feat: Add instance flavor and size arguments for det deploy aws [INFE…

…NG-227] (determined-ai#7931)
jgongd · Sep 22, 2023 · d25612b · d25612b
1 parent e031a2f
commit d25612b
Show file tree

Hide file tree

Showing 4 changed files with 33 additions and 2 deletions.
diff --git a/harness/determined/deploy/aws/cli.py b/harness/determined/deploy/aws/cli.py
@@ -150,6 +150,9 @@ def deploy_aws(command: str, args: argparse.Namespace) -> None:
                 f"The agent-subnet-id was set to '{args.agent_subnet_id}', but the "
                 f"deployment-type={args.deployment_type}."
             )
+    if args.deployment_type == constants.deployment_types.SIMPLE_RDS:
+        if args.db_size is not None and args.db_size < 20:
+            raise ValueError("The db-size must be greater than or equal to 20 GB")
 
     if args.deployment_type == constants.deployment_types.GOVCLOUD:
         if args.region not in ["us-gov-east-1", "us-gov-west-1"]:
@@ -198,6 +201,8 @@ def deploy_aws(command: str, args: argparse.Namespace) -> None:
         constants.cloudformation.VERSION: args.det_version,
         constants.cloudformation.INBOUND_CIDR: args.inbound_cidr,
         constants.cloudformation.DB_PASSWORD: args.db_password,
+        constants.cloudformation.DB_INSTANCE_TYPE: args.db_instance_type,
+        constants.cloudformation.DB_SIZE: args.db_size,
         constants.cloudformation.MAX_IDLE_AGENT_PERIOD: args.max_idle_agent_period,
         constants.cloudformation.MAX_AGENT_STARTING_PERIOD: args.max_agent_starting_period,
         constants.cloudformation.MAX_AUX_CONTAINERS_PER_AGENT: args.max_aux_containers_per_agent,
@@ -437,6 +442,18 @@ def handle_dump_master_config_template(args: argparse.Namespace) -> None:
                     default=constants.defaults.DB_PASSWORD,
                     help="password for master database",
                 ),
+                Arg(
+                    "--db-instance-type",
+                    type=str,
+                    default=constants.defaults.DB_INSTANCE_TYPE,
+                    help="instance type for master database",
+                ),
+                Arg(
+                    "--db-size",
+                    type=int,
+                    default=constants.defaults.DB_SIZE,
+                    help="storage size in GB for master database",
+                ),
                 Arg(
                     "--max-idle-agent-period",
                     type=str,

diff --git a/harness/determined/deploy/aws/constants.py b/harness/determined/deploy/aws/constants.py
@@ -15,6 +15,8 @@ class defaults:
     REGION = "us-west-2"
     STACK_TAG_KEY = "managed-by"
     STACK_TAG_VALUE = "determined"
+    DB_SIZE = 20
+    DB_INSTANCE_TYPE = "db.m7g.large"
 
 
 class cloudformation:
@@ -36,6 +38,8 @@ class cloudformation:
     MASTER_SCHEME = "MasterScheme"
     VERSION = "Version"
     DB_PASSWORD = "DBPassword"
+    DB_INSTANCE_TYPE = "DBInstanceType"
+    DB_SIZE = "DBSize"
     ENABLE_CORS = "EnableCORS"
     MASTER_TLS_CERT = "MasterTLSCert"
     MASTER_TLS_KEY = "MasterTLSKey"

diff --git a/harness/determined/deploy/aws/deployment_types/base.py b/harness/determined/deploy/aws/deployment_types/base.py
@@ -22,6 +22,8 @@
     constants.cloudformation.INBOUND_CIDR,
     constants.cloudformation.VERSION,
     constants.cloudformation.DB_PASSWORD,
+    constants.cloudformation.DB_INSTANCE_TYPE,
+    constants.cloudformation.DB_SIZE,
     constants.cloudformation.MAX_IDLE_AGENT_PERIOD,
     constants.cloudformation.MAX_AGENT_STARTING_PERIOD,
     constants.cloudformation.MAX_AUX_CONTAINERS_PER_AGENT,

diff --git a/harness/determined/deploy/aws/templates/simple-rds.yaml b/harness/determined/deploy/aws/templates/simple-rds.yaml
@@ -99,6 +99,14 @@ Parameters:
     Type: String
     Description: Password for database (eg. "postgres")
     NoEcho: true
+
+  DBSize:
+    Type: Number
+    Description: Initial storage size for database in GB
+
+  DBInstanceType:
+    Type: String
+    Description: Instance type for database (eg. "db.m7g.large")
 
   MaxAuxContainersPerAgent:
     Type: Number
@@ -453,7 +461,7 @@ Resources:
     Type: AWS::RDS::DBInstance
     DeletionPolicy: Delete
     Properties:
-      DBInstanceClass: db.m7g.large
+      DBInstanceClass: !Ref DBInstanceType
       Engine: postgres
       MasterUsername: postgres
       MasterUserPassword: !Ref DBPassword
@@ -462,7 +470,7 @@ Resources:
           Value: !Ref AWS::StackName
       VPCSecurityGroups:
         - !GetAtt DatabaseSecurityGroup.GroupId
-      AllocatedStorage: 20
+      AllocatedStorage: !Ref DBSize
       StorageType: gp2
 
   RetainedLogGroup: