Limit provisioning to one GPU worker node in only one zone

[lionelmace]

I have the following GPU scenario where I want to create a large single GPU worker (e.g. A100) in a single VPC Zone to limit cost.

Any recommandation on how to adapt the code below which by default provisions one worker per zone?

locals {
  # list of subnets in all zones
  subnets = [
    for subnet in ibm_is_subnet.subnets :
    {
      id         = subnet.id
      zone       = subnet.zone
      cidr_block = subnet.ipv4_cidr_block
    }
  ]

  # mapping of cluster worker pool names to subnets
  cluster_vpc_subnets = {
    zone-1 = local.subnets,
    zone-2 = local.subnets,
    zone-3 = local.subnets
  }

  boot_volume_encryption_kms_config = {
    crk             = module.kp_all_inclusive.keys["${local.key_ring}.${local.boot_volume_key}"].key_id
    kms_instance_id = module.kp_all_inclusive.kms_guid
  }

  worker_pools = [
    {
      subnet_prefix                     = "zone-1"
      pool_name                         = "default"   # ibm_container_vpc_cluster automatically names default pool "default" (See https://github.com/IBM-Cloud/terraform-provider-ibm/issues/2849)
      machine_type                      = "bx2.16x64" # ODF Flavors
      workers_per_zone                  = 1
      operating_system                  = "RHCOS"
      enableAutoscaling                 = true
      minSize                           = 1
      maxSize                           = 6
      boot_volume_encryption_kms_config = local.boot_volume_encryption_kms_config
    },
    {
      subnet_prefix                     = "zone-2"
      pool_name                         = "gpu-pool"
      machine_type                      = "gx3.16x80.l4"
      workers_per_zone                  = 1
      secondary_storage                 = "600gb.10iops-tier"
      operating_system                  = "RHCOS"
      boot_volume_encryption_kms_config = local.boot_volume_encryption_kms_config
    },

[lionelmace]

I suggest the following changes marked as LMA

locals {
  # list of subnets in all zones
  subnets = [
    for subnet in ibm_is_subnet.subnets :
    {
      id         = subnet.id
      zone       = subnet.zone
      cidr_block = subnet.ipv4_cidr_block
    }
  ]

  # mapping of cluster worker pool names to subnets
  cluster_vpc_subnets = {
    zone-1 = local.subnets,
    zone-2 = local.subnets,
    zone-3 = local.subnets
  }

  # LMA BEGIN ------------------
  # Subnets for one zone for GPU
  subnets_gpu_zone_1 = [
    for s in ibm_is_subnet.subnets : {
      id         = s.id
      zone       = s.zone
      cidr_block = s.ipv4_cidr_block
    } if s.zone == "${var.region}-1"
  ]

  # GPU subnets
  cluster_vpc_subnets_merged = merge(
    local.cluster_vpc_subnets,
    { "zone-1-gpu" = local.subnets_gpu_zone_1 }
  )
  # LMA END ------------------

  boot_volume_encryption_kms_config = {
    crk             = module.kp_all_inclusive.keys["${local.key_ring}.${local.boot_volume_key}"].key_id
    kms_instance_id = module.kp_all_inclusive.kms_guid
  }

  worker_pools = [
    {
      subnet_prefix                     = "zone-1"
      pool_name                         = "default"   # ibm_container_vpc_cluster automatically names default pool "default" (See https://github.com/IBM-Cloud/terraform-provider-ibm/issues/2849)
      machine_type                      = "bx2.16x64" # ODF Flavors
      workers_per_zone                  = 1
      operating_system                  = "RHCOS"
      enableAutoscaling                 = true
      minSize                           = 1
      maxSize                           = 6
      boot_volume_encryption_kms_config = local.boot_volume_encryption_kms_config
    },
    {
      subnet_prefix                     = "zone-1-gpu"
      pool_name                         = "gpu-pool"
      machine_type                      = "gx3.16x80.l4"
      workers_per_zone                  = 1
      secondary_storage                 = "600gb.10iops-tier"
      operating_system                  = "RHCOS"
      boot_volume_encryption_kms_config = local.boot_volume_encryption_kms_config
    },
    # {
    #   subnet_prefix                     = "zone-3"
    #   pool_name                         = "zone-3"
    #   machine_type                      = "bx2.16x64"
    #   workers_per_zone                  = 1
    #   operating_system                  = "RHCOS"
    #   boot_volume_encryption_kms_config = local.boot_volume_encryption_kms_config
    # }
  ]

  worker_pools_taints = {
    all     = []
    default = []
    zone-2 = [{
      key    = "dedicated"
      value  = "zone-2"
      effect = "NoExecute"
    }]
    zone-3 = [{
      key    = "dedicated"
      value  = "zone-3"
      effect = "NoExecute"
    }]
  }
}

module "ocp_base" {
  source = "terraform-ibm-modules/base-ocp-vpc/ibm"

  cluster_name                        = var.prefix
  resource_group_id                   = module.resource_group.resource_group_id
  region                              = var.region
  force_delete_storage                = true
  vpc_id                              = ibm_is_vpc.vpc.id
  #LMA vpc_subnets                         = local.cluster_vpc_subnets
  vpc_subnets                         = local.cluster_vpc_subnets_merged
  worker_pools                        = local.worker_pools
  ocp_version                         = var.ocp_version
  tags                                = var.resource_tags
  access_tags                         = var.access_tags
  worker_pools_taints                 = local.worker_pools_taints
  ocp_entitlement                     = var.ocp_entitlement
  enable_openshift_version_upgrade    = var.enable_openshift_version_upgrade
  disable_outbound_traffic_protection = var.disable_outbound_traffic_protection

[ocofaigh]

PR: #843