feat: support image building for arm64 (#1087)

*⚠️ Breaking API change on `PUT /resource_pools/{resource_pool_id}`: a new `platform` field is required.

Add support for building images for `linux/arm64` to be used when running sessions at CSCS.

Other changes include support for storing the `platform` of resource pools and for getting `platform` information about container images. These changes help support users who want to run sessions on `linux/arm64`.

---------

Co-authored-by: Tasko Olevski <olevski90@gmail.com>

Author: leafty

bases/renku_data_services/data_api/dependencies.py

@@ -26,6 +26,7 @@
 from renku_data_services.connected_services.db import ConnectedServicesRepository
 from renku_data_services.connected_services.oauth_http import DefaultOAuthHttpClientFactory, OAuthHttpClientFactory
 from renku_data_services.crc import models as crc_models
+from renku_data_services.crc.constants import DEFAULT_RUNTIME_PLATFORM
 from renku_data_services.crc.db import ClusterRepository, ResourcePoolRepository, UserRepository
 from renku_data_services.crc.server_options import (
     ServerOptions,
@@ -101,6 +102,7 @@
     quota=None,
     public=True,
     default=True,
+    platform=DEFAULT_RUNTIME_PLATFORM,
 )
 
 

components/renku_data_services/crc/api.spec.yaml

@@ -1294,6 +1294,8 @@ components:
           $ref: "#/components/schemas/HibernationThreshold"
         cluster_id:
           $ref: "#/components/schemas/Ulid"
+        platform:
+          $ref: "#/components/schemas/RuntimePlatform"
       required: ["classes", "name", "public", "default"]
       example:
         quota:
@@ -1342,6 +1344,8 @@ components:
           $ref: "#/components/schemas/HibernationThreshold"
         cluster_id:
           $ref: "#/components/schemas/Ulid"
+        platform:
+          $ref: "#/components/schemas/RuntimePlatform"
       example:
         quota:
           cpu: 50
@@ -1374,7 +1378,9 @@ components:
           $ref: "#/components/schemas/HibernationThreshold"
         cluster_id:
           $ref: "#/components/schemas/Ulid"
-      required: ["classes", "name", "public", "default"]
+        platform:
+          $ref: "#/components/schemas/RuntimePlatform"
+      required: ["classes", "name", "public", "default", "platform"]
       example:
         quota:
           cpu: 50
@@ -1429,7 +1435,9 @@ components:
             id:
               $ref: "#/components/schemas/Ulid"
           required: ["id"]
-      required: ["classes", "name", "id", "public", "default"]
+        platform:
+          $ref: "#/components/schemas/RuntimePlatform"
+      required: ["classes", "name", "id", "public", "default", "platform"]
       example:
         quota:
           cpu: 50
@@ -1481,7 +1489,9 @@ components:
           $ref: "#/components/schemas/HibernationThreshold"
         cluster_id:
           $ref: "#/components/schemas/Ulid"
-      required: ["classes", "name", "id", "public", "default"]
+        platform:
+          $ref: "#/components/schemas/RuntimePlatform"
+      required: ["classes", "name", "id", "public", "default", "platform"]
       example:
         quota:
           cpu: 50
@@ -1871,6 +1881,11 @@ components:
       uniqueItems: true
       default: []
       minItems: 0
+    RuntimePlatform:
+      type: string
+      description: |
+        A runtime platform, e.g. "linux/amd64" or "linux/arm64".
+      enum: ["linux/amd64", "linux/arm64"]
     ErrorResponse:
       type: object
       properties:

components/renku_data_services/crc/apispec.py

@@ -1,6 +1,6 @@
 # generated by datamodel-codegen:
 #   filename:  api.spec.yaml
-#   timestamp: 2025-10-13T09:06:36+00:00
+#   timestamp: 2025-11-27T12:13:24+00:00
 
 from __future__ import annotations
 
@@ -123,6 +123,11 @@ class NodeAffinityListResponse(RootModel[List[NodeAffinity]]):
     )
 
 
+class RuntimePlatform(Enum):
+    linux_amd64 = "linux/amd64"
+    linux_arm64 = "linux/arm64"
+
+
 class Error(BaseAPISpec):
     code: int = Field(..., examples=[1404], gt=0)
     detail: Optional[str] = Field(
@@ -824,6 +829,7 @@ class ResourcePoolPatch(BaseAPISpec):
         min_length=26,
         pattern="^[0-7][0-9A-HJKMNP-TV-Z]{25}$",
     )
+    platform: Optional[RuntimePlatform] = None
 
 
 class ResourcePool(BaseAPISpec):
@@ -871,6 +877,7 @@ class ResourcePool(BaseAPISpec):
         min_length=26,
         pattern="^[0-7][0-9A-HJKMNP-TV-Z]{25}$",
     )
+    platform: Optional[RuntimePlatform] = None
 
 
 class ResourcePoolPut(BaseAPISpec):
@@ -944,6 +951,7 @@ class ResourcePoolPut(BaseAPISpec):
         min_length=26,
         pattern="^[0-7][0-9A-HJKMNP-TV-Z]{25}$",
     )
+    platform: RuntimePlatform
 
 
 class ResourcePoolWithId(BaseAPISpec):
@@ -991,6 +999,7 @@ class ResourcePoolWithId(BaseAPISpec):
         le=2147483647,
     )
     cluster: Optional[Cluster1] = None
+    platform: RuntimePlatform
 
 
 class ResourcePoolWithIdFiltered(BaseAPISpec):
@@ -1044,6 +1053,7 @@ class ResourcePoolWithIdFiltered(BaseAPISpec):
         min_length=26,
         pattern="^[0-7][0-9A-HJKMNP-TV-Z]{25}$",
     )
+    platform: RuntimePlatform
 
 
 class ResourcePoolsWithId(RootModel[List[ResourcePoolWithId]]):

components/renku_data_services/crc/constants.py

@@ -0,0 +1,8 @@
+"""Constant values used for resource pools and resource classes."""
+
+from typing import Final
+
+from renku_data_services.crc import models
+
+DEFAULT_RUNTIME_PLATFORM: Final[models.RuntimePlatform] = models.RuntimePlatform.linux_amd64
+"""The default runtime platform used by resource pools, "linux/amd64"."""

components/renku_data_services/crc/core.py

@@ -179,6 +179,7 @@ def validate_resource_pool(body: apispec.ResourcePool) -> models.UnsavedResource
         raise errors.ValidationError(message="One default class is required in each resource pool.")
 
     remote = validate_remote(body=body.remote) if body.remote else None
+    platform = __validate_runtime_platform(body=body.platform)
 
     return models.UnsavedResourcePool(
         name=body.name,
@@ -190,6 +191,7 @@ def validate_resource_pool(body: apispec.ResourcePool) -> models.UnsavedResource
         public=body.public,
         remote=remote,
         cluster_id=ULID.from_str(body.cluster_id) if body.cluster_id else None,
+        platform=platform,
     )
 
 
@@ -200,6 +202,7 @@ def validate_resource_pool_patch(body: apispec.ResourcePoolPatch) -> models.Reso
     )
     quota = validate_quota_put_patch(body=body.quota) if body.quota else None
     remote = validate_remote_patch(body=body.remote) if body.remote else None
+    platform = __validate_runtime_platform(body=body.platform) if body.platform else None
     return models.ResourcePoolPatch(
         name=body.name,
         classes=classes,
@@ -210,6 +213,7 @@ def validate_resource_pool_patch(body: apispec.ResourcePoolPatch) -> models.Reso
         public=body.public,
         remote=remote,
         cluster_id=ULID.from_str(body.cluster_id) if body.cluster_id else None,
+        platform=platform,
     )
 
 
@@ -221,6 +225,7 @@ def validate_resource_pool_put(body: apispec.ResourcePoolPut) -> models.Resource
     )
     quota = validate_quota_put_patch(body=body.quota) if body.quota else RESET
     remote = validate_remote_put(body=body.remote)
+    platform = __validate_runtime_platform(body=body.platform) if body.platform else RESET
     return models.ResourcePoolPatch(
         name=body.name,
         classes=classes,
@@ -231,6 +236,7 @@ def validate_resource_pool_put(body: apispec.ResourcePoolPut) -> models.Resource
         public=body.public,
         remote=remote,
         cluster_id=ULID.from_str(body.cluster_id) if body.cluster_id else RESET,
+        platform=platform,
     )
 
 
@@ -460,3 +466,18 @@ def validate_firecrest_api_url(url: str) -> None:
             message=f"The scheme for the firecrest api url {url} is not valid, expected one of {accepted_schemes}",
             quiet=True,
         )
+
+
+def __validate_runtime_platform(body: apispec.RuntimePlatform | None) -> models.RuntimePlatform:
+    """Validate the platform field for resource pools."""
+    platform_str: str = models.RuntimePlatform.linux_amd64
+    if body:
+        platform_str = body.value
+    if platform_str not in models.RuntimePlatform:
+        raise errors.ValidationError(
+            message=(
+                f"Invalid value for the field 'platform': {body}: "
+                f"Valid values are {[e.value for e in models.RuntimePlatform]}"
+            )
+        )
+    return models.RuntimePlatform(platform_str)

components/renku_data_services/crc/db.py

@@ -423,6 +423,8 @@ async def update_resource_pool(
                 rp.hibernation_threshold = None
             elif isinstance(update.hibernation_threshold, int):
                 rp.hibernation_threshold = update.hibernation_threshold
+            if update.platform is not None:
+                rp.platform = update.platform
 
             if update.cluster_id is RESET:
                 rp.cluster_id = None

components/renku_data_services/crc/models.py

@@ -283,6 +283,13 @@ class SavedClusterSettings(ClusterSettings):
     id: ClusterId
 
 
+class RuntimePlatform(StrEnum):
+    """Represents a runtime platform."""
+
+    linux_amd64 = "linux/amd64"
+    linux_arm64 = "linux/arm64"
+
+
 @dataclass(frozen=True, eq=True, kw_only=True)
 class UnsavedResourcePool:
     """Model for a resource pool yet to be saved."""
@@ -296,6 +303,7 @@ class UnsavedResourcePool:
     public: bool = False
     remote: RemoteConfigurationFirecrest | None = None
     cluster_id: ClusterId | None = None
+    platform: RuntimePlatform
 
 
 @dataclass(frozen=True, eq=True, kw_only=True)
@@ -312,6 +320,7 @@ class ResourcePool:
     public: bool = False
     remote: RemoteConfigurationFirecrest | None = None
     cluster: SavedClusterSettings | None = None
+    platform: RuntimePlatform
 
     def get_resource_class(self, resource_class_id: int) -> ResourceClass | None:
         """Find a specific resource class in the resource pool by the resource class id."""
@@ -341,6 +350,7 @@ class ResourcePoolPatch:
     public: bool | None = None
     remote: RemoteConfigurationPatch | None = None
     cluster_id: ClusterId | ResetType | None = None
+    platform: RuntimePlatform | None = None
 
 
 class RemoteConfigurationKind(StrEnum):

components/renku_data_services/crc/orm.py

@@ -4,16 +4,7 @@
 
 from typing import Any, Optional
 
-from sqlalchemy import (
-    JSON,
-    BigInteger,
-    Column,
-    Identity,
-    Integer,
-    MetaData,
-    String,
-    Table,
-)
+from sqlalchemy import JSON, BigInteger, Column, Enum, Identity, Integer, MetaData, String, Table, literal
 from sqlalchemy.dialects.postgresql import JSONB
 from sqlalchemy.orm import DeclarativeBase, Mapped, MappedAsDataclass, mapped_column, relationship
 from sqlalchemy.schema import ForeignKey
@@ -262,6 +253,11 @@ class ResourcePoolORM(BaseORM):
     )
     cluster: Mapped[Optional[ClusterORM]] = relationship(viewonly=True, default=None, lazy="selectin", init=False)
 
+    # NOTE: we reuse the "build_platform" enum type here
+    platform: Mapped[models.RuntimePlatform] = mapped_column(
+        Enum(models.RuntimePlatform, name="build_platform"), default=None, server_default=literal("linux_amd64")
+    )
+
     @classmethod
     def from_unsaved_model(
         cls,
@@ -291,6 +287,7 @@ def from_unsaved_model(
             remote_provider_id=remote_provider_id,
             remote_json=remote_json,
             cluster_id=cluster.id if cluster else None,
+            platform=new_resource_pool.platform,
         )
 
     def dump(
@@ -324,6 +321,7 @@ def dump(
             default=self.default,
             remote=remote,
             cluster=cluster,
+            platform=self.platform,
         )
 
     def _dump_remote(self) -> models.RemoteConfigurationFirecrest | None:

components/renku_data_services/crc/server_options.py

@@ -11,6 +11,7 @@
 from pydantic import BaseModel, ByteSize, Field, validator
 
 from renku_data_services.crc import models
+from renku_data_services.crc.constants import DEFAULT_RUNTIME_PLATFORM
 
 
 def _check_greater_than_zero(cls: Any, v: int | float) -> int | float:
@@ -152,4 +153,5 @@ def generate_default_resource_pool(
         default=True,
         public=True,
         name="default",
+        platform=DEFAULT_RUNTIME_PLATFORM,
     )

components/renku_data_services/migrations/versions/328803606473_feat_add_platform_field_to_resource_.py

@@ -0,0 +1,33 @@
+"""feat: add platform field to resource pools.
+
+Revision ID: 328803606473
+Revises: df2c0e65612a
+Create Date: 2025-11-17 12:50:15.907221
+
+"""
+
+import sqlalchemy as sa
+from alembic import op
+
+# revision identifiers, used by Alembic.
+revision = "328803606473"
+down_revision = "df2c0e65612a"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    op.add_column(
+        "resource_pools",
+        sa.Column(
+            "platform",
+            sa.Enum("linux_amd64", "linux_arm64", name="build_platform"),
+            server_default=sa.literal("linux_amd64"),
+            nullable=False,
+        ),
+        schema="resource_pools",
+    )
+
+
+def downgrade() -> None:
+    op.drop_column("resource_pools", "platform", schema="resource_pools")