feat(BA-2164): Add DB Tables for Model Deployment

changes/5599.feature.md

@@ -0,0 +1 @@
+Add DB Tables for Model Deployment (`model_deployments`, `model_revisions`, `deployment_routes`, `deployment_states`, `model_deployment_tokens`, `model_deployment_auto_scaling_rules`)

src/ai/backend/common/data/model_deployment/types.py

@@ -0,0 +1,33 @@
+from enum import StrEnum
+
+
+class ReadinessStatus(StrEnum):
+    NOT_CHECKED = "NOT_CHECKED"
+    HEALTHY = "HEALTHY"
+    UNHEALTHY = "UNHEALTHY"
+
+
+class LivenessStatus(StrEnum):
+    NOT_CHECKED = "NOT_CHECKED"
+    HEALTHY = "HEALTHY"
+    UNHEALTHY = "UNHEALTHY"
+    DEGRADED = "DEGRADED"
+
+
+class ActivenessStatus(StrEnum):
+    ACTIVE = "ACTIVE"
+    INACTIVE = "INACTIVE"
+
+
+class ModelDeploymentStatus(StrEnum):
+    PENDING = "PENDING"
+    SCALING = "SCALING"
+    DEPLOYING = "DEPLOYING"
+    READY = "READY"
+    STOPPING = "STOPPING"
+    STOPPED = "STOPPED"
+
+
+class DeploymentStrategy(StrEnum):
+    ROLLING = "ROLLING"
+    BLUE_GREEN = "BLUE_GREEN"

src/ai/backend/manager/models/__init__.py

@@ -7,6 +7,8 @@
 from . import association_container_registries_groups as _association_container_registries_groups
 from . import audit_log as _auditlog
 from . import container_registry as _container_registry
+from . import deployment_route as _deployment_route
+from . import deployment_state as _deployment_state
 from . import domain as _domain
 from . import dotfile as _dotfile
 from . import endpoint as _endpoint
@@ -18,6 +20,10 @@
 from . import image as _image
 from . import kernel as _kernel
 from . import keypair as _keypair
+from . import model_deployment as _model_deployment
+from . import model_deployment_auto_scaling_rules as _model_deployment_auto_scaling_rules
+from . import model_deployment_tokens as _model_deployment_tokens
+from . import model_revision as _model_revision
 from . import network as _network
 from . import object_storage as _object_storage
 from . import rbac as _rbac
@@ -61,6 +67,12 @@
     *_object_storage.__all__,
     *_user.__all__,
     *_vfolder.__all__,
+    *_model_deployment.__all__,
+    *_model_revision.__all__,
+    *_deployment_route.__all__,
+    *_model_deployment_auto_scaling_rules.__all__,
+    *_model_deployment_tokens.__all__,
+    *_deployment_state.__all__,
     *_dotfile.__all__,
     *_rbac.__all__,
     *_rbac_models.__all__,
@@ -117,3 +129,9 @@
 from .gql_models.kernel import *  # noqa
 from .gql_models.session import *  # noqa
 from .reservoir_registry import *  # noqa
+from .model_deployment import *  # noqa
+from .model_revision import *  # noqa
+from .deployment_route import *  # noqa
+from .model_deployment_auto_scaling_rules import *  # noqa
+from .model_deployment_tokens import *  # noqa
+from .deployment_state import *  # noqa

src/ai/backend/manager/models/alembic/versions/96ab6ecb5f9d_add_model_deployment_related_tables.py

@@ -0,0 +1,186 @@
+"""Add Model Deployment related tables
+
+Revision ID: 96ab6ecb5f9d
+Revises: 5b171528a6f5
+Create Date: 2025-09-04 11:49:56.488145
+
+"""
+
+import sqlalchemy as sa
+from alembic import op
+from sqlalchemy.dialects import postgresql
+
+from ai.backend.common.data.model_deployment.types import (
+    ActivenessStatus,
+    DeploymentStrategy,
+    LivenessStatus,
+    ModelDeploymentStatus,
+    ReadinessStatus,
+)
+from ai.backend.common.types import RuntimeVariant, VFolderMount
+from ai.backend.manager.models.base import (
+    GUID,
+    DecimalType,
+    ResourceSlotColumn,
+    StrEnumType,
+    StructuredJSONObjectListColumn,
+)
+
+# revision identifiers, used by Alembic.
+revision = "96ab6ecb5f9d"
+down_revision = "5b171528a6f5"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    # ### commands auto generated by Alembic - please adjust! ###
+    op.create_table(
+        "deployment_routes",
+        sa.Column("id", GUID(), server_default=sa.text("uuid_generate_v4()"), nullable=False),
+        sa.Column("deployment_id", GUID(), nullable=False),
+        sa.Column("session_id", GUID(), nullable=False),
+        sa.Column("revision_id", GUID(), nullable=False),
+        sa.Column("weight", sa.Integer(), nullable=False),
+        sa.Column("readiness_status", StrEnumType(ReadinessStatus, length=64), nullable=False),
+        sa.Column("liveness_status", StrEnumType(LivenessStatus, length=64), nullable=False),
+        sa.Column("activeness_status", StrEnumType(ActivenessStatus, length=64), nullable=False),
+        sa.Column("detail", postgresql.JSONB(astext_type=sa.Text()), nullable=True),
+        sa.Column(
+            "created_at",
+            sa.DateTime(timezone=True),
+            server_default=sa.text("now()"),
+            nullable=False,
+        ),
+        sa.Column(
+            "updated_at",
+            sa.DateTime(timezone=True),
+            server_default=sa.text("now()"),
+            nullable=False,
+        ),
+        sa.PrimaryKeyConstraint("id", name=op.f("pk_deployment_routes")),
+    )
+    op.create_table(
+        "deployment_states",
+        sa.Column("id", GUID(), server_default=sa.text("uuid_generate_v4()"), nullable=False),
+        sa.Column("deployment_id", GUID(), nullable=False),
+        sa.Column("prev_revision_id", GUID(), nullable=False),
+        sa.Column("next_revision_id", GUID(), nullable=True),
+        sa.Column("strategy", StrEnumType(DeploymentStrategy, length=64), nullable=False),
+        sa.Column("status", StrEnumType(ModelDeploymentStatus, length=64), nullable=False),
+        sa.Column("created_at", sa.DateTime(), nullable=False),
+        sa.Column("updated_at", sa.DateTime(), nullable=False),
+        sa.PrimaryKeyConstraint("id", name=op.f("pk_deployment_states")),
+    )
+    op.create_table(
+        "model_deployment_auto_scaling_rules",
+        sa.Column("id", GUID(), server_default=sa.text("uuid_generate_v4()"), nullable=False),
+        sa.Column("model_deployment_id", GUID(), nullable=False),
+        sa.Column("metric_source", sa.String(), nullable=False),
+        sa.Column("metric_name", sa.String(), nullable=False),
+        sa.Column("min_threshold", DecimalType(), nullable=True),
+        sa.Column("max_threshold", DecimalType(), nullable=True),
+        sa.Column("step_size", sa.Integer(), nullable=False),
+        sa.Column("time_window", sa.Integer(), nullable=False),
+        sa.Column("min_replicas", sa.Integer(), nullable=True),
+        sa.Column("max_replicas", sa.Integer(), nullable=True),
+        sa.Column(
+            "created_at",
+            sa.DateTime(timezone=True),
+            server_default=sa.text("now()"),
+            nullable=False,
+        ),
+        sa.Column("last_triggered_at", sa.DateTime(timezone=True), nullable=False),
+        sa.PrimaryKeyConstraint("id", name=op.f("pk_model_deployment_auto_scaling_rules")),
+    )
+    op.create_table(
+        "model_deployment_tokens",
+        sa.Column("id", GUID(), server_default=sa.text("uuid_generate_v4()"), nullable=False),
+        sa.Column("token", sa.String(), nullable=False),
+        sa.Column("model_deployment_id", GUID(), nullable=False),
+        sa.Column(
+            "created_at", sa.DateTime(timezone=True), server_default=sa.text("now()"), nullable=True
+        ),
+        sa.PrimaryKeyConstraint("id", name=op.f("pk_model_deployment_tokens")),
+    )
+    op.create_table(
+        "model_deployments",
+        sa.Column("id", GUID(), server_default=sa.text("uuid_generate_v4()"), nullable=False),
+        sa.Column("name", sa.String(), nullable=False),
+        sa.Column("status", StrEnumType(ModelDeploymentStatus, length=64), nullable=False),
+        sa.Column("tags", sa.String(), nullable=False),
+        sa.Column("endpoint_url", sa.String(), nullable=True),
+        sa.Column("preferred_domain_name", sa.String(), nullable=True),
+        sa.Column("open_to_public", sa.Boolean(), nullable=False),
+        sa.Column("desired_replica_count", sa.Integer(), nullable=False),
+        sa.Column("created_user_id", GUID(), nullable=False),
+        sa.Column("current_revision_id", GUID(), nullable=False),
+        sa.Column(
+            "default_deployment_strategy_type",
+            StrEnumType(DeploymentStrategy, length=64),
+            nullable=False,
+        ),
+        sa.Column("deployment_state_id", GUID(), nullable=True),
+        sa.Column(
+            "created_at", sa.DateTime(timezone=True), server_default=sa.text("now()"), nullable=True
+        ),
+        sa.Column(
+            "updated_at", sa.DateTime(timezone=True), server_default=sa.text("now()"), nullable=True
+        ),
+        sa.PrimaryKeyConstraint("id", name=op.f("pk_model_deployments")),
+    )
+    op.create_index(
+        op.f("ix_model_deployments_created_user_id"),
+        "model_deployments",
+        ["created_user_id"],
+        unique=False,
+    )
+    op.create_index(op.f("ix_model_deployments_name"), "model_deployments", ["name"], unique=False)
+    op.create_table(
+        "model_revisions",
+        sa.Column("id", GUID(), server_default=sa.text("uuid_generate_v4()"), nullable=False),
+        sa.Column("name", sa.String(), nullable=False),
+        sa.Column("model_deployment_id", GUID(), nullable=False),
+        sa.Column("tags", sa.TEXT(), nullable=False),
+        sa.Column("runtime_variant", StrEnumType(RuntimeVariant, length=64), nullable=False),
+        sa.Column(
+            "inference_runtime_config", postgresql.JSONB(astext_type=sa.Text()), nullable=False
+        ),
+        sa.Column("environment_variables", postgresql.JSONB(astext_type=sa.Text()), nullable=False),
+        sa.Column("model_vfolder_id", GUID(), nullable=False),
+        sa.Column("model_mount_destination", sa.String(), server_default="/models", nullable=False),
+        sa.Column("model_definition_path", sa.String(), nullable=False),
+        sa.Column("image_id", GUID(), nullable=False),
+        sa.Column("cluster_mode", sa.String(), server_default="single-node", nullable=False),
+        sa.Column("cluster_size", sa.Integer(), server_default="1", nullable=False),
+        sa.Column("resource_group", sa.String(), nullable=False),
+        sa.Column("resource_slots", ResourceSlotColumn(), nullable=False),
+        sa.Column("resource_opts", postgresql.JSONB(astext_type=sa.Text()), nullable=False),
+        sa.Column(
+            "extra_mount",
+            StructuredJSONObjectListColumn(VFolderMount),
+            server_default="[]",
+            nullable=False,
+        ),
+        sa.Column(
+            "created_at",
+            sa.DateTime(timezone=True),
+            server_default=sa.text("now()"),
+            nullable=False,
+        ),
+        sa.PrimaryKeyConstraint("id", name=op.f("pk_model_revisions")),
+    )
+    # ### end Alembic commands ###
+
+
+def downgrade() -> None:
+    # ### commands auto generated by Alembic - please adjust! ###
+    op.drop_table("model_revisions")
+    op.drop_index(op.f("ix_model_deployments_name"), table_name="model_deployments")
+    op.drop_index(op.f("ix_model_deployments_created_user_id"), table_name="model_deployments")
+    op.drop_table("model_deployments")
+    op.drop_table("model_deployment_tokens")
+    op.drop_table("model_deployment_auto_scaling_rules")
+    op.drop_table("deployment_states")
+    op.drop_table("deployment_routes")
+    # ### end Alembic commands ###