ellf-cli 17.0.45__tar.gz → 17.0.51__tar.gz

{ellf_cli-17.0.45/ellf_cli.egg-info → ellf_cli-17.0.51}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: ellf-cli
-Version: 17.0.45
+Version: 17.0.51
 Summary: Ellf Command Line Interface
 Home-page: https://prodi.gy
 Author: ExplosionAI GmbH

{ellf_cli-17.0.45 → ellf_cli-17.0.51}/ellf_cli/about.json

@@ -1,7 +1,7 @@
 {
   "title": "Ellf CLI",
   "name": "ellf-cli",
-  "version": "17.0.45",
+  "version": "17.0.51",
   "summary": "Ellf Command Line Interface",
   "uri": "https://prodi.gy",
   "prog": "ellf",

{ellf_cli-17.0.45 → ellf_cli-17.0.51}/ellf_cli/cluster_config.py

@@ -5,7 +5,7 @@ import tempfile
 from contextlib import contextmanager
 from typing import Annotated, Iterator
 
-from pydantic import BaseModel, Field
+from pydantic import BaseModel, Field, model_validator
 
 from .helm import (
     HelmChartValues,
@@ -35,7 +35,8 @@ class ClusterConfig(BaseModel):
         workspace: str | None = None
 
     class Deployment(BaseModel):
-        gcp: HelmChartValues.GCP
+        gcp: HelmChartValues.GCP | None = None
+        aws: HelmChartValues.AWS | None = None
         domain: str
         pam_domain: Annotated[str, Field(alias="pamDomain")] = ""
         container_registry: str = ""
@@ -53,9 +54,25 @@ class ClusterConfig(BaseModel):
         system_node_pool_size: int = 1
         model_config = {"populate_by_name": True}
 
+        @model_validator(mode="after")
+        def _check_exactly_one_cloud_provider(self) -> "ClusterConfig.Deployment":
+            if bool(self.gcp) == bool(self.aws):
+                raise ValueError(
+                    "deployment must set exactly one of 'gcp' or 'aws', not"
+                    f" {'both' if self.gcp else 'neither'}"
+                )
+            return self
+
+        @property
+        def cloud_provider(self) -> str:
+            return "gcp" if self.gcp else "aws"
+
         @property
         def region(self) -> str:
-            return self.gcp.zone[:-2]
+            if self.gcp:
+                return self.gcp.zone[:-2]
+            assert self.aws
+            return self.aws.region
 
     name: str
     terraform: Terraform
@@ -84,6 +101,7 @@ class ClusterConfigOverrides(BaseModel):
 
     class Deployment(BaseModel):
         gcp: HelmChartValues.GCP | None = None
+        aws: HelmChartValues.AWS | None = None
         domain: str | None = None
         pam_domain: Annotated[str | None, Field(alias="pamDomain")] = None
         container_registry: str | None = None
@@ -103,7 +121,11 @@ class ClusterConfigOverrides(BaseModel):
 
         @property
         def region(self) -> str | None:
-            return self.gcp.zone[:-2] if self.gcp else None
+            if self.gcp:
+                return self.gcp.zone[:-2]
+            if self.aws:
+                return self.aws.region
+            return None
 
     name: str | None = None
     terraform: Terraform | None = None

{ellf_cli-17.0.45 → ellf_cli-17.0.51}/ellf_cli/commands/infra/_helpers.py

@@ -26,23 +26,45 @@ SYSTEM_NODE_TOLERATIONS = [
 ]
 
 
-def get_cluster_credentials(project: str, zone: str, cluster_name: str) -> None:
-    """Run gcloud container clusters get-credentials to configure kubectl."""
-    subprocess.run(
-        [
-            "gcloud",
-            "container",
-            "clusters",
-            "get-credentials",
-            cluster_name,
-            "--zone",
-            zone,
-            "--project",
-            project,
-        ],
-        check=True,
-        text=True,
-    )
+def get_cluster_credentials(
+    cloud_provider: str,
+    cluster_name: str,
+    *,
+    project: str = "",
+    zone: str = "",
+    region: str = "",
+) -> None:
+    """Configure kubectl for the cluster via the provider's own credential command."""
+    if cloud_provider == "aws":
+        subprocess.run(
+            [
+                "aws",
+                "eks",
+                "update-kubeconfig",
+                "--region",
+                region,
+                "--name",
+                cluster_name,
+            ],
+            check=True,
+            text=True,
+        )
+    else:
+        subprocess.run(
+            [
+                "gcloud",
+                "container",
+                "clusters",
+                "get-credentials",
+                cluster_name,
+                "--zone",
+                zone,
+                "--project",
+                project,
+            ],
+            check=True,
+            text=True,
+        )
 
 
 def ensure_cluster_keypair(infra_secret_name: str, namespace: str = "ellf") -> str:

{ellf_cli-17.0.45 → ellf_cli-17.0.51}/ellf_cli/commands/infra/provision.py

@@ -51,9 +51,19 @@ def provision(
     cluster_name = config.deployment.cluster_name
     assert cluster_name, "cluster_name must be set in deployment config"
     auth = get_auth_state()
-    get_cluster_credentials(
-        config.deployment.gcp.project, config.deployment.gcp.zone, cluster_name
-    )
+    if config.deployment.cloud_provider == "aws":
+        assert config.deployment.aws
+        get_cluster_credentials(
+            "aws", cluster_name, region=config.deployment.aws.region
+        )
+    else:
+        assert config.deployment.gcp
+        get_cluster_credentials(
+            "gcp",
+            cluster_name,
+            project=config.deployment.gcp.project,
+            zone=config.deployment.gcp.zone,
+        )
     ensure_cluster_keypair(config.deployment.infra_secret_name)
     cluster_auth = get_cluster_auth(auth.client, config.deployment.domain)
     values = build_values(config, cluster_auth)

{ellf_cli-17.0.45 → ellf_cli-17.0.51}/ellf_cli/commands/infra/start.py

@@ -56,9 +56,20 @@ def start(
     config = ClusterConfig(**config_dict)
     cluster_name = config.deployment.cluster_name
     assert cluster_name, "cluster_name must be set in deployment config"
-    get_cluster_credentials(
-        config.deployment.gcp.project, config.deployment.gcp.zone, cluster_name
-    )
+    cloud_provider = config.deployment.cloud_provider
+    if cloud_provider == "aws":
+        assert config.deployment.aws
+        get_cluster_credentials(
+            "aws", cluster_name, region=config.deployment.aws.region
+        )
+    else:
+        assert config.deployment.gcp
+        get_cluster_credentials(
+            "gcp",
+            cluster_name,
+            project=config.deployment.gcp.project,
+            zone=config.deployment.gcp.zone,
+        )
     public_key = ensure_cluster_keypair(config.deployment.infra_secret_name)
     if offline is not None:
         from ellf_pam_sdk.models import ClusterCredentials, Image, Secrets
@@ -86,6 +97,7 @@ def start(
             auth.org_id,
             config.deployment.domain,
             public_key,
+            cloud_provider=cloud_provider,
         )
         cluster_auth = get_cluster_auth(auth.client, config.deployment.domain)
         values = build_values(config, cluster_auth)

{ellf_cli-17.0.45 → ellf_cli-17.0.51}/ellf_cli/commands/infra/terraform.py

@@ -116,7 +116,7 @@ def apply_terraform(
     if config_key is not None:
         config_dict = config_dict[config_key]
     config = ClusterConfig(**config_dict)
-    tf_dir = src / "gcp"
+    tf_dir = src / config.deployment.cloud_provider
     if yes_tf_init:
         _terraform_init(config, tf_dir)
     if yes_tf_init or yes_tf_apply:
@@ -185,15 +185,31 @@ def _terraform_init(config: ClusterConfig, path: Path) -> None:
 @contextmanager
 def _write_terraform_vars(config: ClusterConfig) -> Iterator[str]:
     """Write terraform variables to a temp file, yielding its path."""
-    data = {
-        "gcp_project": config.deployment.gcp.project,
-        "gcp_zone": config.deployment.gcp.zone,
-        "domain": config.deployment.domain,
-        "enable_ssh": True,
-        "system_node_pool_machine_type": config.deployment.system_node_pool_machine_type,
-        "system_node_pool_size": str(config.deployment.system_node_pool_size),
-        "worker_types": {w.name: w.to_terraform_var() for w in config.node_pools},
+    cloud_provider = config.deployment.cloud_provider
+    worker_types = {
+        w.name: w.to_terraform_var(cloud_provider) for w in config.node_pools
     }
+    if cloud_provider == "aws":
+        assert config.deployment.aws
+        data = {
+            "aws_region": config.deployment.aws.region,
+            "domain": config.deployment.domain,
+            "enable_ssh": True,
+            "system_node_pool_instance_type": config.deployment.system_node_pool_machine_type,
+            "system_node_pool_size": str(config.deployment.system_node_pool_size),
+            "worker_types": worker_types,
+        }
+    else:
+        assert config.deployment.gcp
+        data = {
+            "gcp_project": config.deployment.gcp.project,
+            "gcp_zone": config.deployment.gcp.zone,
+            "domain": config.deployment.domain,
+            "enable_ssh": True,
+            "system_node_pool_machine_type": config.deployment.system_node_pool_machine_type,
+            "system_node_pool_size": str(config.deployment.system_node_pool_size),
+            "worker_types": worker_types,
+        }
     with tempfile.NamedTemporaryFile(
         mode="w", suffix=".tfvars.json", delete=True
     ) as tmp:

{ellf_cli-17.0.45 → ellf_cli-17.0.51}/ellf_cli/ellf.json

@@ -1,7 +1,7 @@
 {
   "prog": "ellf",
   "help": "Ellf Command Line Interface.",
-  "version": "17.0.44",
+  "version": "17.0.50",
   "extra_key": "_extra",
   "commands": {
     "actions": {

{ellf_cli-17.0.45 → ellf_cli-17.0.51}/ellf_cli/helm.py

@@ -66,6 +66,9 @@ class NodePool(BaseModel):
         preemptible: bool = False
         spot: bool = False
 
+    class AWSConfig(BaseModel):
+        spot: bool = False
+
     name: str
     node_class: str
     machine_type: str
@@ -74,6 +77,7 @@ class NodePool(BaseModel):
     max_size: int | None = None
     gpu: GPU | None = None
     gcp: GCPConfig = Field(default_factory=GCPConfig)
+    aws: AWSConfig = Field(default_factory=AWSConfig)
     per_job: PerJob
 
     @property
@@ -92,18 +96,39 @@ class NodePool(BaseModel):
             per_job=self.per_job,
         )
 
-    def to_terraform_var(self) -> dict:
-        """Convert to a terraform worker_types variable entry."""
+    def _min_max_size(self) -> tuple[int, int]:
+        min_size = self.min_size if self.min_size is not None else 0
+        max_size = (
+            self.max_size if self.max_size is not None else max(self.target_size, 1)
+        )
+        return min_size, max_size
+
+    def to_terraform_var(self, cloud_provider: str) -> dict:
+        """Convert to a terraform worker_types variable entry.
+
+        The two workspaces' `worker_types` variable shapes differ (AWS has no
+        separate preemptible/spot distinction and calls the GPU key `gpu`
+        rather than `guest_accelerator`), so the emitted dict is provider-specific.
+        """
+        min_size, max_size = self._min_max_size()
+        if cloud_provider == "aws":
+            return {
+                "name": self.name,
+                "node_class": self.node_class,
+                "instance_type": self.machine_type,
+                "spot": self.aws.spot,
+                "min_size": min_size,
+                "max_size": max_size,
+                "gpu": self.gpu.model_dump() if self.gpu else None,
+            }
         return {
             "name": self.name,
             "node_class": self.node_class,
             "machine_type": self.machine_type,
             "preemptible": self.preemptible,
             "spot": self.spot,
-            "min_size": self.min_size if self.min_size is not None else 0,
-            "max_size": self.max_size
-            if self.max_size is not None
-            else max(self.target_size, 1),
+            "min_size": min_size,
+            "max_size": max_size,
             "guest_accelerator": self.gpu.model_dump() if self.gpu else None,
         }
 
@@ -204,6 +229,9 @@ class HelmChartValues(BaseModel):
         project: str
         zone: str
 
+    class AWS(BaseModel):
+        region: str
+
     class ConfigFile(BaseModel):
         cluster_config: Annotated[str, Field(alias="clusterConfig")]
         cpl_config: Annotated[str, Field(alias="cplConfig")]
@@ -285,6 +313,7 @@ class HelmChartValues(BaseModel):
     service: Service
     postgresql: Postgresql
     gcp: GCP | None = None
+    aws: AWS | None = None
     automount_service_account_token: Annotated[
         bool | None, Field(alias="automountServiceAccountToken")
     ]
@@ -412,6 +441,9 @@ class HelmChartOverrides(BaseModel):
         project: str | None = None
         zone: str | None = None
 
+    class AWS(BaseModel):
+        region: str | None = None
+
     class ConfigFile(BaseModel):
         cluster_config: Annotated[str | None, Field(alias="clusterConfig")] = None
         cpl_config: Annotated[str | None, Field(alias="cplConfig")] = None
@@ -510,6 +542,7 @@ class HelmChartOverrides(BaseModel):
     service: Service | None = None
     postgresql: Postgresql | None = None
     gcp: GCP | None = None
+    aws: AWS | None = None
     automount_service_account_token: Annotated[
         bool | None, Field(alias="automountServiceAccountToken")
     ] = None
@@ -651,14 +684,17 @@ def build_values(
             infra_secret_name=config.deployment.infra_secret_name,
         ),
         gcp=HelmChartOverrides.GCP(**config.deployment.gcp.model_dump())
-        if config.deployment.gcp.project
+        if config.deployment.gcp and config.deployment.gcp.project
+        else None,
+        aws=HelmChartOverrides.AWS(**config.deployment.aws.model_dump())
+        if config.deployment.aws
         else None,
         service_account=HelmChartOverrides.ServiceAccount(
             annotations={
                 "iam.gke.io/gcp-service-account": f"cluster-gke-sa@{config.deployment.gcp.project}.iam.gserviceaccount.com",
             },
         )
-        if config.deployment.gcp.project
+        if config.deployment.gcp and config.deployment.gcp.project
         else None,
         ingress=HelmChartOverrides.Ingress(
             tls=HelmChartOverrides.Ingress.TLS(

{ellf_cli-17.0.45 → ellf_cli-17.0.51/ellf_cli.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: ellf-cli
-Version: 17.0.45
+Version: 17.0.51
 Summary: Ellf Command Line Interface
 Home-page: https://prodi.gy
 Author: ExplosionAI GmbH