU cards are greater than that needed, and therefore should be avoided. To prevent waste, lower the requested CPU and memory values./b>
   Any total value exceeds the maximum specifications.	An error occurs, and resource matching fails.

You can better understand the GPU specification calculation methods for pods from the following examples.

• Example 1

  ## eks.tke.cloud.tencent.com/gpu-type: V100
  resources:
    limits:
      cpu: "8"
      memory: 32Gi
      nvidia.com/gpu: "1"
    requests:
      cpu: "4"
      memory: 16Gi
      nvidia.com/gpu: "1"

  • Total GPU value: 1
  2. Total CPU value: max(4,8) = 8 cores
  3. Total memory value: max(16,32) = 32 GiB
     Result: 8 cores and 32 GiB are less than the CPU and memory values (8 cores and 40 GiB) of the V100 GPU specification (one card) in Resource Specifications. The ultimately selected pod specification is 8 CPU cores, 40-GiB memory, and 1x V100.
• Example 2

  ## eks.tke.cloud.tencent.com/gpu-type: V100
  ## container1
  resources:
    limits:
      cpu: "8"
      memory: 32Gi
      nvidia.com/gpu: "1"
    requests:
      cpu: "4"
      memory: 16Gi
      nvidia.com/gpu: "1"
  ## container2
  resources:
    limits:
      cpu: "32"
      memory: 128Gi
      nvidia.com/gpu: "1"
    requests:
      cpu: "16"
      memory: 64Gi
      nvidia.com/gpu: "1"

  • Total GPU value: 1+1 = 2
  2. Total CPU value: max((4+16),max(8,32)) = 32 cores
  3. Toal memory value: max((16+64),max(32,128)) = 128 GiB
  Result: 32 cores and 128 GiB are greater than the CPU and memory values (18 cores and 80 GiB) of the V100 GPU specification (two cards) but less than the CPU and memory values (36 cores and 160 GiB) of the V100 GPU specification (four cards).
  The ultimately selected pod specification is 36 CPU cores, 160-GiB memory, and 4x V100, resulting in the waste of two GPU cards. In this case, the waste should be avoided.