Allow skypilot to configure step or run full pipeline in one VM (zenm…

…l-io#2276)

* Allow skypilot to configure step or run  full pipeline in one VM

* Chnage behaviour of using step based resources

* Update src/zenml/integrations/skypilot/orchestrators/skypilot_base_vm_orchestrator.py

Co-authored-by: Alex Strick van Linschoten <strickvl@users.noreply.github.com>

* Refactor SkypilotBaseOrchestrator to support step-based VM configuration

* Update docs/book/stacks-and-components/component-guide/orchestrators/skypilot-vm.md

Co-authored-by: Hamza Tahir <hamza@zenml.io>

---------

Co-authored-by: Alex Strick van Linschoten <strickvl@users.noreply.github.com>
Co-authored-by: Hamza Tahir <hamza@zenml.io>

docs/book/stacks-and-components/component-guide/orchestrators/skypilot-vm.md

@@ -40,6 +40,12 @@ on-demand and managed spot VMs. While you can select the VM type you want to use
 also includes an optimizer that automatically selects the cheapest VM/zone/region/cloud for your workloads.
 Finally, the orchestrator includes an autostop feature that cleans up idle clusters, preventing unnecessary cloud costs.
 
+{% hint style="info" %}
+You can configure the SkyPilot VM Orchestrator to use a specific VM type, and 
+resources for each step of your pipeline can be configured individually.
+Read more about how to configure step-specific resources [here](#configuring-step-specific-resources).
+{% endhint %}
+
 {% hint style="warning" %}
 The SkyPilot VM Orchestrator does not currently support the ability to [schedule pipelines runs](/docs/book/user-guide/advanced-guide/pipelining-features/schedule-pipeline-runs.md)
 {% endhint %}
@@ -378,7 +384,11 @@ One of the key features of the SkyPilot VM Orchestrator is the ability to run ea
 
 The SkyPilot VM Orchestrator allows you to configure resources for each step individually. This means you can specify different VM types, CPU and memory requirements, and even use spot instances for certain steps while using on-demand instances for others.
 
-To configure step-specific resources, you can pass a `SkypilotBaseOrchestratorSettings` object to the `settings` parameter of the `@step` decorator. This object allows you to define various attributes such as `instance_type`, `cpus`, `memory`, `use_spot`, `region`, and more.
+If no step-specific settings are specified, the orchestrator will use the resources specified in the orchestrator settings for each step and run the entire pipeline in one VM. If step-specific settings are specified, an orchestrator VM will be spun up first, which will subsequently spin out new VMs dependant on the step settings. You can disable this behavior by setting the `disable_step_based_settings` parameter to `True` in the orchestrator configuration, using the following command:
+
+```shell
+zenml orchestrator update <ORCHESTRATOR_NAME> --disable_step_based_settings=True
+```
 
 Here's an example of how to configure specific resources for a step for the AWS cloud:
 

src/zenml/integrations/skypilot/flavors/skypilot_orchestrator_base_vm_config.py

@@ -113,7 +113,16 @@ class SkypilotBaseOrchestratorSettings(BaseSettings):
 class SkypilotBaseOrchestratorConfig(  # type: ignore[misc] # https://github.com/pydantic/pydantic/issues/4173
     BaseOrchestratorConfig, SkypilotBaseOrchestratorSettings
 ):
-    """Skypilot orchestrator base config."""
+    """Skypilot orchestrator base config.
+
+    Attributes:
+        disable_step_based_settings: whether to disable step-based settings.
+            If True, the orchestrator will run all steps with the pipeline
+            settings in one single VM. If False, the orchestrator will run
+            each step with its own settings in separate VMs if provided.
+    """
+
+    disable_step_based_settings: bool = False
 
     @property
     def is_local(self) -> bool:

src/zenml/integrations/skypilot/orchestrators/skypilot_base_vm_orchestrator.py

@@ -17,12 +17,15 @@
 import re
 from abc import abstractmethod
 from typing import TYPE_CHECKING, Any, Dict, Optional, Tuple, cast
+from uuid import uuid4
 
 import sky
 
+from zenml.entrypoints import PipelineEntrypointConfiguration
 from zenml.enums import StackComponentType
 from zenml.environment import Environment
 from zenml.integrations.skypilot.flavors.skypilot_orchestrator_base_vm_config import (
+    SkypilotBaseOrchestratorConfig,
     SkypilotBaseOrchestratorSettings,
 )
 from zenml.integrations.skypilot.orchestrators.skypilot_orchestrator_entrypoint_configuration import (
@@ -109,6 +112,15 @@ def get_orchestrator_run_id(self) -> str:
                 f"{ENV_ZENML_SKYPILOT_ORCHESTRATOR_RUN_ID}."
             )
 
+    @property
+    def config(self) -> SkypilotBaseOrchestratorConfig:
+        """Returns the `SkypilotBaseOrchestratorConfig` config.
+
+        Returns:
+            The configuration.
+        """
+        return cast(SkypilotBaseOrchestratorConfig, self._config)
+
     @property
     @abstractmethod
     def cloud(self) -> sky.clouds.Cloud:
@@ -167,6 +179,12 @@ def prepare_or_run_pipeline(
                 "and the pipeline will be run immediately."
             )
 
+        # Set up some variables for configuration
+        orchestrator_run_id = str(uuid4())
+        environment[
+            ENV_ZENML_SKYPILOT_ORCHESTRATOR_RUN_ID
+        ] = orchestrator_run_id
+
         settings = cast(
             SkypilotBaseOrchestratorSettings,
             self.get_settings(deployment),
@@ -189,14 +207,47 @@ def prepare_or_run_pipeline(
                 deployment=deployment, step_name=pipeline_step_name
             )
 
-        # Build entrypoint command and args for the orchestrator pod.
-        # This will internally also build the command/args for all step pods.
-        command = SkypilotOrchestratorEntrypointConfiguration.get_entrypoint_command()
+        different_settings_found = False
+
+        if not self.config.disable_step_based_settings:
+            for _, step in deployment.step_configurations.items():
+                step_settings = cast(
+                    SkypilotBaseOrchestratorSettings,
+                    self.get_settings(step),
+                )
+                if step_settings != settings:
+                    different_settings_found = True
+                    logger.info(
+                        "At least one step has different settings than the "
+                        "pipeline. The step with different settings will be "
+                        "run in a separate VM.\n"
+                        "You can configure the orchestrator to disable this "
+                        "behavior by updating the `disable_step_based_settings` "
+                        "in your orchestrator configuration "
+                        "by running the following command: "
+                        "`zenml orchestrator update --disable-step-based-settings=True`"
+                    )
+                    break
+
+        # Decide which configuration to use based on whether different settings were found
+        if (
+            not self.config.disable_step_based_settings
+            and different_settings_found
+        ):
+            # Run each step in a separate VM using SkypilotOrchestratorEntrypointConfiguration
+            command = SkypilotOrchestratorEntrypointConfiguration.get_entrypoint_command()
+            args = SkypilotOrchestratorEntrypointConfiguration.get_entrypoint_arguments(
+                run_name=orchestrator_run_name,
+                deployment_id=deployment.id,
+            )
+        else:
+            # Run the entire pipeline in one VM using PipelineEntrypointConfiguration
+            command = PipelineEntrypointConfiguration.get_entrypoint_command()
+            args = PipelineEntrypointConfiguration.get_entrypoint_arguments(
+                deployment_id=deployment.id
+            )
+
         entrypoint_str = " ".join(command)
-        args = SkypilotOrchestratorEntrypointConfiguration.get_entrypoint_arguments(
-            run_name=orchestrator_run_name,
-            deployment_id=deployment.id,
-        )
         arguments_str = " ".join(args)
 
         docker_environment_str = " ".join(