feat: add aws govcloud support for det-deploy (#2049)

deploy/determined_deploy/aws/cli.py

@@ -7,7 +7,7 @@
 import boto3
 
 from determined_deploy.aws import aws, constants
-from determined_deploy.aws.deployment_types import base, secure, simple, vpc
+from determined_deploy.aws.deployment_types import base, govcloud, secure, simple, vpc
 
 
 def validate_spot_max_price() -> Callable:
@@ -282,6 +282,7 @@ def deploy_aws(args: argparse.Namespace) -> None:
         constants.deployment_types.VPC: vpc.VPC,
         constants.deployment_types.EFS: vpc.EFS,
         constants.deployment_types.FSX: vpc.FSx,
+        constants.deployment_types.GOVCLOUD: govcloud.Govcloud,
     }  # type: Dict[str, Union[Type[base.DeterminedDeployment]]]
 
     if args.deployment_type != constants.deployment_types.SIMPLE:
@@ -292,6 +293,13 @@ def deploy_aws(args: argparse.Namespace) -> None:
                 f"deployment-type={args.deployment_type}."
             )
 
+    if args.deployment_type == constants.deployment_types.GOVCLOUD:
+        if args.region not in ["us-gov-east-1", "us-gov-west-1"]:
+            raise ValueError(
+                "When deploying to GovCloud, set the region to either us-gov-east-1 "
+                "or us-gov-west-1."
+            )
+
     master_tls_cert = master_tls_key = ""
     if args.master_tls_cert:
         with open(args.master_tls_cert, "rb") as f:

deploy/determined_deploy/aws/constants.py

@@ -4,7 +4,8 @@ class deployment_types:
     VPC = "vpc"
     EFS = "efs"
     FSX = "fsx"
-    DEPLOYMENT_TYPES = [SIMPLE, SECURE, VPC, EFS, FSX]
+    GOVCLOUD = "govcloud"
+    DEPLOYMENT_TYPES = [SIMPLE, SECURE, VPC, EFS, FSX, GOVCLOUD]
 
 
 class defaults:
@@ -74,4 +75,6 @@ class misc:
         "us-east-1",
         "us-east-2",
         "us-west-2",
+        "us-gov-east-1",
+        "us-gov-west-1",
     ]

deploy/determined_deploy/aws/deployment_types/govcloud.py

@@ -0,0 +1,72 @@
+import boto3
+
+from determined_deploy.aws import aws, constants
+from determined_deploy.aws.deployment_types import base
+
+
+class Govcloud(base.DeterminedDeployment):
+    ssh_command = "SSH to master Instance: ssh -i <pem-file> ubuntu@{master_ip}"
+    det_ui = (
+        "Configure the Determined CLI: export DET_MASTER={master_ip}\n"
+        "View the Determined UI: http://{master_ip}:8080\n"
+        "View Logs at: https://{region}.console.amazonaws-us-gov.com/cloudwatch/home?"
+        "region={region}#logsV2:log-groups/log-group/{log_group}"
+    )
+
+    template = "govcloud.yaml"
+
+    template_parameter_keys = [
+        constants.cloudformation.ENABLE_CORS,
+        constants.cloudformation.MASTER_TLS_CERT,
+        constants.cloudformation.MASTER_TLS_KEY,
+        constants.cloudformation.MASTER_CERT_NAME,
+        constants.cloudformation.KEYPAIR,
+        constants.cloudformation.MASTER_INSTANCE_TYPE,
+        constants.cloudformation.CPU_AGENT_INSTANCE_TYPE,
+        constants.cloudformation.GPU_AGENT_INSTANCE_TYPE,
+        constants.cloudformation.INBOUND_CIDR,
+        constants.cloudformation.VERSION,
+        constants.cloudformation.DB_PASSWORD,
+        constants.cloudformation.MAX_IDLE_AGENT_PERIOD,
+        constants.cloudformation.MAX_AGENT_STARTING_PERIOD,
+        constants.cloudformation.MAX_CPU_CONTAINERS_PER_AGENT,
+        constants.cloudformation.MAX_DYNAMIC_AGENTS,
+        constants.cloudformation.SPOT_ENABLED,
+        constants.cloudformation.SPOT_MAX_PRICE,
+        constants.cloudformation.SUBNET_ID_KEY,
+        constants.cloudformation.SCHEDULER_TYPE,
+        constants.cloudformation.PREEMPTION_ENABLED,
+        constants.cloudformation.CPU_ENV_IMAGE,
+        constants.cloudformation.GPU_ENV_IMAGE,
+        constants.cloudformation.LOG_GROUP_PREFIX,
+        constants.cloudformation.RETAIN_LOG_GROUP,
+    ]
+
+    def deploy(self) -> None:
+        cfn_parameters = self.consolidate_parameters()
+        self.before_deploy_print()
+        with open(self.template_path) as f:
+            template = f.read()
+
+        aws.deploy_stack(
+            stack_name=self.parameters[constants.cloudformation.CLUSTER_ID],
+            template_body=template,
+            keypair=self.parameters[constants.cloudformation.KEYPAIR],
+            boto3_session=self.parameters[constants.cloudformation.BOTO3_SESSION],
+            parameters=cfn_parameters,
+        )
+        self.print_results(
+            self.parameters[constants.cloudformation.CLUSTER_ID],
+            self.parameters[constants.cloudformation.BOTO3_SESSION],
+        )
+
+    def print_results(self, stack_name: str, boto3_session: boto3.session.Session) -> None:
+        output = aws.get_output(stack_name, boto3_session)
+        master_ip = output[constants.cloudformation.DET_ADDRESS]
+        region = output[constants.cloudformation.REGION]
+        log_group = output[constants.cloudformation.LOG_GROUP]
+        ui_command = self.det_ui.format(master_ip=master_ip, region=region, log_group=log_group)
+        print(ui_command)
+
+        ssh_command = self.ssh_command.format(master_ip=master_ip)
+        print(ssh_command)