Archon aws ias test

.gitignore

@@ -19,4 +19,5 @@ analysis/
 test.sh
 tests/data/
 data/*
-.scratch
+.scratch
+aws-batch-arch-for-protein-folding.code-workspace

Notes.md

@@ -0,0 +1,6 @@
+
+# Package infrastructure for deployment
+aws cloudformation package --template-file infrastructure/cloudformation/batch-protein-folding-cfn-root.yaml --s3-bucket archon-iac --s3-prefix main --output-template-file ../protein-folding.cfn.yaml --profile archon
+
+# Update stack
+aws cloudformation update-stack --stack-name archon-iac-galaxy --template-body file://../protein-folding.cfn.yaml --capabilities CAPABILITY_NAMED_IAM --profile archon --disable-rollback --output json --profile archon

infrastructure/cloudformation/batch-protein-folding-cfn-batch.yaml

@@ -35,11 +35,11 @@ Parameters:
   CreateAWSServiceRoleForEC2Spot:
     Description: Create the AWSServiceRoleForEC2Spot service-linked role
     Type: String
-    Default: "N"
+    Default: "Y"
   CreateAWSServiceRoleForEC2SpotFleet:
     Description: Create the AWSServiceRoleForEC2SpotFleet service-linked role
     Type: String
-    Default: "N"
+    Default: "Y"
   CreateP4dComputeEnvironment:
     Description: Create a compute environment for p4d.24xlarge instance types? Note that availabilty is region-specific
     Type: String
@@ -136,8 +136,6 @@ Resources:
                 Value: !Ref ApplicationName
               - Key: StackId
                 Value: !Ref AWS::StackId
-              - Key: SolutionId
-                Value: SO9164
         UserData:
           Fn::Base64:
             Fn::Join:
@@ -189,8 +187,6 @@ Resources:
           Value: !Ref ApplicationName
         - Key: StackId
           Value: !Ref AWS::StackId
-        - Key: SolutionId
-          Value: SO9164
 
   AWSServiceRoleForEC2SpotServiceLinkedRole:
     Type: AWS::IAM::ServiceLinkedRole
@@ -302,6 +298,38 @@ Resources:
         StackId: !Ref "AWS::StackId"
         SolutionId: SO9164
 
+  G4dnSpotComputeEnvironment:
+    Type: AWS::Batch::ComputeEnvironment
+    Properties:
+      ComputeResources:
+        AllocationStrategy: SPOT_CAPACITY_OPTIMIZED
+        BidPercentage: 100
+        InstanceRole: !Ref InstanceProfile
+        InstanceTypes:
+          - g4dn
+        LaunchTemplate:
+          LaunchTemplateId:
+            Ref: InstanceLaunchTemplate
+          Version: $Latest
+        MaxvCpus: 256
+        MinvCpus: 0
+        SecurityGroupIds:
+          - Ref: DefaultSecurityGroupID
+        SpotIamFleetRole:
+          "Fn::GetAtt": AmazonEC2SpotFleetTaggingRole.Arn
+        Subnets: !Ref Subnets
+        Type: SPOT
+        Tags:
+          Application: !Ref ApplicationName
+          StackId: !Ref "AWS::StackId"
+          SolutionId: SO9164
+      State: ENABLED
+      Type: MANAGED
+      Tags:
+        Application: !Ref ApplicationName
+        StackId: !Ref "AWS::StackId"
+        SolutionId: SO9164
+
   G5ComputeEnvironment:
     Type: AWS::Batch::ComputeEnvironment
     Condition: CreateG5ComputeEnvironment
@@ -387,7 +415,6 @@ Resources:
       Tags:
         Application: !Ref ApplicationName
         StackId: !Ref "AWS::StackId"
-        SolutionId: SO9164
 
   G4dnJobQueue:
     Type: AWS::Batch::JobQueue
@@ -400,7 +427,18 @@ Resources:
       Tags:
         Application: !Ref ApplicationName
         StackId: !Ref "AWS::StackId"
-        SolutionId: SO9164
+
+  G4dnSpotJobQueue:
+    Type: AWS::Batch::JobQueue
+    Properties:
+      ComputeEnvironmentOrder:
+        - ComputeEnvironment: !Ref G4dnSpotComputeEnvironment
+          Order: 1
+      Priority: 10
+      State: ENABLED
+      Tags:
+        Application: !Ref ApplicationName
+        StackId: !Ref "AWS::StackId"
 
   G5JobQueue:
     Type: AWS::Batch::JobQueue
@@ -414,7 +452,6 @@ Resources:
       Tags:
         Application: !Ref ApplicationName
         StackId: !Ref "AWS::StackId"
-        SolutionId: SO9164
 
   P4dJobQueue:
     Type: AWS::Batch::JobQueue
@@ -460,7 +497,6 @@ Resources:
       Tags:
         Application: !Ref ApplicationName
         StackId: !Ref "AWS::StackId"
-        SolutionId: SO9164
       Type: container
 
 Outputs:
@@ -480,6 +516,10 @@ Outputs:
     Description: Job queue for on-demand GPU instances.
     Value:
       Ref: G4dnJobQueue
+  G4dnSpotJobQueue:
+    Description: Job queue for on-demand GPU instances.
+    Value:
+      Ref: G4dnSpotJobQueue
   G5JobQueue:
     Condition: CreateG5ComputeEnvironment
     Description: Job queue for on-demand G5 instances.

infrastructure/cloudformation/batch-protein-folding-cfn-container.yaml

@@ -19,12 +19,12 @@ Parameters:
   DownloadFsxData:
     Description: Automatically populate FSx for Lustre file system with model parameters and sequence databases?
     Type: String
-    Default: "Y"
+    Default: "N"
     AllowedValues: [Y, N]
   CodeRepoS3BucketName:
     Description: S3 Bucket Name to retrieve code zip
     Type: String    
-    Default: "aws-hcls-ml"
+    Default: "archon-iac"
   S3BucketARN:
     Description: S3 Bucket ARN
     Type: String
@@ -68,8 +68,6 @@ Resources:
           Value: !Ref ApplicationName
         - Key: StackId
           Value: !Ref AWS::StackId
-        - Key: SolutionId
-          Value: SO9164
 
   DownloadContainerRegistry:
     Type: AWS::ECR::Repository
@@ -83,8 +81,6 @@ Resources:
           Value: !Ref ApplicationName
         - Key: StackId
           Value: !Ref "AWS::StackId"    
-        - Key: SolutionId
-          Value: SO9164 
 
   CodeBuildRole:
     Type: AWS::IAM::Role
@@ -163,8 +159,6 @@ Resources:
           Value: !Ref ApplicationName
         - Key: StackId
           Value: !Ref "AWS::StackId"
-        - Key: SolutionId
-          Value: SO9164
 
   EncryptionKey:
     Type: "AWS::KMS::Key"
@@ -228,8 +222,6 @@ Resources:
           Value: !Ref ApplicationName
         - Key: StackId
           Value: !Ref "AWS::StackId"
-        - Key: SolutionId
-          Value: SO9164
 
   CodeBuildLambdaExecutionRole:
     Type: AWS::IAM::Role
@@ -258,7 +250,8 @@ Resources:
   StartCodeBuildLambda:
     Type: AWS::Lambda::Function
     Properties:
-      ReservedConcurrentExecutions: 10
+#      ReservedConcurrentExecutions: 10
+#      ReservedConcurrentExecutions: 0
       Code:
         ZipFile: |
           import logging          
@@ -302,9 +295,7 @@ Resources:
         - Key: Application
           Value: !Ref ApplicationName
         - Key: StackId
-          Value: !Ref "AWS::StackId"
-        - Key: SolutionId
-          Value: SO9164      
+          Value: !Ref "AWS::StackId"  
 
   ##################################################
   # Delete container repo on stack deletion
@@ -337,7 +328,8 @@ Resources:
   DeleteContainerRepoLambda:
     Type: AWS::Lambda::Function
     Properties:
-      ReservedConcurrentExecutions: 10
+#      ReservedConcurrentExecutions: 10
+#      ReservedConcurrentExecutions: 0
       Code:
         ZipFile: |
           import logging          
@@ -386,8 +378,6 @@ Resources:
           Value: !Ref ApplicationName
         - Key: StackId
           Value: !Ref "AWS::StackId"
-        - Key: SolutionId
-          Value: SO9164
 
 ##################################################
 # Create Download Container
@@ -429,9 +419,7 @@ Resources:
         - Key: Application
           Value: !Ref ApplicationName
         - Key: StackId
-          Value: !Ref "AWS::StackId"        
-        - Key: SolutionId
-          Value: SO9164                  
+          Value: !Ref "AWS::StackId"                     
 
   StartDownloadContainerBuild:
     Type: Custom::ResourceForBuildingContainer

infrastructure/cloudformation/batch-protein-folding-cfn-download.yaml

@@ -69,20 +69,20 @@ Resources:
 
                       download_scripts = [
                           "/scripts/download_alphafold_params.sh",
-                          "/scripts/download_bfd.sh",
-                          "/scripts/download_esmfold_params.sh",
-                          "/scripts/download_mgnify.sh",
-                          "/scripts/download_openfold_params.sh",
-                          "/scripts/download_pdb_mmcif.sh",
-                          "/scripts/download_pdb_seqres.sh",
-                          "/scripts/download_pdb70.sh",
-                          "/scripts/download_small_bfd.sh",
-                          "/scripts/download_uniref30.sh",
-                          "/scripts/download_uniprot.sh",                          
-                          "/scripts/download_uniref90.sh",
-                          "/scripts/download_omegafold_params.sh",
-                          "/scripts/download_rfdesign_params.sh",
-                          "/scripts/download_diffdock_params.sh",
+                          # "/scripts/download_bfd.sh",
+                          # "/scripts/download_esmfold_params.sh",
+                          # "/scripts/download_mgnify.sh",
+                          # "/scripts/download_openfold_params.sh",
+                          # "/scripts/download_pdb_mmcif.sh",
+                          # "/scripts/download_pdb_seqres.sh",
+                          # "/scripts/download_pdb70.sh",
+                          # "/scripts/download_small_bfd.sh",
+                          # "/scripts/download_uniref30.sh",
+                          # "/scripts/download_uniprot.sh",                          
+                          # "/scripts/download_uniref90.sh",
+                          # "/scripts/download_omegafold_params.sh",
+                          # "/scripts/download_rfdesign_params.sh",
+                          # "/scripts/download_diffdock_params.sh",
                           "/scripts/download_rfdiffusion_params.sh"
                       ]
 
@@ -188,8 +188,6 @@ Resources:
           Value: !Ref ApplicationName
         - Key: StackId
           Value: !Ref AWS::StackId
-        - Key: SolutionId
-          Value: SO9164      
 
   DownloadFsxData:
     Type: Custom::ResourceForDownloadingData

infrastructure/cloudformation/batch-protein-folding-cfn-fsx-lustre.yaml

@@ -29,7 +29,7 @@ Parameters:
       - 250
       - 500
       - 1000
-    Default: 500
+    Default: 125
   Subnet:
     Description: Subnet ID for the file system.
     Type: String
@@ -65,9 +65,7 @@ Resources:
             Ref: ApplicationName
         - Key: StackId
           Value:
-            Ref: AWS::StackId
-        - Key: SolutionId
-          Value: SO9164            
+            Ref: AWS::StackId      
 
   DataRepoAssociation:
     Type: AWS::FSx::DataRepositoryAssociation

infrastructure/cloudformation/batch-protein-folding-cfn-module-alphafold2.yaml

@@ -44,8 +44,6 @@ Resources:
           Value: !Ref ApplicationName
         - Key: StackId
           Value: !Ref "AWS::StackId"       
-        - Key: SolutionId
-          Value: SO9164  
 
   CodeBuildProjectAlphaFold2:
     Type: AWS::CodeBuild::Project
@@ -82,8 +80,6 @@ Resources:
           Value: !Ref ApplicationName
         - Key: StackId
           Value: !Ref "AWS::StackId"           
-        - Key: SolutionId
-          Value: SO9164  
 
   StartAlphaFold2ContainerBuild:
     Type: Custom::ResourceForBuildingContainer
@@ -132,7 +128,6 @@ Resources:
       Tags:
         Application: !Ref ApplicationName
         StackId: !Ref "AWS::StackId"
-        SolutionId: SO9164  
       Type: container    
 
   DeleteAlphaFold2ContainerRepo:

infrastructure/cloudformation/batch-protein-folding-cfn-module-proteinmpnn.yaml

@@ -43,8 +43,6 @@ Resources:
           Value: !Ref ApplicationName
         - Key: StackId
           Value: !Ref "AWS::StackId" 
-        - Key: SolutionId
-          Value: SO9164  
 
   CodeBuildProjectProteinMPNN:
     Type: AWS::CodeBuild::Project
@@ -81,8 +79,6 @@ Resources:
           Value: !Ref ApplicationName
         - Key: StackId
           Value: !Ref "AWS::StackId"   
-        - Key: SolutionId
-          Value: SO9164  
 
   StartProteinMPNNContainerBuild:
     Type: Custom::ResourceForBuildingContainer

infrastructure/cloudformation/batch-protein-folding-cfn-module-rfdiffusion.yaml

@@ -43,8 +43,6 @@ Resources:
           Value: !Ref ApplicationName
         - Key: StackId
           Value: !Ref "AWS::StackId" 
-        - Key: SolutionId
-          Value: SO9164  
 
   CodeBuildProjectRFDiffusion:
     Type: AWS::CodeBuild::Project
@@ -126,7 +124,6 @@ Resources:
       Tags:
         Application: !Ref ApplicationName
         StackId: !Ref "AWS::StackId"
-        SolutionId: SO9164  
       Type: container
 
   DeleteRFDiffusionContainerRepo: