Revert "feat(sdk): make endpoint/app lifecycle operations CLI-only"

src/runpod_flash/cli/main.py

@@ -17,7 +17,6 @@
     update,
 )
 from .update_checker import start_background_check
-from ..core.cli_context import mark_cli_invocation
 
 
 def get_version() -> str:
@@ -79,11 +78,6 @@ def main(
     version: bool = typer.Option(False, "--version", "-v", help="Show version"),
 ):
     """Runpod Flash CLI - Distributed inference and serving framework."""
-    # Mark this process as a CLI invocation so resource lifecycle operations
-    # (deploy/undeploy/app/env management) are permitted. Direct SDK calls
-    # outside the CLI raise FlashUsageError. See core.cli_context.
-    mark_cli_invocation()
-
     if version:
         console.print(f"Runpod Flash CLI v{get_version()}")
         raise typer.Exit()

src/runpod_flash/core/exceptions.py

@@ -4,24 +4,7 @@
 """
 
 
-class FlashError(Exception):
-    """Base class for all runpod_flash domain errors.
-
-    Catch this to handle any flash-raised error; catch a subclass for a
-    specific failure mode.
-    """
-
-
-class FlashUsageError(FlashError):
-    """Raised when the SDK is used in a way that is not supported.
-
-    Currently raised when a CLI-managed lifecycle operation (deploy, undeploy,
-    app/environment management) is invoked directly from the SDK instead of
-    through the flash CLI. The message names the equivalent CLI command.
-    """
-
-
-class RunpodAPIKeyError(FlashError):
+class RunpodAPIKeyError(Exception):
     """Raised when RUNPOD_API_KEY environment variable is missing or invalid.
 
     This exception provides helpful guidance on how to obtain and configure

src/runpod_flash/core/resources/app.py

@@ -5,7 +5,6 @@
 import logging
 
 from ..api.runpod import RunpodGraphQLClient
-from ..cli_context import cli_only
 
 from .constants import (
     TARBALL_CONTENT_TYPE,
@@ -326,7 +325,6 @@ async def _hydrate(self) -> None:
             self._hydrated = True
             return
 
-    @cli_only("flash env create")
     async def create_environment(self, environment_name: str) -> Dict[str, Any]:
         """Create an environment within an app.
 
@@ -364,7 +362,6 @@ async def _get_tarball_upload_url(self, tarball_size: int) -> Dict[str, str]:
                 {"flashAppId": self.id, "tarballSize": tarball_size}
             )
 
-    @cli_only("flash deploy")
     async def deploy_build_to_environment(
         self,
         build_id: str,
@@ -497,14 +494,12 @@ async def from_name(cls, app_name: str) -> "FlashApp":
         return cls(app_name, id=result["id"], eager_hydrate=False)
 
     @classmethod
-    @cli_only("flash app create")
     async def create(cls, app_name: str) -> "FlashApp":
         async with RunpodGraphQLClient() as client:
             result = await client.create_flash_app({"name": app_name})
         return cls(app_name, id=result["id"], eager_hydrate=False)
 
     @classmethod
-    @cli_only("flash app create")
     async def get_or_create(cls, app_name: str) -> "FlashApp":
         try:
             return await cls.from_name(app_name)
@@ -514,7 +509,6 @@ async def get_or_create(cls, app_name: str) -> "FlashApp":
             return cls(app_name, id=result["id"], eager_hydrate=False)
 
     @classmethod
-    @cli_only("flash app create")
     async def create_environment_and_app(
         cls, app_name: str, environment_name: str
     ) -> Tuple["FlashApp", Dict]:
@@ -528,7 +522,6 @@ async def list(cls):
             return await client.list_flash_apps()
 
     @classmethod
-    @cli_only("flash app delete")
     async def delete(
         cls, app_name: Optional[str] = None, app_id: Optional[str] = None
     ) -> bool:
@@ -547,7 +540,6 @@ async def delete(
             result = await client.delete_flash_app(app_id)
         return result.get("success", False)
 
-    @cli_only("flash env delete")
     async def delete_environment(self, environment_name: str) -> bool:
         """Delete an environment from this flash app.
 

src/runpod_flash/core/resources/load_balancer_sls_resource.py

@@ -45,8 +45,7 @@ class LoadBalancerSlsResource(ServerlessResource):
             workersMin=1,
             workersMax=3,
         )
-
-    Deploy via the CLI (`flash deploy`); SDK lifecycle calls are CLI-only.
+        await lb.deploy()
     """
 
     # Override default type to LB
@@ -210,8 +209,7 @@ class CpuLoadBalancerSlsResource(CpuEndpointMixin, LoadBalancerSlsResource):
             workersMin=1,
             workersMax=3,
         )
-
-    Deploy via the CLI (`flash deploy`); SDK lifecycle calls are CLI-only.
+        await cpu_lb.deploy()
     """
 
     instanceIds: Optional[List[CpuInstanceType]] = [CpuInstanceType.CPU3G_2_8]

src/runpod_flash/core/resources/network_volume.py

@@ -15,7 +15,6 @@
 from ..urls import RUNPOD_CONSOLE_URL
 from .base import DeployableResource
 from .resource_manager import ResourceManager
-from ..cli_context import cli_only
 
 log = logging.getLogger(__name__)
 
@@ -215,7 +214,6 @@ async def _do_deploy(self) -> "DeployableResource":
             log.error(f"{self} failed to deploy: {e}")
             raise
 
-    @cli_only("flash deploy")
     async def deploy(self) -> "DeployableResource":
         resource_manager = ResourceManager()
         resource = await resource_manager.get_or_deploy_resource(self)

src/runpod_flash/core/resources/serverless.py

@@ -20,7 +20,6 @@
 from runpod.endpoint.runner import Job
 
 from ..api.runpod import RunpodGraphQLClient
-from ..cli_context import cli_only
 from ..exceptions import RunpodAPIKeyError
 from ..utils.backoff import get_backoff_delay
 from .base import DeployableResource
@@ -1115,7 +1114,6 @@ async def _do_deploy(self) -> "DeployableResource":
             log.error(f"{self} failed to deploy: {e}")
             raise
 
-    @cli_only("flash deploy")
     async def update(self, new_config: "ServerlessResource") -> "ServerlessResource":
         """Update existing endpoint with new configuration.
 
@@ -1319,7 +1317,6 @@ def _has_structural_changes(self, new_config: "ServerlessResource") -> bool:
 
         return False
 
-    @cli_only("flash deploy")
     async def deploy(self) -> "DeployableResource":
         resource_manager = ResourceManager()
         resource = await resource_manager.get_or_deploy_resource(self)
@@ -1372,7 +1369,6 @@ async def _do_undeploy(self) -> bool:
 
             return False
 
-    @cli_only("flash undeploy")
     async def undeploy(self) -> Dict[str, Any]:
         resource_manager = ResourceManager()
         result = await resource_manager.undeploy_resource(self.resource_id)

tests/conftest.py

@@ -17,7 +17,6 @@
 
 import pytest
 
-from runpod_flash.core.cli_context import _invoked_by_cli, allow_lifecycle_operations
 from runpod_flash.core.resources.resource_manager import ResourceManager
 from runpod_flash.core.utils.singleton import SingletonMixin
 
@@ -30,36 +29,6 @@ def pytest_configure(config):
     """
     # This hook is called early in pytest initialization
     # xdist will check for this during test distribution
-    config.addinivalue_line(
-        "markers",
-        "no_cli_context: run with the flash CLI context disabled so guarded "
-        "lifecycle methods raise FlashUsageError (used by guard tests).",
-    )
-
-
-@pytest.fixture(autouse=True)
-def _flash_cli_context(request):
-    """Permit CLI-managed lifecycle operations during tests.
-
-    Lifecycle methods (deploy/undeploy/app/env management) are restricted to
-    flash CLI invocation. Most tests drive them directly, so enable the context
-    by default. Guard tests opt out with @pytest.mark.no_cli_context.
-
-    The opt-out branch actively forces the context to False (rather than merely
-    skipping the enable) and restores it afterwards. ``mark_cli_invocation()`` is
-    set-and-leave, so a CLI test that ran earlier in the same worker could leave
-    the ContextVar True; forcing False here makes guard tests independent of test
-    order instead of relying on each guard module to reset the var itself.
-    """
-    if request.node.get_closest_marker("no_cli_context"):
-        token = _invoked_by_cli.set(False)
-        try:
-            yield
-        finally:
-            _invoked_by_cli.reset(token)
-    else:
-        with allow_lifecycle_operations():
-            yield
 
 
 def pytest_collection_modifyitems(config, items):