|
50 | 50 | accelerator_type: nvidia-b200 |
51 | 51 | version_prefix: "1.32." |
52 | 52 |
|
53 | | - enable_periodic_health_checks: false # Make this true to run CHS (healthchecks) |
54 | | - health_check_schedule: "0 0 * * 0" # Run the health check at 12:00 AM (midnight) every Sunday |
55 | | - |
56 | | - permissions_file_staged_path: $(ghpc_stage("./chs-permissions.yaml.tftpl")) |
57 | | - chs_output_bucket_name: chs-result |
58 | | - chs_pvc_claim_name: chs-output-pvc |
59 | | - chs_cronjob_rendered_path: $(ghpc_stage("./chs-cronjob.yaml.tftpl")) |
60 | | - chs_pvc_rendered_path: $(ghpc_stage("./chs-pvc.yaml.tftpl")) |
61 | 53 |
|
62 | 54 | deployment_groups: |
63 | 55 | - group: primary |
@@ -140,7 +132,6 @@ deployment_groups: |
140 | 132 | - stackdriver.resourceMetadata.writer |
141 | 133 | - storage.objectAdmin |
142 | 134 | - artifactregistry.reader |
143 | | - - container.admin |
144 | 135 |
|
145 | 136 | - id: training_bucket |
146 | 137 | source: community/modules/file-system/cloud-storage-bucket |
@@ -240,27 +231,6 @@ deployment_groups: |
240 | 231 | source: modules/management/kubectl-apply |
241 | 232 | use: [a4-cluster] |
242 | 233 | settings: |
243 | | - apply_manifests: |
244 | | - - source: $(vars.permissions_file_staged_path) |
245 | | - template_vars: |
246 | | - project_id: $(vars.project_id) |
247 | | - - source: $(vars.chs_pvc_rendered_path) |
248 | | - enable: $(vars.enable_periodic_health_checks) |
249 | | - template_vars: |
250 | | - pvc_name: $(vars.chs_pvc_claim_name) |
251 | | - access_mode: ReadWriteOnce |
252 | | - capacity: 1Gi |
253 | | - storage_class_name: standard-rwo |
254 | | - - source: $(vars.chs_cronjob_rendered_path) |
255 | | - enable: $(vars.enable_periodic_health_checks) |
256 | | - template_vars: |
257 | | - project_id: $(vars.project_id) |
258 | | - deployment_name: $(vars.deployment_name) |
259 | | - region: $(vars.region) |
260 | | - machine_type: a4-highgpu-8g |
261 | | - gcs_bucket: $(vars.chs_output_bucket_name) |
262 | | - gcs_pvc: $(vars.chs_pvc_claim_name) |
263 | | - cronjob_schedule: $(vars.health_check_schedule) |
264 | 234 | kueue: |
265 | 235 | install: true |
266 | 236 | config_path: $(vars.kueue_configuration_path) |
|
0 commit comments