Worked on ml- prefixed tests

Author: simrankaurb

tools/cloud-build/daily-tests/tests/ml-a3-highgpu-slurm-cluster.yml

@@ -17,8 +17,8 @@
 # region, zone, nfs_ip, remote_mount_homefs, must be defined in build file
 # with --extra-vars flag!
 test_name: a3h-cluster
-deployment_name: a3hc-{{ build }}
-slurm_cluster_name: "a3hc{{ build[0:4] }}"
+deployment_name: g{{ build[0:3] }}{{ build[-3:] }}-a3hc
+slurm_cluster_name: "g{{ build[0:3] }}{{ build[-3:] }}a3h"
 workspace: /workspace
 blueprint_yaml: "{{ workspace }}/examples/machine-learning/a3-highgpu-8g/ml-slurm-a3-2-cluster.yaml"
 login_node: "{{ slurm_cluster_name }}-login-*"

tools/cloud-build/daily-tests/tests/ml-a3-highgpu-slurm-image.yml

@@ -15,16 +15,20 @@
 ---
 
 test_name: a3h-image
-deployment_name: a3himg{{ build }}
+deployment_name: f{{ build[0:3] }}{{ build[-3:] }}a3himg
+slurm_cluster_name: "i{{ build[0:3] }}{{ build[-3:] }}a3h"
 workspace: /workspace
 blueprint_yaml: "{{ workspace }}/examples/machine-learning/a3-highgpu-8g/ml-slurm-a3-1-image.yaml"
 packer_group_name: slurm-build
+network: "{{ deployment_name }}-net-0"
+sub_network: "{{ deployment_name }}-sub-net-0"
 packer_module_id: slurm-image
 delete_image: false
 cli_deployment_vars:
-  network_name_system: default
-  subnetwork_name_system: default
+  network_name_system: "{{ network }}"
+  subnetwork_name_system: "{{ sub_network }}"
   region: us-west1
   zone: us-west1-a
   source_image_project_id: deeplearning-platform
   source_image: dlvm-tcpd-cu120-718492384-ubuntu-2004-py310
+  slurm_cluster_name: "{{ slurm_cluster_name }}"

tools/cloud-build/daily-tests/tests/ml-a3-megagpu-slurm-ubuntu.yml

@@ -16,8 +16,8 @@
 
 # region, zone must be defined in build file with --extra-vars flag!
 test_name: a3m-slurm
-deployment_name: a3m-slurm-{{ build }}
-slurm_cluster_name: "a3m{{ build[0:4] }}"
+deployment_name: e{{ build[0:3] }}{{ build[-3:] }}-a3m-slurm
+slurm_cluster_name: "e{{ build[0:3] }}{{ build[-3:] }}a3m"
 workspace: /workspace
 blueprint_yaml: "{{ workspace }}/examples/machine-learning/a3-megagpu-8g/a3mega-slurm-blueprint.yaml"
 login_node: "{{ slurm_cluster_name }}-login-*"

tools/cloud-build/daily-tests/tests/ml-a3-ultragpu-jbvms.yml

@@ -16,13 +16,13 @@
 
 # region, zone must be defined in build file with --extra-vars flag!
 test_name: a3u-jbvms
-deployment_name: a3u-jbvms-{{ build }}
+deployment_name: "a{{ build[0:3] }}{{ build[-3:] }}-a3u-jbvms"
 hostname_prefix: "{{ deployment_name }}-beowulf"
 workspace: /workspace
 blueprint_yaml: "{{ workspace }}/examples/machine-learning/a3-ultragpu-8g/a3ultra-vm.yaml"
 region: europe-west1
 zone: europe-west1-b
-network: "{{ test_name }}-net-0"
+network: "{{ deployment_name }}-net-0"
 remote_node: "{{ hostname_prefix }}-0"
 post_deploy_tests:
 - test-validation/test-mounts.yml
@@ -36,4 +36,4 @@ cli_deployment_vars:
   disk_size_gb: 200
   a3u_reservation_name: hpc-exr-2
   a3u_provisioning_model: RESERVATION_BOUND
-  base_network_name: "{{ test_name }}"
+  base_network_name: "{{ deployment_name }}"

tools/cloud-build/daily-tests/tests/ml-a3-ultragpu-slurm.yml

@@ -16,15 +16,15 @@
 
 # region, zone must be defined in build file with --extra-vars flag!
 test_name: a3u-slurm
-deployment_name: a3u-slurm-{{ build }}
-slurm_cluster_name: "a3u{{ build[0:4] }}"
+deployment_name: b{{ build[0:3] }}{{ build[-3:] }}-a3u-slurm
+slurm_cluster_name: "b{{ build[0:3] }}{{ build[-3:] }}a3u"
 workspace: /workspace
 blueprint_yaml: "{{ workspace }}/examples/machine-learning/a3-ultragpu-8g/a3ultra-slurm-blueprint.yaml"
 login_node: "{{ slurm_cluster_name }}-slurm-login-*"
 controller_node: "{{ slurm_cluster_name }}-controller"
 region: europe-west1
 zone: europe-west1-b
-network: "{{ test_name }}-net-0"
+network: "{{ deployment_name }}-net-0"
 post_deploy_tests:
 - test-validation/test-mounts.yml
 - test-validation/test-partitions.yml
@@ -49,4 +49,4 @@ cli_deployment_vars:
   disk_size_gb: 200
   a3u_cluster_size: 2
   a3u_reservation_name: hpc-exr-2
-  base_network_name: "{{ test_name }}"
+  base_network_name: "{{ deployment_name }}"

tools/cloud-build/daily-tests/tests/ml-a4-highgpu-slurm-flex.yml

@@ -16,15 +16,15 @@
 
 # region, zone must be defined in build file with --extra-vars flag!
 test_name: a4h-slurm-flex
-deployment_name: a4h-slurm-flex-{{ build }}
-slurm_cluster_name: "a4hf{{ build[0:4] }}"
+deployment_name: c{{ build[0:3] }}{{ build[-3:] }}-a4h-slurm-flex
+slurm_cluster_name: "c{{ build[0:3] }}{{ build[-3:] }}a4h"
 workspace: /workspace
 blueprint_yaml: "{{ workspace }}/examples/machine-learning/a4-highgpu-8g/a4high-slurm-blueprint.yaml"
 login_node: "{{ slurm_cluster_name }}-slurm-login-*"
 controller_node: "{{ slurm_cluster_name }}-controller"
 region: us-central1
 zone: us-central1-b
-network: "{{ test_name }}-net-0"
+network: "{{ deployment_name }}-net-0"
 post_deploy_tests:
 - test-validation/test-mounts.yml
 - test-validation/test-partitions.yml
@@ -47,5 +47,5 @@ cli_deployment_vars:
   slurm_cluster_name: "{{ slurm_cluster_name }}"
   disk_size_gb: 200
   a4h_cluster_size: 2
-  base_network_name: "{{ test_name }}"
+  base_network_name: "{{ deployment_name }}"
   a4h_dws_flex_enabled: true

tools/cloud-build/daily-tests/tests/ml-a4-highgpu-slurm.yml

@@ -16,15 +16,15 @@
 
 # region, zone must be defined in build file with --extra-vars flag!
 test_name: a4h-slurm
-deployment_name: a4h-slurm-{{ build }}
-slurm_cluster_name: "a4h{{ build[0:4] }}"
+deployment_name: d{{ build[0:3] }}{{ build[-3:] }}-a4h-slurm
+slurm_cluster_name: "d{{ build[0:3] }}{{ build[-3:] }}a4h"
 workspace: /workspace
 blueprint_yaml: "{{ workspace }}/examples/machine-learning/a4-highgpu-8g/a4high-slurm-blueprint.yaml"
 login_node: "{{ slurm_cluster_name }}-slurm-login-*"
 controller_node: "{{ slurm_cluster_name }}-controller"
 region: us-central1
 zone: us-central1-b
-network: "{{ test_name }}-net-0"
+network: "{{ deployment_name }}-net-0"
 post_deploy_tests:
 - test-validation/test-mounts.yml
 - test-validation/test-partitions.yml
@@ -49,4 +49,4 @@ cli_deployment_vars:
   disk_size_gb: 200
   a4h_cluster_size: 2
   a4h_reservation_name: nvidia-b200-db38b25a-c93d-4a7e-a3c5-ba4135be357e
-  base_network_name: "{{ test_name }}"
+  base_network_name: "{{ deployment_name }}"

tools/cloud-build/daily-tests/tests/ml-gke-e2e.yml

@@ -13,7 +13,7 @@
 # limitations under the License.
 ---
 test_name: ml-gke-e2e
-deployment_name: ml-gke-e2e-{{ build }}
+deployment_name: g{{ build[0:3] }}{{ build[-3:] }}-ml-gke-e2e
 region: asia-southeast1
 zone: asia-southeast1-b  # for remote node
 workspace: /workspace

tools/cloud-build/daily-tests/tests/ml-gke.yml

@@ -13,7 +13,7 @@
 # limitations under the License.
 ---
 test_name: ml-gke
-deployment_name: ml-gke-{{ build }}
+deployment_name: t{{ build[0:3] }}{{ build[-3:] }}-ml-gke
 region: asia-southeast1
 zone: asia-southeast1-b  # for remote node
 workspace: /workspace

tools/cloud-build/daily-tests/tests/ml-slurm.yml

@@ -15,8 +15,8 @@
 ---
 
 test_name: ml-slurm-v6
-deployment_name: ml-slurm-v6-{{ build }}
-network: "{{ test_name }}-net"
+deployment_name: d{{ build[0:3] }}{{ build[-3:] }}-ml-slurm-v6
+network: "{{ deployment_name }}-net"
 workspace: /workspace
 blueprint_yaml: "{{ workspace }}/examples/ml-slurm.yaml"
 packer_group_name: packer