ALIYUN::EHPC::AutoScaleConfig

Configures auto scaling for a specified E-HPC cluster.

Syntax

{
  "Type": "ALIYUN::EHPC::AutoScaleConfig",
  "Properties": {
    "ExtraNodesGrowRatio": Integer,
    "EnableAutoGrow": Boolean,
    "ClusterId": String,
    "ShrinkIdleTimes": Integer,
    "MaxNodesInCluster": Integer,
    "EnableAutoShrink": Boolean,
    "GrowRatio": Integer,
    "GrowIntervalInMinutes": Integer,
    "GrowTimeoutInMinutes": Integer,
    "ImageId": String,
    "ShrinkIntervalInMinutes": Integer,
    "SpotPriceLimit": Number,
    "Queues": List,
    "ExcludeNodes": String,
    "SpotStrategy": String
  }
}

Properties

Property	Type	Required	Update allowed	Description	Constraints
ExtraNodesGrowRatio	Integer	No	Yes	The percentage of extra nodes.	Example: If a job needs 100 nodes and this value is 2, 102 nodes are added. Valid values: 0 to 100. Default value: 0.
EnableAutoGrow	Boolean	No	Yes	Enables or disables auto scale-out.	Valid values: true: Enables scale-out. false (default): Disables scale-out.
ClusterId	String	Yes	No	The cluster ID.	None
ShrinkIdleTimes	Integer	No	Yes	Consecutive idle count before a node is considered for release. A node is idle when not running jobs.	Number of consecutive idle checks before scale-in. Valid values: 2 to 5. Default value: 3. Example: With ShrinkIdleTimes set to 3 and the default ShrinkIntervalInMinutes of 2 minutes, a node is released after being idle for 6+ minutes.
MaxNodesInCluster	Integer	No	Yes	Maximum compute nodes allowed in the cluster.	Valid values: 0 to 500. Default value: 100.
EnableAutoShrink	Boolean	No	Yes	Enables or disables auto scale-in.	Valid values: true: Enables scale-in. false (default): Disables scale-in.
GrowRatio	Integer	No	Yes	The scale-out ratio.	Example: If a job needs 10 nodes and this value is 50, nodes are added at 50% per round. Valid values: 1 to 100. Default value: 100.
GrowIntervalInMinutes	Integer	No	Yes	Interval between consecutive scale-out rounds.	Unit: minutes. Valid values: 2 to 10. Default value: 2. Note A scale-out task may span multiple rounds. This parameter controls the interval between rounds.
GrowTimeoutInMinutes	Integer	No	Yes	The scale-out timeout period.	Unit: minutes. Valid values: 10 to 60. Default value: 20. Nodes not running after the timeout are released.
ImageId	String	No	Yes	The image ID.	Note If you set both QueueImageId and ImageId, QueueImageId is used by default. If you set QueueImageId or ImageId, the specified image is used. If you leave both QueueImageId and ImageId empty, the image used in the last scale-out is used. If no scale-out has been performed, the image used to create the cluster is used.
ShrinkIntervalInMinutes	Integer	No	Yes	Interval between consecutive scale-in rounds.	Unit: minutes. Valid values: 2 to 10. Default value: 2.
SpotPriceLimit	Number	No	Yes	The maximum hourly price for the nodes.	The value can be accurate to three decimal places. This parameter takes effect only when SpotStrategy is set to SpotWithPriceLimit.
Queues	List	No	Yes	Queue configurations for scale-out.	Queues properties.
ExcludeNodes	String	No	Yes	Nodes excluded from auto scaling.	Separate multiple nodes with commas (,). Note Excluded nodes are never released during scale-in.
SpotStrategy	String	No	Yes	The preemption policy for compute nodes.	Valid values: NoSpot (default): The nodes are pay-as-you-go instances. SpotWithPriceLimit: The nodes are preemptible instances with a user-defined maximum hourly price. SpotAsPriceGo: The nodes are preemptible instances for which the system automatically bids based on the current market price.

Queues syntax

"Queues": [
  {
    "HostNameSuffix": String,
    "MinNodesInQueue": Integer,
    "MaxNodesInQueue": Integer,
    "EnableAutoGrow": Boolean,
    "DataDisks": List,
    "SystemDiskSize": Integer,
    "SystemDiskLevel": String,
    "QueueImageId": String,
    "MinNodesPerCycle": Integer,
    "MaxNodesPerCycle": Integer,
    "EnableAutoShrink": Boolean,
    "SystemDiskCategory": String,
    "HostNamePrefix": String,
    "SpotPriceLimit": Number,
    "InstanceTypes": List,
    "InstanceType": String,
    "QueueName": String,
    "SpotStrategy": String
  }
]

Queues properties

Property Name	Type	Required	Update allowed	Description	Constraints
HostNameSuffix	String	No	Yes	Hostname suffix for scaled-out nodes.	None
MinNodesInQueue	Integer	No	Yes	Minimum node count after scale-in.	Valid values: 0 to 50. Default value: 0.
MaxNodesInQueue	Integer	No	Yes	Maximum node count after scale-out.	Valid values: 0 to 500. Default value: 100.
EnableAutoGrow	Boolean	No	Yes	Enables or disables queue-level auto scale-out.	Valid values: true: Enables scale-out. false (default): Disables scale-out.
DataDisks	List	No	Yes	Data disk configurations.	DataDisks properties.
SystemDiskSize	Integer	No	Yes	System disk size for scaled-out nodes.	Unit: GB. Valid values: 40 to 500. Default value: 40.
SystemDiskLevel	String	No	Yes	System disk performance level for scaled-out nodes.	Valid values: PL0: A single disk can deliver up to 10,000 random read/write input/output operations per second (IOPS). PL1 (default): A single disk can deliver up to 50,000 random read/write IOPS. PL2: A single disk can deliver up to 100,000 random read/write IOPS. PL3: A single disk can deliver up to 1,000,000 random read/write IOPS.
QueueImageId	String	No	Yes	Image ID for queue scale-out.	Note If you set both QueueImageId and ImageId, QueueImageId is used by default. If you set QueueImageId or ImageId, the specified image is used. If you leave both QueueImageId and ImageId empty, the image used in the last scale-out is used. If no scale-out has been performed, the image used to create the cluster is used.
MinNodesPerCycle	Integer	No	Yes	Minimum nodes to add per scale-out round.	Valid values: 1 to 99. Default value: 1. If the required count is less than this minimum, auto scaling adjusts upward to ensure resource delivery. Note This adjustment applies to the current cycle only.
MaxNodesPerCycle	Integer	No	Yes	Maximum nodes to add per scale-out round.	Valid values: 0 to 99. Default value: 0.
EnableAutoShrink	Boolean	No	Yes	Enables or disables queue-level auto scale-in.	Valid values: true: Enables scale-in. false (default): Disables scale-in.
SystemDiskCategory	String	No	Yes	System disk type for scaled-out nodes.	Valid values: cloud_efficiency (default): Ultra Disk. cloud_ssd: Standard SSD. cloud_essd: enterprise SSD. cloud: basic disk. This disk type is discontinued.
HostNamePrefix	String	No	Yes	Hostname prefix for scaled-out nodes.	None
SpotPriceLimit	Number	No	Yes	Maximum hourly price for auto-scaled nodes.	The value can be accurate to three decimal places. This parameter takes effect only when Queues.N.SpotStrategy is set to SpotWithPriceLimit.
InstanceTypes	List	No	Yes	Instance type configurations for auto-scaled nodes.	InstanceTypes properties.
InstanceType	String	No	Yes	Instance type for auto-scaled nodes in the queue.	None
QueueName	String	No	Yes	The name of the queue.	You can set the names of N queues simultaneously.
SpotStrategy	String	No	Yes	Preemption policy for auto-scaled nodes in the queue.	Valid values: NoSpot (default): The nodes are pay-as-you-go instances. SpotWithPriceLimit: The nodes are preemptible instances with a user-defined maximum hourly price. SpotAsPriceGo: The nodes are preemptible instances for which the system automatically bids based on the current market price.

DataDisks syntax

"DataDisks": [
  {
    "DataDiskKMSKeyId": String,
    "DataDiskPerformanceLevel": String,
    "DataDiskEncrypted": Boolean,
    "DataDiskDeleteWithInstance": Boolean,
    "DataDiskSize": Integer,
    "DataDiskCategory": String
  }
]

DataDisks properties

Property Name	Type	Required	Update allowed	Description	Constraints
DataDiskKMSKeyId	String	No	Yes	KMS key ID for the data disk.	None
DataDiskPerformanceLevel	String	No	Yes	Performance level of the data disk (ESSD only).	Valid only when DataDiskCategory is set to cloud_essd. Valid values: PL0: A single disk can deliver up to 10,000 random read/write IOPS. PL1 (default): A single disk can deliver up to 50,000 random read/write IOPS. PL2: A single disk can deliver up to 100,000 random read/write IOPS. PL3: A single disk can deliver up to 1,000,000 random read/write IOPS.
DataDiskEncrypted	Boolean	No	Yes	Specifies whether to encrypt the data disk.	Valid values: true: Encrypts the data disk. false (default): Does not encrypt the data disk.
DataDiskDeleteWithInstance	Boolean	No	Yes	Specifies whether to release the data disk with the instance.	Valid values: true (default): Releases the data disk with the instance. false: Retains the data disk when the instance is released.
DataDiskSize	Integer	No	Yes	Size of the data disk.	Unit: GB. Valid values: 40 to 500. Default value: 40.
DataDiskCategory	String	No	Yes	Type of the data disk.	Valid values: cloud_efficiency (default): ultra disk. cloud_ssd: Standard SSD. cloud_essd: enterprise SSD. cloud: basic disk.

InstanceTypes syntax

"InstanceTypes": [
  {
    "ZoneId": String,
    "VSwitchId": String,
    "SpotPriceLimit": Number,
    "InstanceType": String,
    "SpotStrategy": String
  }
]

InstanceTypes properties

Property Name	Type	Required	Update allowed	Description	Constraints
ZoneId	String	Yes	No	Zone ID for auto-scaled nodes.	Takes effect during queue scale-out. N ranges from 0 to 500.
VSwitchId	String	Yes	No	vSwitch ID for auto-scaled nodes.	Takes effect during queue scale-out. N ranges from 0 to 500.
SpotPriceLimit	Number	No	Yes	Maximum hourly price for auto-scaled nodes.	The value can be accurate to three decimal places. Valid only when SpotStrategy is set to SpotWithPriceLimit. Takes effect during queue scale-out. N ranges from 0 to 500.
InstanceType	String	Yes	Yes	Instance type for auto-scaled nodes.	Takes effect during queue scale-out. N ranges from 0 to 500.
SpotStrategy	String	No	Yes	Preemption policy for auto-scaled nodes.	Valid values: NoSpot (default): The nodes are pay-as-you-go instances. SpotWithPriceLimit: The nodes are preemptible instances with a user-defined maximum hourly price. SpotAsPriceGo: The nodes are preemptible instances for which the system automatically bids based on the current market price. You can configure the bidding strategy for N compute nodes simultaneously. Takes effect during queue scale-out. Up to 500 instance types can be specified.

Return value

Fn::GetAtt

ClusterId: the cluster ID.

Examples

JSON format

{
  "ROSTemplateFormatVersion": "2015-09-01",
  "Parameters": {
    "ClusterId": {
      "Type": "String",
      "Description": "The cluster ID."
    },
    "Queues": {
      "Type": "Json",
      "Description": "",
      "MaxLength": 8
    }
  },
  "Resources": {
    "AutoScaleConfig": {
      "Type": "ALIYUN::EHPC::AutoScaleConfig",
      "Properties": {
        "ClusterId": {
          "Ref": "ClusterId"
        },
        "Queues": {
          "Ref": "Queues"
        }
      }
    }
  },
  "Outputs": {
    "ClusterId": {
      "Description": "The cluster ID.",
      "Value": {
        "Fn::GetAtt": [
          "AutoScaleConfig",
          "ClusterId"
        ]
      }
    }
  }
}

YAML format

ROSTemplateFormatVersion: '2015-09-01'
Parameters:
  ClusterId:
    Type: String
    Description: The cluster ID.
  Queues:
    Type: Json
    Description: ''
    MaxLength: 8
Resources:
  AutoScaleConfig:
    Type: ALIYUN::EHPC::AutoScaleConfig
    Properties:
      ClusterId:
        Ref: ClusterId
      Queues:
        Ref: Queues
Outputs:
  ClusterId:
    Description: The cluster ID.
    Value:
      Fn::GetAtt:
        - AutoScaleConfig
        - ClusterId