gcloud compute resource-policies create workload-policy - create a Compute Engine workload resource policy

gcloud compute resource-policies create workload-policyNAME--type=TYPE[--description=DESCRIPTION][--region=REGION][--accelerator-topology=ACCELERATOR_TOPOLOGY    |--max-topology-distance=MAX_TOPOLOGY_DISTANCE][GCLOUD_WIDE_FLAG …]

NAME

Name of the resource policy to operate on.

--type=TYPE

Type of the workload policy defining the high-level intent of the cluster.TYPE must be one of:

HIGH_AVAILABILITY

For workloads that aim to be highly available. Common examples are web / MLserving, or distributed database clusters. Compute Engine spreads VMs atbest-effort to improve reliability of the distributed infrastructure.

HIGH_THROUGHPUT

For high throughput distributed workloads eg. HPC or ML training. Compute Enginecollocates VMs at best-effort to reduce network latency between VMs.

--description=DESCRIPTION

An optional, textual description for the backend.

--region=REGION

Region of the resource policy to operate on. If not specified, you might beprompted to select a region (interactive mode only).

To avoid prompting when this flag is omitted, you can set thecompute/region property:

gcloudconfigsetcompute/regionREGION

A list of regions can be fetched by running:

gcloudcomputeregionslist

To unset the property, run:

gcloudconfigunsetcompute/region

Alternatively, the region can be stored in the environment variableCLOUDSDK_COMPUTE_REGION.

At most one of these can be specified:

--accelerator-topology=ACCELERATOR_TOPOLOGY

Specifies the topology of placement and interconnection performance required tocreate a slice of VMs with interconnected accelerators.

--max-topology-distance=MAX_TOPOLOGY_DISTANCE

Specifies the topology of placement and interconnection network performance ofthe group of VMs (MIG / Multi-MIGs).MAX_TOPOLOGY_DISTANCE must be one of:

BLOCK

VMs are placed within the same block of capacity with improved latency comparedto Cluster.

CLUSTER

VMs are placed within the same cluster of capacity with improved latency betweenthem.

SUBBLOCK

Tightest collocation of VMs that provides minimized network latency. VMs areplaced within the same rack of capacity with improved latency compared to Block.

Run$gcloud help for details.