Allow multi step configuration for skypilot

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

@@ -65,16 +65,16 @@ To use the SkyPilot VM Orchestrator, you need:
 * One of the SkyPilot integrations installed. You can install the SkyPilot integration for your cloud provider of choice using the following command:
   ```shell
     # For AWS
-    pip install "zenml[connectors-gcp]"
-    zenml integration install aws vm_aws 
+    pip install "zenml[connectors-aws]"
+    zenml integration install aws skypilot_aws 
 
     # for GCP
     pip install "zenml[connectors-gcp]"
-    zenml integration install gcp vm_gcp # for GCP
+    zenml integration install gcp skypilot_gcp # for GCP
 
     # for Azure
     pip install "zenml[connectors-azure]"
-    zenml integration install azure vm_azure # for Azure
+    zenml integration install azure skypilot_azure # for Azure
   ```
 * [Docker](https://www.docker.com) installed and running.
 * A [remote artifact store](../artifact-stores/artifact-stores.md) as part of your stack.
@@ -371,6 +371,38 @@ skypilot_settings = SkypilotAzureOrchestratorSettings(
 {% endtab %}
 {% endtabs %}
 
+
+One of the key features of the SkyPilot VM Orchestrator is the ability to run each step of a pipeline on a separate VM with its own specific settings. This allows for fine-grained control over the resources allocated to each step, ensuring that each part of your pipeline has the necessary compute power while optimizing for cost and efficiency.
+
+## Configuring Step-Specific Resources
+
+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.
+
+Here's an example of how to configure specific resources for a step for the AWS cloud:
+
+```python
+from zenml.integrations.skypilot.flavors.skypilot_orchestrator_aws_vm_flavor import SkypilotAWSOrchestratorSettings
+
+# Settings for a specific step that requires more resources
+high_resource_settings = SkypilotAWSOrchestratorSettings(
+    instance_type='t2.2xlarge',
+    cpus=8,
+    memory=32,
+    use_spot=False,
+    region='us-east-1',
+    # ... other settings
+)
+
+@step(settings={"orchestrator.vm_aws": high_resource_settings})
+def my_resource_intensive_step():
+    # Step implementation
+    pass
+```
+
+By using the `settings` parameter, you can tailor the resources for each step according to its specific needs. This flexibility allows you to optimize your pipeline execution for both performance and cost.
+
 Check out
 the [SDK docs](https://sdkdocs.zenml.io/latest/integration\_code\_docs/integrations-skypilot/#zenml.integrations.skypilot.flavors.skypilot\_orchestrator\_base\_vm\_flavor.SkypilotBaseOrchestratorSettings)
 for a full list of available attributes and [this docs page](/docs/book/user-guide/advanced-guide/pipelining-features/configure-steps-pipelines.md) for more

src/zenml/integrations/aws/service_connectors/aws_service_connector.py

@@ -27,6 +27,7 @@
 import base64
 import datetime
 import json
+import os
 import re
 from typing import Any, Dict, List, Optional, Tuple, cast
 
@@ -1371,6 +1372,21 @@ def _configure_local_client(
                     "aws_session_token"
                 ] = credentials.token
 
+            aws_credentials_path = os.path.join(
+                users_home, ".aws", "credentials"
+            )
+
+            # Ensure the .aws directory exists
+            os.makedirs(os.path.dirname(aws_credentials_path), exist_ok=True)
+
+            # Ensure the credentials file exists
+            if not os.path.isfile(aws_credentials_path):
+                open(aws_credentials_path, "a").close()
+
+            # Set the appropriate permissions for the .aws directory and credentials file
+            os.chmod(os.path.dirname(aws_credentials_path), 0o700)
+            os.chmod(aws_credentials_path, 0o600)
+
             common.rewrite_credentials_file(all_profiles, users_home)
 
             logger.info(

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

@@ -14,24 +14,26 @@
 """Implementation of the Skypilot base VM orchestrator."""
 
 import os
-import time
+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 (
     SkypilotBaseOrchestratorSettings,
 )
+from zenml.integrations.skypilot.orchestrators.skypilot_orchestrator_entrypoint_configuration import (
+    SkypilotOrchestratorEntrypointConfiguration,
+)
 from zenml.logger import get_logger
 from zenml.orchestrators import (
     ContainerizedOrchestrator,
 )
+from zenml.orchestrators.utils import get_orchestrator_run_name
 from zenml.stack import StackValidator
-from zenml.utils import string_utils
 
 if TYPE_CHECKING:
     from zenml.models import PipelineDeploymentResponse
@@ -136,7 +138,7 @@ def prepare_or_run_pipeline(
         stack: "Stack",
         environment: Dict[str, str],
     ) -> Any:
-        """Runs all pipeline steps in Skypilot containers.
+        """Runs each pipeline step in a separate Skypilot container.
 
         Args:
             deployment: The pipeline deployment to prepare or run.
@@ -147,39 +149,59 @@ def prepare_or_run_pipeline(
         Raises:
             Exception: If the pipeline run fails.
         """
+        # First check whether the code is running in a notebook.
+        if Environment.in_notebook():
+            raise RuntimeError(
+                "The Kubernetes orchestrator cannot run pipelines in a notebook "
+                "environment. The reason is that it is non-trivial to create "
+                "a Docker image of a notebook. Please consider refactoring "
+                "your notebook cells into separate scripts in a Python module "
+                "and run the code outside of a notebook when using this "
+                "orchestrator."
+            )
         if deployment.schedule:
             logger.warning(
                 "Skypilot Orchestrator currently does not support the"
                 "use of schedules. The `schedule` will be ignored "
                 "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),
         )
 
-        entrypoint = PipelineEntrypointConfiguration.get_entrypoint_command()
-        entrypoint_str = " ".join(entrypoint)
-        arguments = PipelineEntrypointConfiguration.get_entrypoint_arguments(
-            deployment_id=deployment.id
+        pipeline_name = deployment.pipeline_configuration.name
+        orchestrator_run_name = get_orchestrator_run_name(pipeline_name)
+
+        assert stack.container_registry
+
+        # Get Docker image for the orchestrator pod
+        try:
+            image = self.get_image(deployment=deployment)
+        except KeyError:
+            # If no generic pipeline image exists (which means all steps have
+            # custom builds) we use a random step image as all of them include
+            # dependencies for the active stack
+            pipeline_step_name = next(iter(deployment.step_configurations))
+            image = self.get_image(
+                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()
+        entrypoint_str = " ".join(command)
+        args = SkypilotOrchestratorEntrypointConfiguration.get_entrypoint_arguments(
+            run_name=orchestrator_run_name,
+            deployment_id=deployment.id,
         )
-        arguments_str = " ".join(arguments)
+        arguments_str = " ".join(args)
 
-        # Set up docker run command
-        image = self.get_image(deployment=deployment)
         docker_environment_str = " ".join(
             f"-e {k}={v}" for k, v in environment.items()
         )
 
-        start_time = time.time()
-
         instance_type = settings.instance_type or self.DEFAULT_INSTANCE_TYPE
 
         # Set up credentials
@@ -188,8 +210,7 @@ def prepare_or_run_pipeline(
         # Guaranteed by stack validation
         assert stack is not None and stack.container_registry is not None
 
-        docker_creds = stack.container_registry.credentials
-        if docker_creds:
+        if docker_creds := stack.container_registry.credentials:
             docker_username, docker_password = docker_creds
             setup = (
                 f"docker login --username $DOCKER_USERNAME --password "
@@ -232,7 +253,10 @@ def prepare_or_run_pipeline(
                 )
             )
 
-            cluster_name = settings.cluster_name
+            # Set the cluster name
+            cluster_name = (
+                settings.cluster_name
+            )  # or self.sanitize_cluster_name(orchestrator_run_name)
             if cluster_name is None:
                 # Find existing cluster
                 for i in sky.status(refresh=True):
@@ -261,8 +285,15 @@ def prepare_or_run_pipeline(
             # Unset the service connector AWS profile ENV variable
             self.prepare_environment_variable(set=False)
 
-        run_duration = time.time() - start_time
-        logger.info(
-            "Pipeline run has finished in `%s`.",
-            string_utils.get_human_readable_time(run_duration),
-        )
+    def sanitize_cluster_name(self, name: str) -> str:
+        """Sanitize the value to be used in a cluster name.
+
+        Args:
+            name: Arbitrary input cluster name.
+
+        Returns:
+            Sanitized cluster name.
+        """
+        name = re.sub(r"[^a-z0-9-]", "-", name.lower())
+        name = re.sub(r"^[-]+", "", name)
+        return re.sub(r"[-]+", "-", name)