lightning-sdk 0.2.7__py3-none-any.whl → 0.2.9__py3-none-any.whl

lightning_sdk/__init__.py

@@ -31,6 +31,6 @@ __all__ = [
     "User",
 ]
 
-__version__ = "0.2.7"
+__version__ = "0.2.9"
 _check_version_and_prompt_upgrade(__version__)
 _set_tqdm_envvars_noninteractive()

lightning_sdk/api/cluster_api.py

@@ -0,0 +1,22 @@
+from lightning_sdk.lightning_cloud.openapi import Externalv1Cluster
+from lightning_sdk.lightning_cloud.rest_client import LightningClient
+
+
+class ClusterApi:
+    """Internal API client for API requests to cluster endpoints."""
+
+    def __init__(self) -> None:
+        self._client = LightningClient(max_tries=7)
+
+    def get_cluster(self, cluster_id: str, project_id: str, org_id: str) -> Externalv1Cluster:
+        """Gets the cluster from given params cluster_id, project_id and owner.
+
+        :param cluster_id: cluster ID test
+        :param project_id: the project the cluster is supposed to be associated with
+        :param org_id: The owning org of this cluster
+        :return:
+        """
+        res = self._client.cluster_service_get_cluster(id=cluster_id, org_id=org_id, project_id=project_id)
+        if not res:
+            raise ValueError(f"Cluster {cluster_id} does not exist")
+        return res

lightning_sdk/api/lit_container_api.py

@@ -7,7 +7,6 @@ import requests
 from rich.console import Console
 
 from lightning_sdk.api.utils import _get_registry_url
-from lightning_sdk.lightning_cloud.env import LIGHTNING_CLOUD_URL
 from lightning_sdk.lightning_cloud.openapi.models import V1DeleteLitRepositoryResponse
 from lightning_sdk.lightning_cloud.rest_client import LightningClient
 from lightning_sdk.teamspace import Teamspace
@@ -103,11 +102,14 @@ class LitContainerApi:
         """Lists containers of the project ID.
 
         :param project_id: The non-human readable project ID used internally to identify projects.
+        :param cloud_account: The cluster ID of the cloud account. If None, will use the default cluster.
         :return:
         """
-        project = self._client.lit_registry_service_get_lit_project_registry(
-            project_id, cluster_id=cloud_account
-        )  # cloud account on the CLI is cluster_id
+        if cloud_account is None:
+            project = self._client.lit_registry_service_get_lit_project_registry(project_id)
+        else:
+            project = self._client.lit_registry_service_get_lit_project_registry(project_id, cluster_id=cloud_account)
+
         return project.repositories
 
     def delete_container(self, project_id: str, container: str) -> V1DeleteLitRepositoryResponse:
@@ -138,7 +140,7 @@ class LitContainerApi:
             Named cloud-account in the CLI options.
         :param platform: If empty will be linux/amd64. This is important because our entire deployment infra runs on
             linux/amd64. Will show user a warning otherwise.
-        :return: Generator[dict, None, None]
+        :return: Generator[dict, None, dict]
         """
         try:
             self._docker_client.images.get(f"{container}:{tag}")
@@ -164,13 +166,6 @@ class LitContainerApi:
             raise ValueError(f"Could not tag container {container}:{tag} with {repository}:{tag}")
         yield from self._push_with_retry(repository, tag=tag)
 
-        yield {
-            "finish": True,
-            "url": f"{LIGHTNING_CLOUD_URL}/{teamspace.owner.name}/{teamspace.name}/containers/{container_basename}"
-            f"{f'?clusterId={cloud_account}' if cloud_account is not None else ''}",
-            "repository": repository,
-        }
-
     def _push_with_retry(self, repository: str, tag: str, max_retries: int = 3) -> Iterator[Dict[str, Any]]:
         def is_auth_error(error_msg: str) -> bool:
             auth_errors = ["unauthorized", "authentication required", "unauth"]

lightning_sdk/api/teamspace_api.py

@@ -177,6 +177,7 @@ class TeamspaceApi:
     def create_model(
         self,
         name: str,
+        version: Optional[str],
         metadata: Dict[str, str],
         private: bool,
         teamspace_id: str,
@@ -191,7 +192,7 @@ class TeamspaceApi:
             )
         assert len(models) == 1, "Multiple models with the same name found"
         return self.models.models_store_create_model_version(
-            body=ModelIdVersionsBody(cluster_id=cloud_account),
+            body=ModelIdVersionsBody(cluster_id=cloud_account, version=version),
             project_id=teamspace_id,
             model_id=models[0].id,
         )

lightning_sdk/cli/clusters_menu.py

@@ -0,0 +1,46 @@
+import sys
+from typing import List
+
+from rich.console import Console
+from simple_term_menu import TerminalMenu
+
+from lightning_sdk import Teamspace
+from lightning_sdk.api.cluster_api import ClusterApi
+from lightning_sdk.lightning_cloud.openapi import Externalv1Cluster, V1ProjectClusterBinding
+
+
+class _ClustersMenu:
+    def _get_cluster_from_interactive_menu(self, possible_clusters: List[V1ProjectClusterBinding]) -> str:
+        terminal_menu = self._prepare_terminal_menu_teamspaces([cluster.cluster_id for cluster in possible_clusters])
+        terminal_menu.show()
+
+        return possible_clusters[terminal_menu.chosen_menu_index].cluster_id
+
+    @staticmethod
+    def _prepare_terminal_menu_teamspaces(cluster_ids: List[str]) -> TerminalMenu:
+        title = "Please select a cluster from the following:"
+
+        return TerminalMenu(cluster_ids, title=title, clear_menu_on_exit=True)
+
+    def _resolve_cluster(self, teamspace: Teamspace) -> Externalv1Cluster:
+        selected_cluster_id = None
+        console = Console()
+        try:
+            selected_cluster_id = self._get_cluster_from_interactive_menu(
+                possible_clusters=teamspace.cloud_account_objs
+            )
+            cluster_api = ClusterApi()
+
+            return cluster_api.get_cluster(
+                cluster_id=selected_cluster_id, org_id=teamspace.owner.id, project_id=teamspace.id
+            )
+        except KeyboardInterrupt:
+            console.print("Operation cancelled by user")
+            sys.exit(0)
+
+        except Exception:
+            console.print(
+                f"[red]Could not find the given Cluster:[/red] {selected_cluster_id}. "
+                "Please contact Lightning AI directly to resolve this issue."
+            )
+            sys.exit(1)

lightning_sdk/cli/list.py

@@ -7,8 +7,9 @@ from rich.table import Table
 from typing_extensions import Literal
 
 from lightning_sdk import Job, Machine, Studio, Teamspace
+from lightning_sdk.cli.clusters_menu import _ClustersMenu
 from lightning_sdk.cli.teamspace_menu import _TeamspacesMenu
-from lightning_sdk.lightning_cloud.openapi import V1MultiMachineJob
+from lightning_sdk.lightning_cloud.openapi import V1ClusterType, V1MultiMachineJob
 from lightning_sdk.lit_container import LitContainer
 from lightning_sdk.utils.resolve import _get_authed_user
 
@@ -232,18 +233,37 @@ def mmts(
         "If not provided, can be selected in an interactive menu."
     ),
 )
-def containers(teamspace: Optional[str] = None) -> None:
+@click.option(
+    "--cloud-account",
+    "--cloud_account",  # The UI will present the above variant, using this as a secondary to be consistent w/ models
+    default=None,
+    help="The name of the cloud account where containers are stored in.",
+)
+def containers(teamspace: Optional[str] = None, cloud_account: Optional[str] = None) -> None:
     """Display the list of available containers."""
     api = LitContainer()
     menu = _TeamspacesMenu()
+    clusters_menu = _ClustersMenu()
     resolved_teamspace = menu._resolve_teamspace(teamspace=teamspace)
-    result = api.list_containers(teamspace=resolved_teamspace.name, org=resolved_teamspace.owner.name)
+
+    if not cloud_account:
+        cloud_account_obj = clusters_menu._resolve_cluster(resolved_teamspace)
+        cloud_account = "" if cloud_account_obj.spec.cluster_type == V1ClusterType.GLOBAL else cloud_account_obj.id
+
+    result = api.list_containers(
+        teamspace=resolved_teamspace.name, org=resolved_teamspace.owner.name, cloud_account=cloud_account
+    )
+
+    if not result:
+        return
+
     table = Table(pad_edge=True, box=None)
     table.add_column("REPOSITORY")
-    table.add_column("IMAGE ID")
+    table.add_column("CLOUD ACCOUNT")
+    table.add_column("LATEST TAG")
     table.add_column("CREATED")
     for repo in result:
-        table.add_row(repo["REPOSITORY"], repo["IMAGE ID"], repo["CREATED"])
+        table.add_row(repo["REPOSITORY"], repo["CLOUD ACCOUNT"], repo["LATEST TAG"], repo["CREATED"])
     Console().print(table)
 
 

lightning_sdk/cli/serve.py

@@ -1,8 +1,11 @@
 import os
 import subprocess
+import time
+import webbrowser
 from datetime import datetime
 from pathlib import Path
 from typing import Optional, Union
+from urllib.parse import urlencode
 
 import click
 from rich.console import Console
@@ -10,24 +13,37 @@ from rich.progress import Progress, SpinnerColumn, TextColumn, TimeElapsedColumn
 from rich.prompt import Confirm
 
 from lightning_sdk import Machine, Teamspace
+from lightning_sdk.api import UserApi
 from lightning_sdk.api.lit_container_api import LitContainerApi
 from lightning_sdk.cli.teamspace_menu import _TeamspacesMenu
-from lightning_sdk.serve import _LitServeDeployer, authenticate
+from lightning_sdk.lightning_cloud import env
+from lightning_sdk.lightning_cloud.login import Auth, AuthServer
+from lightning_sdk.serve import _LitServeDeployer
+from lightning_sdk.utils.resolve import _get_authed_user, _resolve_teamspace
 
 _MACHINE_VALUES = tuple([machine.name for machine in Machine.__dict__.values() if isinstance(machine, Machine)])
 
 
-@click.group("serve")
+class _ServeGroup(click.Group):
+    def parse_args(self, ctx: click.Context, args: list) -> click.Group:
+        # Check if first arg is a file path and not a command name
+        if args and os.path.exists(args[0]) and args[0] not in self.commands:
+            # Insert the 'api' command before the file path
+            args.insert(0, "api")
+        return super().parse_args(ctx, args)
+
+
+@click.group("serve", cls=_ServeGroup)
 def serve() -> None:
     """Serve a LitServe model.
 
     Example:
-        lightning serve api server.py  # serve locally
+        lightning serve server.py  # deploy to the cloud
 
     Example:
-        lightning serve api server.py --cloud --name litserve-api  # deploy to the cloud
+        lightning serve server.py --local  # serve locally
 
-    You can deploy the API to the cloud by running `lightning serve api server.py --cloud`.
+    You can deploy the API to the cloud by running `lightning serve server.py`.
     This will build a docker container for the server.py script and deploy it to the Lightning AI platform.
     """
 
@@ -42,11 +58,11 @@ def serve() -> None:
     help="Generate a client for the model",
 )
 @click.option(
-    "--cloud",
+    "--local",
     is_flag=True,
     default=False,
     flag_value=True,
-    help="Deploy the model to the Lightning AI platform",
+    help="Run the model locally",
 )
 @click.option("--name", default=None, help="Name of the deployed API (e.g., 'classification-api', 'Llama-api')")
 @click.option(
@@ -107,7 +123,7 @@ def serve() -> None:
 def api(
     script_path: str,
     easy: bool,
-    cloud: bool,
+    local: bool,
     name: Optional[str],
     non_interactive: bool,
     machine: str,
@@ -126,7 +142,7 @@ def api(
     return api_impl(
         script_path=script_path,
         easy=easy,
-        cloud=cloud,
+        local=local,
         repository=name,
         non_interactive=non_interactive,
         machine=machine,
@@ -146,7 +162,7 @@ def api(
 def api_impl(
     script_path: Union[str, Path],
     easy: bool = False,
-    cloud: bool = False,
+    local: bool = False,
     repository: [str] = None,
     tag: Optional[str] = None,
     non_interactive: bool = False,
@@ -176,7 +192,7 @@ def api_impl(
         timestr = datetime.now().strftime("%b-%d-%H_%M")
         repository = f"litserve-{timestr}".lower()
 
-    if cloud:
+    if not local:
         repository = repository or "litserve-model"
         machine = Machine.from_str(machine)
         return _handle_cloud(
@@ -209,6 +225,69 @@ def api_impl(
         raise RuntimeError(error_msg) from None
 
 
+class _AuthServer(AuthServer):
+    def get_auth_url(self, port: int) -> str:
+        redirect_uri = f"http://localhost:{port}/login-complete"
+        params = urlencode({"redirectTo": redirect_uri, "inviteCode": "litserve"})
+        return f"{env.LIGHTNING_CLOUD_URL}/sign-in?{params}"
+
+
+class _Auth(Auth):
+    def __init__(self, shall_confirm: bool = False) -> None:
+        super().__init__()
+        self._shall_confirm = shall_confirm
+
+    def _run_server(self) -> None:
+        if self._shall_confirm:
+            proceed = Confirm.ask(
+                "Authenticating with Lightning AI. This will open a browser window. Continue?", default=True
+            )
+            if not proceed:
+                raise RuntimeError(
+                    "Login cancelled. Please login to Lightning AI to deploy the API."
+                    " Run `lightning login` to login."
+                ) from None
+        print("Opening browser for authentication...")
+        print("Please come back to the terminal after logging in.")
+        time.sleep(3)
+        _AuthServer().login_with_browser(self)
+
+
+def authenticate(shall_confirm: bool = True) -> None:
+    auth = _Auth(shall_confirm)
+    auth.authenticate()
+
+
+def select_teamspace(teamspace: Optional[str], org: Optional[str], user: Optional[str]) -> Teamspace:
+    if teamspace is None:
+        user = _get_authed_user()
+        menu = _TeamspacesMenu()
+        possible_teamspaces = menu._get_possible_teamspaces(user)
+        if len(possible_teamspaces) == 1:
+            name = next(iter(possible_teamspaces.values()))["name"]
+            return Teamspace(name=name, org=org, user=user)
+
+        return menu._resolve_teamspace(teamspace)
+
+    return _resolve_teamspace(teamspace=teamspace, org=org, user=user)
+
+
+def poll_verified_status() -> bool:
+    """Polls the verified status of the user until it is True or a timeout occurs."""
+    user_api = UserApi()
+    user = _get_authed_user()
+    start_time = datetime.now()
+    timeout = 600  # 10 minutes
+    while True:
+        user_resp = user_api.get_user(name=user.name)
+        if user_resp.status.verified:
+            return True
+        if (datetime.now() - start_time).total_seconds() > timeout:
+            break
+        time.sleep(5)
+    return False
+
+
 def _handle_cloud(
     script_path: Union[str, Path],
     console: Console,
@@ -233,16 +312,8 @@ def _handle_cloud(
     if non_interactive:
         console.print("[italic]non-interactive[/italic] mode enabled, skipping confirmation prompts", style="blue")
 
-    # Authenticate with LitServe affiliate
-    authenticate()
-    if teamspace is None:
-        menu = _TeamspacesMenu()
-        resolved_teamspace = menu._resolve_teamspace(teamspace)
-    else:
-        resolved_teamspace = Teamspace(name=teamspace, org=org, user=user)
-
     port = port or 8000
-    ls_deployer = _LitServeDeployer(name=deployment_name, teamspace=resolved_teamspace)
+    ls_deployer = _LitServeDeployer(name=deployment_name, teamspace=None)
     path = ls_deployer.dockerize_api(script_path, port=port, gpu=not machine.is_cpu(), tag=tag, print_success=False)
 
     console.print(f"\nPlease review the Dockerfile at [u]{path}[/u] and make sure it is correct.", style="bold")
@@ -251,6 +322,36 @@ def _handle_cloud(
         console.print("Please fix the Dockerfile and try again.", style="red")
         return
 
+    with Progress(
+        SpinnerColumn(),
+        TextColumn("[progress.description]{task.description}"),
+        TimeElapsedColumn(),
+        console=console,
+        transient=True,
+    ) as progress:
+        try:
+            # Build the container
+            build_task = progress.add_task("Building Docker image", total=None)
+            for line in ls_deployer.build_container(path, repository, tag):
+                console.print(line.strip())
+                progress.update(build_task, advance=1)
+            progress.update(build_task, description="[green]Build completed![/green]", completed=1.0)
+            progress.remove_task(build_task)
+
+        except Exception as e:
+            console.print(f"❌ Deployment failed: {e}", style="red")
+            return
+
+    # Push the container to the registry
+    console.print("\nPushing container to registry. It may take a while...", style="bold")
+    # Authenticate with LitServe affiliate
+    authenticate(shall_confirm=not non_interactive)
+    resolved_teamspace = select_teamspace(teamspace, org, user)
+    verified = poll_verified_status()
+    if not verified:
+        console.print("❌ Verify phone number to continue. Visit lightning.ai.", style="red")
+        return
+
     # list containers to create the project if it doesn't exist
     lit_cr = LitContainerApi()
     lit_cr.list_containers(resolved_teamspace.id, cloud_account=cloud_account)
@@ -263,17 +364,21 @@ def _handle_cloud(
         transient=True,
     ) as progress:
         try:
-            # TODO: @aniketmaurya improve the console prints here
-            ls_deployer.build_container(path, repository, tag, console, progress)
-            push_status = ls_deployer.push_container(
-                repository, tag, resolved_teamspace, lit_cr, progress, cloud_account=cloud_account
-            )
+            push_task = progress.add_task("Pushing to registry", total=None)
+            push_status = {}
+            for line in ls_deployer.push_container(
+                repository, tag, resolved_teamspace, lit_cr, cloud_account=cloud_account
+            ):
+                push_status = line
+                progress.update(push_task, advance=1)
+                if not ("Pushing" in line["status"] or "Waiting" in line["status"]):
+                    console.print(line["status"])
+            progress.update(push_task, description="[green]Push completed![/green]")
         except Exception as e:
             console.print(f"❌ Deployment failed: {e}", style="red")
             return
     console.print(f"\n✅ Image pushed to {repository}:{tag}")
-    console.print(f"🔗 You can access the container at: [i]{push_status.get('url')}[/i]")
-    image = push_status.get("repository")
+    image = push_status.get("image")
 
     deployment_status = ls_deployer.run_on_cloud(
         deployment_name=deployment_name,
@@ -290,3 +395,4 @@ def _handle_cloud(
         include_credentials=include_credentials,
     )
     console.print(f"🚀 Deployment started, access at [i]{deployment_status.get('url')}[/i]")
+    webbrowser.open(deployment_status.get("url"))

lightning_sdk/lightning_cloud/openapi/__init__.py

@@ -83,6 +83,7 @@ from lightning_sdk.lightning_cloud.openapi.models.cloud_space_id_versionpublicat
 from lightning_sdk.lightning_cloud.openapi.models.cloud_space_id_versions_body import CloudSpaceIdVersionsBody
 from lightning_sdk.lightning_cloud.openapi.models.cloudspace_id_metric_body import CloudspaceIdMetricBody
 from lightning_sdk.lightning_cloud.openapi.models.cloudspace_id_runs_body import CloudspaceIdRunsBody
+from lightning_sdk.lightning_cloud.openapi.models.cloudspace_id_systemmetrics_body import CloudspaceIdSystemmetricsBody
 from lightning_sdk.lightning_cloud.openapi.models.cloudspaces_id_body import CloudspacesIdBody
 from lightning_sdk.lightning_cloud.openapi.models.cluster_id_capacityblock_body import ClusterIdCapacityblockBody
 from lightning_sdk.lightning_cloud.openapi.models.cluster_id_capacityreservations_body import ClusterIdCapacityreservationsBody
@@ -495,6 +496,7 @@ from lightning_sdk.lightning_cloud.openapi.models.v1_get_agent_job_env_response
 from lightning_sdk.lightning_cloud.openapi.models.v1_get_agent_job_logs_metadata_response import V1GetAgentJobLogsMetadataResponse
 from lightning_sdk.lightning_cloud.openapi.models.v1_get_artifacts_page_response import V1GetArtifactsPageResponse
 from lightning_sdk.lightning_cloud.openapi.models.v1_get_cloud_space_instance_status_response import V1GetCloudSpaceInstanceStatusResponse
+from lightning_sdk.lightning_cloud.openapi.models.v1_get_cloud_space_instance_system_metrics_aggregate_response import V1GetCloudSpaceInstanceSystemMetricsAggregateResponse
 from lightning_sdk.lightning_cloud.openapi.models.v1_get_cloud_space_size_response import V1GetCloudSpaceSizeResponse
 from lightning_sdk.lightning_cloud.openapi.models.v1_get_cluster_accelerator_demand_response import V1GetClusterAcceleratorDemandResponse
 from lightning_sdk.lightning_cloud.openapi.models.v1_get_cluster_credentials_response import V1GetClusterCredentialsResponse
@@ -775,6 +777,7 @@ from lightning_sdk.lightning_cloud.openapi.models.v1_refresh_response import V1R
 from lightning_sdk.lightning_cloud.openapi.models.v1_region_state import V1RegionState
 from lightning_sdk.lightning_cloud.openapi.models.v1_regional_load_balancer import V1RegionalLoadBalancer
 from lightning_sdk.lightning_cloud.openapi.models.v1_report_cloud_space_instance_stop_at_response import V1ReportCloudSpaceInstanceStopAtResponse
+from lightning_sdk.lightning_cloud.openapi.models.v1_report_cloud_space_instance_system_metrics_response import V1ReportCloudSpaceInstanceSystemMetricsResponse
 from lightning_sdk.lightning_cloud.openapi.models.v1_report_logs_activity_response import V1ReportLogsActivityResponse
 from lightning_sdk.lightning_cloud.openapi.models.v1_report_restart_timings_response import V1ReportRestartTimingsResponse
 from lightning_sdk.lightning_cloud.openapi.models.v1_request_cluster_access_request import V1RequestClusterAccessRequest
@@ -840,6 +843,7 @@ from lightning_sdk.lightning_cloud.openapi.models.v1_subnet_spec import V1Subnet
 from lightning_sdk.lightning_cloud.openapi.models.v1_switch_cloud_space_instance_response import V1SwitchCloudSpaceInstanceResponse
 from lightning_sdk.lightning_cloud.openapi.models.v1_system_info import V1SystemInfo
 from lightning_sdk.lightning_cloud.openapi.models.v1_system_metrics import V1SystemMetrics
+from lightning_sdk.lightning_cloud.openapi.models.v1_system_metrics_aggregated import V1SystemMetricsAggregated
 from lightning_sdk.lightning_cloud.openapi.models.v1_system_metrics_list import V1SystemMetricsList
 from lightning_sdk.lightning_cloud.openapi.models.v1_telemetry import V1Telemetry
 from lightning_sdk.lightning_cloud.openapi.models.v1_timestamp_code_telemetry import V1TimestampCodeTelemetry
@@ -903,6 +907,7 @@ from lightning_sdk.lightning_cloud.openapi.models.v1_verify_verification_respons
 from lightning_sdk.lightning_cloud.openapi.models.v1_voltage_park_direct_v1 import V1VoltageParkDirectV1
 from lightning_sdk.lightning_cloud.openapi.models.v1_volume import V1Volume
 from lightning_sdk.lightning_cloud.openapi.models.v1_vultr_direct_v1 import V1VultrDirectV1
+from lightning_sdk.lightning_cloud.openapi.models.v1_weka_data_connection import V1WekaDataConnection
 from lightning_sdk.lightning_cloud.openapi.models.v1_work import V1Work
 from lightning_sdk.lightning_cloud.openapi.models.validate import Validate
 from lightning_sdk.lightning_cloud.openapi.models.validateautojoindomain_domain_body import ValidateautojoindomainDomainBody