[queue] section - Amazon ParallelCluster
Services or capabilities described in Amazon Web Services documentation might vary by Region. To see the differences applicable to the China Regions, see Getting Started with Amazon Web Services in China (PDF).

[queue] section

Defines configuration settings for a single queue. [queue] sections are only supported when scheduler is set to slurm.

The format is [queue <queue-name>]. queue-name must start with a lowercase letter, contain no more than 30 characters, and only contain lowercase letters, numbers, and hyphens (-).

[queue q1] compute_resource_settings = i1,i2 placement_group = DYNAMIC enable_efa = true disable_hyperthreading = false compute_type = spot
Note

Support for the [queue] section was added in Amazon ParallelCluster version 2.9.0.

compute_resource_settings

(Required) Identifies the [compute_resource] sections containing the compute resources configurations for this queue. The section names must start with a letter, contain no more than 30 characters, and only contain letters, numbers, hyphens (-), and underscores (_).

Up to three (3) [compute_resource] sections are supported for each [queue] section.

For example, the following setting specifies that the sections that start [compute_resource cr1] and [compute_resource cr2] are used.

compute_resource_settings = cr1, cr2

Update policy: If this setting is changed, the update is not allowed.

compute_type

(Optional) Defines the type of instances to launch for this queue. This setting replaces the cluster_type setting.

Valid options are: ondemand, and spot.

The default value is ondemand.

For more information about Spot Instances, see Working with Spot Instances.

Note

Using Spot Instances requires that the AWSServiceRoleForEC2Spot service-linked role exist in your account. To create this role in your account using the Amazon CLI, run the following command:

aws iam create-service-linked-role --aws-service-name spot.amazonaws.com

For more information, see Service-linked role for Spot Instance requests in the Amazon EC2 User Guide.

The following example uses SpotInstances for the compute nodes in this queue.

compute_type = spot

Update policy: The compute fleet must be stopped for this setting to be changed for an update.

disable_hyperthreading

(Optional) Disables hyperthreading on the nodes in this queue. Not all instance types can disable hyperthreading. For a list of instance types that support disabling hyperthreading, see CPU cores and threads for each CPU core per instance type in the Amazon EC2 User Guide. If the disable_hyperthreading setting in the [cluster] section is defined, then this setting cannot be defined.

The default value is false.

disable_hyperthreading = true

Update policy: The compute fleet must be stopped for this setting to be changed for an update.

enable_efa

(Optional) If set to true, specifies that Elastic Fabric Adapter (EFA) is enabled for the nodes in this queue. To view the list of EC2 instances that support EFA, see Supported instance types in the Amazon EC2 User Guide for Linux Instances. If the enable_efa setting in the [cluster] section is defined, then this setting cannot be defined. A cluster placement group should be used to minimize latencies between instances. For more information, see placement and placement_group.

enable_efa = true

Update policy: The compute fleet must be stopped for this setting to be changed for an update.

enable_efa_gdr

(Optional) Starting with Amazon ParallelCluster version 2.11.3, this setting has no effect. Elastic Fabric Adapter (EFA) support for GPUDirect RDMA (remote direct memory access) is enabled for the compute nodes is always enabled if it's supported by the instance type.

Note

Amazon ParallelCluster version 2.10.0 through 2.11.2: If true, specifies that Elastic Fabric Adapter (EFA) GPUDirect RDMA (remote direct memory access) is enabled for the nodes in this queue. Setting this to true requires that the enable_efa setting is set to true .EFA GPUDirect RDMA is supported by the following instance types (p4d.24xlarge) on these operating systems (alinux2, centos7, ubuntu1804, or ubuntu2004). If the enable_efa_gdr setting in the [cluster] section is defined, then this setting cannot be defined. A cluster placement group should be used to minimize latencies between instances. For more information, see placement and placement_group.

The default value is false.

enable_efa_gdr = true
Note

Support for enable_efa_gdr was added in Amazon ParallelCluster version 2.10.0.

Update policy: The compute fleet must be stopped for this setting to be changed for an update.

placement_group

(Optional) If present, defines the placement group for this queue. This setting replaces the placement_group setting.

Valid options are the following values:

  • DYNAMIC

  • An existing Amazon EC2 cluster placement group name

When set to DYNAMIC, a unique placement group for this queue is created and deleted as part of the cluster stack.

For more information about placement groups, see Placement groups in the Amazon EC2 User Guide. If the same placement group is used for different instance types, it’s more likely that the request might fail due to an insufficient capacity error. For more information, see Insufficient instance capacity in the Amazon EC2 User Guide.

There is no default value.

Not all instance types support cluster placement groups. For example, t2.micro doesn't support cluster placement groups. For information about the list of instance types that support cluster placement groups, see Cluster placement group rules and limitations in the Amazon EC2 User Guide. See Placement groups and instance launch issues for tips when working with placement groups.

placement_group = DYNAMIC

Update policy: The compute fleet must be stopped for this setting to be changed for an update.