Merge pull request #3659 from abbas1902/ofi_fix

abbas1902 · web-flow · commit c491a4ab2878 · 2025-02-10T17:56:12.000-08:00
Hotfix: Boolean logic on when cloud rdma ofi tunables run
diff --git a/cmd/root.go b/cmd/root.go
@@ -53,7 +53,7 @@ HPC deployments on the Google Cloud Platform.`,
 				logging.Fatal("cmd.Help function failed: %s", err)
 			}
 		},
-		Version:     "v1.46.0",
+		Version:     "v1.46.1",
 		Annotations: annotation,
 	}
 )
diff --git a/community/examples/hpc-slurm-h4d.yaml b/community/examples/hpc-slurm-h4d.yaml
@@ -128,14 +128,8 @@ deployment_groups:
       machine_type: n2-standard-4
       enable_login_public_ips: true
 
-  - id: slurm_controller_startup
-    source: modules/scripts/startup-script
-    settings:
-      set_ofi_cloud_rdma_tunables: true
-
   - id: slurm_controller
     source: community/modules/scheduler/schedmd-slurm-gcp-v6-controller
     use: [h4d-slurm-net-0, h4d_partition, slurm_login, homefs, appsfs]
     settings:
       enable_controller_public_ips: true
-      controller_startup_script: $(slurm_controller_startup.startup_script)
diff --git a/community/modules/compute/htcondor-execute-point/versions.tf b/community/modules/compute/htcondor-execute-point/versions.tf
@@ -25,6 +25,6 @@ terraform {
   }
 
   provider_meta "google" {
-    module_name = "blueprints/terraform/hpc-toolkit:htcondor-execute-point/v1.46.0"
+    module_name = "blueprints/terraform/hpc-toolkit:htcondor-execute-point/v1.46.1"
   }
 }
diff --git a/community/modules/compute/mig/versions.tf b/community/modules/compute/mig/versions.tf
@@ -22,6 +22,6 @@ terraform {
     }
   }
   provider_meta "google" {
-    module_name = "blueprints/terraform/hpc-toolkit:mig/v1.46.0"
+    module_name = "blueprints/terraform/hpc-toolkit:mig/v1.46.1"
   }
 }
diff --git a/community/modules/compute/schedmd-slurm-gcp-v6-nodeset-dynamic/versions.tf b/community/modules/compute/schedmd-slurm-gcp-v6-nodeset-dynamic/versions.tf
@@ -24,6 +24,6 @@ terraform {
     }
   }
   provider_meta "google" {
-    module_name = "blueprints/terraform/hpc-toolkit:schedmd-slurm-gcp-v6-nodeset-dynamic/v1.46.0"
+    module_name = "blueprints/terraform/hpc-toolkit:schedmd-slurm-gcp-v6-nodeset-dynamic/v1.46.1"
   }
 }
diff --git a/community/modules/compute/schedmd-slurm-gcp-v6-nodeset-tpu/versions.tf b/community/modules/compute/schedmd-slurm-gcp-v6-nodeset-tpu/versions.tf
@@ -18,6 +18,6 @@ terraform {
   required_version = ">= 1.3"
 
   provider_meta "google" {
-    module_name = "blueprints/terraform/hpc-toolkit:schedmd-slurm-gcp-v6-nodeset-tpu/v1.46.0"
+    module_name = "blueprints/terraform/hpc-toolkit:schedmd-slurm-gcp-v6-nodeset-tpu/v1.46.1"
   }
 }
diff --git a/community/modules/compute/schedmd-slurm-gcp-v6-nodeset/versions.tf b/community/modules/compute/schedmd-slurm-gcp-v6-nodeset/versions.tf
@@ -24,6 +24,6 @@ terraform {
     }
   }
   provider_meta "google" {
-    module_name = "blueprints/terraform/hpc-toolkit:schedmd-slurm-gcp-v6-nodeset/v1.46.0"
+    module_name = "blueprints/terraform/hpc-toolkit:schedmd-slurm-gcp-v6-nodeset/v1.46.1"
   }
 }
diff --git a/community/modules/compute/schedmd-slurm-gcp-v6-partition/versions.tf b/community/modules/compute/schedmd-slurm-gcp-v6-partition/versions.tf
@@ -18,6 +18,6 @@ terraform {
   required_version = ">= 1.3"
 
   provider_meta "google" {
-    module_name = "blueprints/terraform/hpc-toolkit:schedmd-slurm-gcp-v6-partition/v1.46.0"
+    module_name = "blueprints/terraform/hpc-toolkit:schedmd-slurm-gcp-v6-partition/v1.46.1"
   }
 }
diff --git a/community/modules/database/slurm-cloudsql-federation/versions.tf b/community/modules/database/slurm-cloudsql-federation/versions.tf
@@ -26,10 +26,10 @@ terraform {
     }
   }
   provider_meta "google" {
-    module_name = "blueprints/terraform/hpc-toolkit:slurm-cloudsql-federation/v1.46.0"
+    module_name = "blueprints/terraform/hpc-toolkit:slurm-cloudsql-federation/v1.46.1"
   }
   provider_meta "google-beta" {
-    module_name = "blueprints/terraform/hpc-toolkit:slurm-cloudsql-federation/v1.46.0"
+    module_name = "blueprints/terraform/hpc-toolkit:slurm-cloudsql-federation/v1.46.1"
   }
 
   required_version = ">= 0.13.0"
diff --git a/community/modules/file-system/cloud-storage-bucket/versions.tf b/community/modules/file-system/cloud-storage-bucket/versions.tf
@@ -30,10 +30,10 @@ terraform {
     }
   }
   provider_meta "google" {
-    module_name = "blueprints/terraform/hpc-toolkit:cloud-storage-bucket/v1.46.0"
+    module_name = "blueprints/terraform/hpc-toolkit:cloud-storage-bucket/v1.46.1"
   }
   provider_meta "google-beta" {
-    module_name = "blueprints/terraform/hpc-toolkit:cloud-storage-bucket/v1.46.0"
+    module_name = "blueprints/terraform/hpc-toolkit:cloud-storage-bucket/v1.46.1"
   }
   required_version = ">= 0.14.0"
 }
diff --git a/community/modules/file-system/nfs-server/versions.tf b/community/modules/file-system/nfs-server/versions.tf
@@ -30,7 +30,7 @@ terraform {
     }
   }
   provider_meta "google" {
-    module_name = "blueprints/terraform/hpc-toolkit:nfs-server/v1.46.0"
+    module_name = "blueprints/terraform/hpc-toolkit:nfs-server/v1.46.1"
   }
 
   required_version = ">= 0.14.0"
diff --git a/community/modules/files/fsi-montecarlo-on-batch/versions.tf b/community/modules/files/fsi-montecarlo-on-batch/versions.tf
@@ -35,9 +35,9 @@ terraform {
     }
   }
   provider_meta "google" {
-    module_name = "blueprints/terraform/hpc-toolkit:fsi-montecarlo-on-batch/v1.46.0"
+    module_name = "blueprints/terraform/hpc-toolkit:fsi-montecarlo-on-batch/v1.46.1"
   }
   provider_meta "google-beta" {
-    module_name = "blueprints/terraform/hpc-toolkit:fsi-montecarlo-on-batch/v1.46.0"
+    module_name = "blueprints/terraform/hpc-toolkit:fsi-montecarlo-on-batch/v1.46.1"
   }
 }
diff --git a/community/modules/network/private-service-access/versions.tf b/community/modules/network/private-service-access/versions.tf
@@ -30,11 +30,11 @@ terraform {
     }
   }
   provider_meta "google" {
-    module_name = "blueprints/terraform/hpc-toolkit:private-service-access/v1.46.0"
+    module_name = "blueprints/terraform/hpc-toolkit:private-service-access/v1.46.1"
   }
 
   provider_meta "google-beta" {
-    module_name = "blueprints/terraform/hpc-toolkit:private-service-access/v1.46.0"
+    module_name = "blueprints/terraform/hpc-toolkit:private-service-access/v1.46.1"
   }
 
   required_version = ">= 1.2"
diff --git a/community/modules/project/service-enablement/versions.tf b/community/modules/project/service-enablement/versions.tf
@@ -22,7 +22,7 @@ terraform {
     }
   }
   provider_meta "google" {
-    module_name = "blueprints/terraform/hpc-toolkit:service-enablement/v1.46.0"
+    module_name = "blueprints/terraform/hpc-toolkit:service-enablement/v1.46.1"
   }
 
   required_version = ">= 0.14.0"
diff --git a/community/modules/pubsub/bigquery-sub/versions.tf b/community/modules/pubsub/bigquery-sub/versions.tf
@@ -26,10 +26,10 @@ terraform {
     }
   }
   provider_meta "google" {
-    module_name = "blueprints/terraform/hpc-toolkit:bigquery-sub/v1.46.0"
+    module_name = "blueprints/terraform/hpc-toolkit:bigquery-sub/v1.46.1"
   }
   provider_meta "google-beta" {
-    module_name = "blueprints/terraform/hpc-toolkit:bigquery-sub/v1.46.0"
+    module_name = "blueprints/terraform/hpc-toolkit:bigquery-sub/v1.46.1"
   }
   required_version = ">= 1.0"
 }
diff --git a/community/modules/pubsub/topic/versions.tf b/community/modules/pubsub/topic/versions.tf
@@ -27,6 +27,6 @@ terraform {
     }
   }
   provider_meta "google" {
-    module_name = "blueprints/terraform/hpc-toolkit:topic/v1.46.0"
+    module_name = "blueprints/terraform/hpc-toolkit:topic/v1.46.1"
   }
 }
diff --git a/community/modules/scheduler/htcondor-access-point/versions.tf b/community/modules/scheduler/htcondor-access-point/versions.tf
@@ -26,7 +26,7 @@ terraform {
     }
   }
   provider_meta "google" {
-    module_name = "blueprints/terraform/hpc-toolkit:htcondor-access-point/v1.46.0"
+    module_name = "blueprints/terraform/hpc-toolkit:htcondor-access-point/v1.46.1"
   }
 
   required_version = ">= 1.1"
diff --git a/community/modules/scheduler/htcondor-central-manager/versions.tf b/community/modules/scheduler/htcondor-central-manager/versions.tf
@@ -22,7 +22,7 @@ terraform {
     }
   }
   provider_meta "google" {
-    module_name = "blueprints/terraform/hpc-toolkit:htcondor-central-manager/v1.46.0"
+    module_name = "blueprints/terraform/hpc-toolkit:htcondor-central-manager/v1.46.1"
   }
 
   required_version = ">= 1.1.0"
diff --git a/community/modules/scheduler/htcondor-pool-secrets/versions.tf b/community/modules/scheduler/htcondor-pool-secrets/versions.tf
@@ -26,7 +26,7 @@ terraform {
     }
   }
   provider_meta "google" {
-    module_name = "blueprints/terraform/hpc-toolkit:htcondor-pool-secrets/v1.46.0"
+    module_name = "blueprints/terraform/hpc-toolkit:htcondor-pool-secrets/v1.46.1"
   }
 
   required_version = ">= 1.3.0"
diff --git a/community/modules/scheduler/schedmd-slurm-gcp-v6-controller/versions.tf b/community/modules/scheduler/schedmd-slurm-gcp-v6-controller/versions.tf
@@ -24,6 +24,6 @@ terraform {
     }
   }
   provider_meta "google" {
-    module_name = "blueprints/terraform/hpc-toolkit:schedmd-slurm-gcp-v6-controller/v1.46.0"
+    module_name = "blueprints/terraform/hpc-toolkit:schedmd-slurm-gcp-v6-controller/v1.46.1"
   }
 }
diff --git a/community/modules/scheduler/schedmd-slurm-gcp-v6-login/versions.tf b/community/modules/scheduler/schedmd-slurm-gcp-v6-login/versions.tf
@@ -24,6 +24,6 @@ terraform {
     }
   }
   provider_meta "google" {
-    module_name = "blueprints/terraform/hpc-toolkit:schedmd-slurm-gcp-v6-login/v1.46.0"
+    module_name = "blueprints/terraform/hpc-toolkit:schedmd-slurm-gcp-v6-login/v1.46.1"
   }
 }
diff --git a/community/modules/scripts/wait-for-startup/versions.tf b/community/modules/scripts/wait-for-startup/versions.tf
@@ -22,7 +22,7 @@ terraform {
     }
   }
   provider_meta "google" {
-    module_name = "blueprints/terraform/hpc-toolkit:wait-for-startup/v1.46.0"
+    module_name = "blueprints/terraform/hpc-toolkit:wait-for-startup/v1.46.1"
   }
 
   required_version = ">= 0.14.0"
diff --git a/community/modules/scripts/windows-startup-script/versions.tf b/community/modules/scripts/windows-startup-script/versions.tf
@@ -16,7 +16,7 @@
 
 terraform {
   provider_meta "google" {
-    module_name = "blueprints/terraform/hpc-toolkit:windows-startup-script/v1.46.0"
+    module_name = "blueprints/terraform/hpc-toolkit:windows-startup-script/v1.46.1"
   }
 
   required_version = ">= 0.14.0"
diff --git a/modules/compute/gke-node-pool/versions.tf b/modules/compute/gke-node-pool/versions.tf
@@ -30,6 +30,6 @@ terraform {
     }
   }
   provider_meta "google" {
-    module_name = "blueprints/terraform/hpc-toolkit:gke-node-pool/v1.46.0"
+    module_name = "blueprints/terraform/hpc-toolkit:gke-node-pool/v1.46.1"
   }
 }
diff --git a/modules/compute/vm-instance/versions.tf b/modules/compute/vm-instance/versions.tf
@@ -31,10 +31,10 @@ terraform {
     }
   }
   provider_meta "google" {
-    module_name = "blueprints/terraform/hpc-toolkit:vm-instance/v1.46.0"
+    module_name = "blueprints/terraform/hpc-toolkit:vm-instance/v1.46.1"
   }
   provider_meta "google-beta" {
-    module_name = "blueprints/terraform/hpc-toolkit:vm-instance/v1.46.0"
+    module_name = "blueprints/terraform/hpc-toolkit:vm-instance/v1.46.1"
   }
 
   required_version = ">= 1.3.0"
diff --git a/modules/file-system/filestore/versions.tf b/modules/file-system/filestore/versions.tf
@@ -26,10 +26,10 @@ terraform {
     }
   }
   provider_meta "google" {
-    module_name = "blueprints/terraform/hpc-toolkit:filestore/v1.46.0"
+    module_name = "blueprints/terraform/hpc-toolkit:filestore/v1.46.1"
   }
   provider_meta "google-beta" {
-    module_name = "blueprints/terraform/hpc-toolkit:filestore/v1.46.0"
+    module_name = "blueprints/terraform/hpc-toolkit:filestore/v1.46.1"
   }
 
   required_version = ">= 1.3.0"
diff --git a/modules/file-system/gke-persistent-volume/versions.tf b/modules/file-system/gke-persistent-volume/versions.tf
@@ -29,6 +29,6 @@ terraform {
     }
   }
   provider_meta "google" {
-    module_name = "blueprints/terraform/hpc-toolkit:gke-persistent-volume/v1.46.0"
+    module_name = "blueprints/terraform/hpc-toolkit:gke-persistent-volume/v1.46.1"
   }
 }
diff --git a/modules/file-system/gke-storage/versions.tf b/modules/file-system/gke-storage/versions.tf
@@ -16,6 +16,6 @@ terraform {
   required_version = ">= 1.5"
 
   provider_meta "google" {
-    module_name = "blueprints/terraform/hpc-toolkit:gke-storage/v1.46.0"
+    module_name = "blueprints/terraform/hpc-toolkit:gke-storage/v1.46.1"
   }
 }
diff --git a/modules/monitoring/dashboard/versions.tf b/modules/monitoring/dashboard/versions.tf
@@ -22,7 +22,7 @@ terraform {
     }
   }
   provider_meta "google" {
-    module_name = "blueprints/terraform/hpc-toolkit:dashboard/v1.46.0"
+    module_name = "blueprints/terraform/hpc-toolkit:dashboard/v1.46.1"
   }
 
   required_version = ">= 0.14.0"
diff --git a/modules/network/firewall-rules/versions.tf b/modules/network/firewall-rules/versions.tf
@@ -22,7 +22,7 @@ terraform {
     }
   }
   provider_meta "google" {
-    module_name = "blueprints/terraform/hpc-toolkit:firewall-rules/v1.46.0"
+    module_name = "blueprints/terraform/hpc-toolkit:firewall-rules/v1.46.1"
   }
 
   required_version = ">= 1.5"
diff --git a/modules/network/pre-existing-subnetwork/versions.tf b/modules/network/pre-existing-subnetwork/versions.tf
@@ -22,7 +22,7 @@ terraform {
     }
   }
   provider_meta "google" {
-    module_name = "blueprints/terraform/hpc-toolkit:pre-existing-subnetwork/v1.46.0"
+    module_name = "blueprints/terraform/hpc-toolkit:pre-existing-subnetwork/v1.46.1"
   }
 
   required_version = ">= 1.5"
diff --git a/modules/network/pre-existing-vpc/versions.tf b/modules/network/pre-existing-vpc/versions.tf
@@ -22,7 +22,7 @@ terraform {
     }
   }
   provider_meta "google" {
-    module_name = "blueprints/terraform/hpc-toolkit:pre-existing-vpc/v1.46.0"
+    module_name = "blueprints/terraform/hpc-toolkit:pre-existing-vpc/v1.46.1"
   }
 
   required_version = ">= 1.5"
diff --git a/modules/scheduler/batch-login-node/versions.tf b/modules/scheduler/batch-login-node/versions.tf
@@ -22,7 +22,7 @@ terraform {
     }
   }
   provider_meta "google" {
-    module_name = "blueprints/terraform/hpc-toolkit:batch-login-node/v1.46.0"
+    module_name = "blueprints/terraform/hpc-toolkit:batch-login-node/v1.46.1"
   }
 
   required_version = ">= 0.14.0"
diff --git a/modules/scheduler/gke-cluster/versions.tf b/modules/scheduler/gke-cluster/versions.tf
@@ -30,6 +30,6 @@ terraform {
     }
   }
   provider_meta "google" {
-    module_name = "blueprints/terraform/hpc-toolkit:gke-cluster/v1.46.0"
+    module_name = "blueprints/terraform/hpc-toolkit:gke-cluster/v1.46.1"
   }
 }
diff --git a/modules/scheduler/pre-existing-gke-cluster/versions.tf b/modules/scheduler/pre-existing-gke-cluster/versions.tf
@@ -23,7 +23,7 @@ terraform {
   }
 
   provider_meta "google" {
-    module_name = "blueprints/terraform/hpc-toolkit:pre-existing-gke-cluster/v1.46.0"
+    module_name = "blueprints/terraform/hpc-toolkit:pre-existing-gke-cluster/v1.46.1"
   }
 
   required_version = ">= 1.3"
diff --git a/modules/scripts/startup-script/main.tf b/modules/scripts/startup-script/main.tf
@@ -93,7 +93,7 @@ locals {
     }
   ]
 
-  ofi_runner = var.set_ofi_cloud_rdma_tunables == "" ? [] : [
+  ofi_runner = !var.set_ofi_cloud_rdma_tunables ? [] : [
     {
       type        = "data"
       destination = "/etc/profile.d/set_ofi_cloud_rdma_tunables.sh"
@@ -102,7 +102,8 @@ locals {
         export FI_PROVIDER="verbs;ofi_rxm"
         export FI_OFI_RXM_USE_RNDV_WRITE=1
         export FI_VERBS_INLINE_SIZE=39
-        export I_MPI_FABRICS="shm:rxm"
+        export I_MPI_FABRICS="shm:ofi"
+        export FI_UNIVERSE_SIZE=3072
         EOT
     },
   ]
diff --git a/modules/scripts/startup-script/versions.tf b/modules/scripts/startup-script/versions.tf
@@ -30,7 +30,7 @@ terraform {
     }
   }
   provider_meta "google" {
-    module_name = "blueprints/terraform/hpc-toolkit:startup-script/v1.46.0"
+    module_name = "blueprints/terraform/hpc-toolkit:startup-script/v1.46.1"
   }
 
   required_version = ">= 1.5"

Original file line number	Diff line number	Diff line change
@@ -53,7 +53,7 @@ HPC deployments on the Google Cloud Platform.`,
`53`	`53`	`logging.Fatal("cmd.Help function failed: %s", err)`
`54`	`54`	`}`
`55`	`55`	`},`
`56`		`- Version: "v1.46.0",`
	`56`	`+ Version: "v1.46.1",`
`57`	`57`	`Annotations: annotation,`
`58`	`58`	`}`
`59`	`59`	`)`
Original file line number	Diff line number	Diff line change
`@@ -25,6 +25,6 @@ terraform {`
`25`	`25`	`}`
`26`	`26`
`27`	`27`	`provider_meta "google" {`
`28`		`- module_name = "blueprints/terraform/hpc-toolkit:htcondor-execute-point/v1.46.0"`
	`28`	`+ module_name = "blueprints/terraform/hpc-toolkit:htcondor-execute-point/v1.46.1"`
`29`	`29`	`}`
`30`	`30`	`}`
Original file line number	Diff line number	Diff line change
`@@ -22,6 +22,6 @@ terraform {`
`22`	`22`	`}`
`23`	`23`	`}`
`24`	`24`	`provider_meta "google" {`
`25`		`- module_name = "blueprints/terraform/hpc-toolkit:mig/v1.46.0"`
	`25`	`+ module_name = "blueprints/terraform/hpc-toolkit:mig/v1.46.1"`
`26`	`26`	`}`
`27`	`27`	`}`
Original file line number	Diff line number	Diff line change
`@@ -24,6 +24,6 @@ terraform {`
`24`	`24`	`}`
`25`	`25`	`}`
`26`	`26`	`provider_meta "google" {`
`27`		`- module_name = "blueprints/terraform/hpc-toolkit:schedmd-slurm-gcp-v6-nodeset-dynamic/v1.46.0"`
	`27`	`+ module_name = "blueprints/terraform/hpc-toolkit:schedmd-slurm-gcp-v6-nodeset-dynamic/v1.46.1"`
`28`	`28`	`}`
`29`	`29`	`}`
Original file line number	Diff line number	Diff line change
`@@ -18,6 +18,6 @@ terraform {`
`18`	`18`	`required_version = ">= 1.3"`
`19`	`19`
`20`	`20`	`provider_meta "google" {`
`21`		`- module_name = "blueprints/terraform/hpc-toolkit:schedmd-slurm-gcp-v6-nodeset-tpu/v1.46.0"`
	`21`	`+ module_name = "blueprints/terraform/hpc-toolkit:schedmd-slurm-gcp-v6-nodeset-tpu/v1.46.1"`
`22`	`22`	`}`
`23`	`23`	`}`
Original file line number	Diff line number	Diff line change
`@@ -26,10 +26,10 @@ terraform {`
`26`	`26`	`}`
`27`	`27`	`}`
`28`	`28`	`provider_meta "google" {`
`29`		`- module_name = "blueprints/terraform/hpc-toolkit:slurm-cloudsql-federation/v1.46.0"`
	`29`	`+ module_name = "blueprints/terraform/hpc-toolkit:slurm-cloudsql-federation/v1.46.1"`
`30`	`30`	`}`
`31`	`31`	`provider_meta "google-beta" {`
`32`		`- module_name = "blueprints/terraform/hpc-toolkit:slurm-cloudsql-federation/v1.46.0"`
	`32`	`+ module_name = "blueprints/terraform/hpc-toolkit:slurm-cloudsql-federation/v1.46.1"`
`33`	`33`	`}`
`34`	`34`
`35`	`35`	`required_version = ">= 0.13.0"`
Original file line number	Diff line number	Diff line change
`@@ -30,10 +30,10 @@ terraform {`
`30`	`30`	`}`
`31`	`31`	`}`
`32`	`32`	`provider_meta "google" {`
`33`		`- module_name = "blueprints/terraform/hpc-toolkit:cloud-storage-bucket/v1.46.0"`
	`33`	`+ module_name = "blueprints/terraform/hpc-toolkit:cloud-storage-bucket/v1.46.1"`
`34`	`34`	`}`
`35`	`35`	`provider_meta "google-beta" {`
`36`		`- module_name = "blueprints/terraform/hpc-toolkit:cloud-storage-bucket/v1.46.0"`
	`36`	`+ module_name = "blueprints/terraform/hpc-toolkit:cloud-storage-bucket/v1.46.1"`
`37`	`37`	`}`
`38`	`38`	`required_version = ">= 0.14.0"`
`39`	`39`	`}`