dstack 0.19.23rc1__py3-none-any.whl → 0.19.25__py3-none-any.whl

dstack/_internal/server/migrations/versions/74a1f55209bd_store_enums_as_strings.py

@@ -0,0 +1,484 @@
+"""Store enums as strings
+
+Revision ID: 74a1f55209bd
+Revises: 728b1488b1b4
+Create Date: 2025-08-06 13:49:28.785378
+
+"""
+
+import sqlalchemy as sa
+from alembic import op
+from sqlalchemy.dialects import postgresql
+
+# revision identifiers, used by Alembic.
+revision = "74a1f55209bd"
+down_revision = "728b1488b1b4"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    # ### commands auto generated by Alembic - please adjust! ###
+    with op.batch_alter_table("users", schema=None) as batch_op:
+        batch_op.alter_column(
+            "global_role",
+            existing_type=postgresql.ENUM("ADMIN", "USER", name="globalrole"),
+            type_=sa.String(length=100),
+            existing_nullable=False,
+        )
+
+    with op.batch_alter_table("members", schema=None) as batch_op:
+        batch_op.alter_column(
+            "project_role",
+            existing_type=postgresql.ENUM("ADMIN", "MANAGER", "USER", name="projectrole"),
+            type_=sa.String(length=100),
+            existing_nullable=False,
+        )
+
+    with op.batch_alter_table("repos", schema=None) as batch_op:
+        batch_op.alter_column(
+            "type",
+            existing_type=postgresql.ENUM("REMOTE", "LOCAL", "VIRTUAL", name="repotype"),
+            type_=sa.String(length=100),
+            existing_nullable=False,
+        )
+
+    with op.batch_alter_table("runs", schema=None) as batch_op:
+        batch_op.alter_column(
+            "status",
+            existing_type=postgresql.ENUM(
+                "PENDING",
+                "SUBMITTED",
+                "PROVISIONING",
+                "RUNNING",
+                "TERMINATING",
+                "TERMINATED",
+                "FAILED",
+                "DONE",
+                name="runstatus",
+            ),
+            type_=sa.String(length=100),
+            existing_nullable=False,
+        )
+        batch_op.alter_column(
+            "termination_reason",
+            existing_type=postgresql.ENUM(
+                "ALL_JOBS_DONE",
+                "JOB_FAILED",
+                "RETRY_LIMIT_EXCEEDED",
+                "STOPPED_BY_USER",
+                "ABORTED_BY_USER",
+                "SERVER_ERROR",
+                name="runterminationreason",
+            ),
+            type_=sa.String(length=100),
+            existing_nullable=True,
+        )
+
+    with op.batch_alter_table("jobs", schema=None) as batch_op:
+        batch_op.alter_column(
+            "status",
+            existing_type=postgresql.ENUM(
+                "SUBMITTED",
+                "PROVISIONING",
+                "PULLING",
+                "RUNNING",
+                "TERMINATING",
+                "TERMINATED",
+                "ABORTED",
+                "FAILED",
+                "DONE",
+                name="jobstatus",
+            ),
+            type_=sa.String(length=100),
+            existing_nullable=False,
+        )
+        batch_op.alter_column(
+            "termination_reason",
+            existing_type=postgresql.ENUM(
+                "FAILED_TO_START_DUE_TO_NO_CAPACITY",
+                "INTERRUPTED_BY_NO_CAPACITY",
+                "INSTANCE_UNREACHABLE",
+                "WAITING_INSTANCE_LIMIT_EXCEEDED",
+                "WAITING_RUNNER_LIMIT_EXCEEDED",
+                "TERMINATED_BY_USER",
+                "VOLUME_ERROR",
+                "GATEWAY_ERROR",
+                "SCALED_DOWN",
+                "DONE_BY_RUNNER",
+                "ABORTED_BY_USER",
+                "TERMINATED_BY_SERVER",
+                "INACTIVITY_DURATION_EXCEEDED",
+                "TERMINATED_DUE_TO_UTILIZATION_POLICY",
+                "CONTAINER_EXITED_WITH_ERROR",
+                "PORTS_BINDING_FAILED",
+                "CREATING_CONTAINER_ERROR",
+                "EXECUTOR_ERROR",
+                "MAX_DURATION_EXCEEDED",
+                name="jobterminationreason",
+            ),
+            type_=sa.String(length=100),
+            existing_nullable=True,
+        )
+
+    with op.batch_alter_table("fleets", schema=None) as batch_op:
+        batch_op.alter_column(
+            "status",
+            existing_type=postgresql.ENUM(
+                "SUBMITTED", "ACTIVE", "TERMINATING", "TERMINATED", "FAILED", name="fleetstatus"
+            ),
+            type_=sa.String(length=100),
+            existing_nullable=False,
+        )
+
+    with op.batch_alter_table("gateways", schema=None) as batch_op:
+        batch_op.alter_column(
+            "status",
+            existing_type=postgresql.ENUM(
+                "SUBMITTED", "PROVISIONING", "RUNNING", "FAILED", name="gatewaystatus"
+            ),
+            type_=sa.String(length=100),
+            existing_nullable=False,
+        )
+
+    with op.batch_alter_table("instances", schema=None) as batch_op:
+        batch_op.alter_column(
+            "status",
+            existing_type=postgresql.ENUM(
+                "PENDING",
+                "PROVISIONING",
+                "IDLE",
+                "BUSY",
+                "TERMINATING",
+                "TERMINATED",
+                name="instancestatus",
+            ),
+            type_=sa.String(length=100),
+            existing_nullable=False,
+        )
+
+    with op.batch_alter_table("volumes", schema=None) as batch_op:
+        batch_op.alter_column(
+            "status",
+            existing_type=postgresql.ENUM(
+                "SUBMITTED", "PROVISIONING", "ACTIVE", "FAILED", name="volumestatus"
+            ),
+            type_=sa.String(length=100),
+            existing_nullable=False,
+        )
+
+    sa.Enum("ADMIN", "USER", name="globalrole").drop(op.get_bind())
+    sa.Enum(
+        "ALL_JOBS_DONE",
+        "JOB_FAILED",
+        "RETRY_LIMIT_EXCEEDED",
+        "STOPPED_BY_USER",
+        "ABORTED_BY_USER",
+        "SERVER_ERROR",
+        name="runterminationreason",
+    ).drop(op.get_bind())
+    sa.Enum("SUBMITTED", "PROVISIONING", "RUNNING", "FAILED", name="gatewaystatus").drop(
+        op.get_bind()
+    )
+    sa.Enum("SUBMITTED", "PROVISIONING", "ACTIVE", "FAILED", name="volumestatus").drop(
+        op.get_bind()
+    )
+    sa.Enum(
+        "PENDING",
+        "SUBMITTED",
+        "PROVISIONING",
+        "RUNNING",
+        "TERMINATING",
+        "TERMINATED",
+        "FAILED",
+        "DONE",
+        name="runstatus",
+    ).drop(op.get_bind())
+    sa.Enum("REMOTE", "LOCAL", "VIRTUAL", name="repotype").drop(op.get_bind())
+    sa.Enum(
+        "SUBMITTED",
+        "PROVISIONING",
+        "PULLING",
+        "RUNNING",
+        "TERMINATING",
+        "TERMINATED",
+        "ABORTED",
+        "FAILED",
+        "DONE",
+        name="jobstatus",
+    ).drop(op.get_bind())
+    sa.Enum(
+        "PENDING",
+        "PROVISIONING",
+        "IDLE",
+        "BUSY",
+        "TERMINATING",
+        "TERMINATED",
+        name="instancestatus",
+    ).drop(op.get_bind())
+    sa.Enum("SUBMITTED", "ACTIVE", "TERMINATING", "TERMINATED", "FAILED", name="fleetstatus").drop(
+        op.get_bind()
+    )
+    sa.Enum("ADMIN", "MANAGER", "USER", name="projectrole").drop(op.get_bind())
+    sa.Enum(
+        "FAILED_TO_START_DUE_TO_NO_CAPACITY",
+        "INTERRUPTED_BY_NO_CAPACITY",
+        "INSTANCE_UNREACHABLE",
+        "WAITING_INSTANCE_LIMIT_EXCEEDED",
+        "WAITING_RUNNER_LIMIT_EXCEEDED",
+        "TERMINATED_BY_USER",
+        "VOLUME_ERROR",
+        "GATEWAY_ERROR",
+        "SCALED_DOWN",
+        "DONE_BY_RUNNER",
+        "ABORTED_BY_USER",
+        "TERMINATED_BY_SERVER",
+        "INACTIVITY_DURATION_EXCEEDED",
+        "TERMINATED_DUE_TO_UTILIZATION_POLICY",
+        "CONTAINER_EXITED_WITH_ERROR",
+        "PORTS_BINDING_FAILED",
+        "CREATING_CONTAINER_ERROR",
+        "EXECUTOR_ERROR",
+        "MAX_DURATION_EXCEEDED",
+        name="jobterminationreason",
+    ).drop(op.get_bind())
+    # ### end Alembic commands ###
+
+
+def downgrade() -> None:
+    # ### commands auto generated by Alembic - please adjust! ###
+    sa.Enum(
+        "FAILED_TO_START_DUE_TO_NO_CAPACITY",
+        "INTERRUPTED_BY_NO_CAPACITY",
+        "INSTANCE_UNREACHABLE",
+        "WAITING_INSTANCE_LIMIT_EXCEEDED",
+        "WAITING_RUNNER_LIMIT_EXCEEDED",
+        "TERMINATED_BY_USER",
+        "VOLUME_ERROR",
+        "GATEWAY_ERROR",
+        "SCALED_DOWN",
+        "DONE_BY_RUNNER",
+        "ABORTED_BY_USER",
+        "TERMINATED_BY_SERVER",
+        "INACTIVITY_DURATION_EXCEEDED",
+        "TERMINATED_DUE_TO_UTILIZATION_POLICY",
+        "CONTAINER_EXITED_WITH_ERROR",
+        "PORTS_BINDING_FAILED",
+        "CREATING_CONTAINER_ERROR",
+        "EXECUTOR_ERROR",
+        "MAX_DURATION_EXCEEDED",
+        name="jobterminationreason",
+    ).create(op.get_bind())
+    sa.Enum("ADMIN", "MANAGER", "USER", name="projectrole").create(op.get_bind())
+    sa.Enum(
+        "SUBMITTED", "ACTIVE", "TERMINATING", "TERMINATED", "FAILED", name="fleetstatus"
+    ).create(op.get_bind())
+    sa.Enum(
+        "PENDING",
+        "PROVISIONING",
+        "IDLE",
+        "BUSY",
+        "TERMINATING",
+        "TERMINATED",
+        name="instancestatus",
+    ).create(op.get_bind())
+    sa.Enum(
+        "SUBMITTED",
+        "PROVISIONING",
+        "PULLING",
+        "RUNNING",
+        "TERMINATING",
+        "TERMINATED",
+        "ABORTED",
+        "FAILED",
+        "DONE",
+        name="jobstatus",
+    ).create(op.get_bind())
+    sa.Enum("REMOTE", "LOCAL", "VIRTUAL", name="repotype").create(op.get_bind())
+    sa.Enum(
+        "PENDING",
+        "SUBMITTED",
+        "PROVISIONING",
+        "RUNNING",
+        "TERMINATING",
+        "TERMINATED",
+        "FAILED",
+        "DONE",
+        name="runstatus",
+    ).create(op.get_bind())
+    sa.Enum("SUBMITTED", "PROVISIONING", "ACTIVE", "FAILED", name="volumestatus").create(
+        op.get_bind()
+    )
+    sa.Enum("SUBMITTED", "PROVISIONING", "RUNNING", "FAILED", name="gatewaystatus").create(
+        op.get_bind()
+    )
+    sa.Enum(
+        "ALL_JOBS_DONE",
+        "JOB_FAILED",
+        "RETRY_LIMIT_EXCEEDED",
+        "STOPPED_BY_USER",
+        "ABORTED_BY_USER",
+        "SERVER_ERROR",
+        name="runterminationreason",
+    ).create(op.get_bind())
+    sa.Enum("ADMIN", "USER", name="globalrole").create(op.get_bind())
+    with op.batch_alter_table("volumes", schema=None) as batch_op:
+        batch_op.alter_column(
+            "status",
+            existing_type=sa.String(length=100),
+            type_=postgresql.ENUM(
+                "SUBMITTED", "PROVISIONING", "ACTIVE", "FAILED", name="volumestatus"
+            ),
+            existing_nullable=False,
+            postgresql_using="status::VARCHAR::volumestatus",
+        )
+
+    with op.batch_alter_table("users", schema=None) as batch_op:
+        batch_op.alter_column(
+            "global_role",
+            existing_type=sa.String(length=100),
+            type_=postgresql.ENUM("ADMIN", "USER", name="globalrole"),
+            existing_nullable=False,
+            postgresql_using="global_role::VARCHAR::globalrole",
+        )
+
+    with op.batch_alter_table("runs", schema=None) as batch_op:
+        batch_op.alter_column(
+            "termination_reason",
+            existing_type=sa.String(length=100),
+            type_=postgresql.ENUM(
+                "ALL_JOBS_DONE",
+                "JOB_FAILED",
+                "RETRY_LIMIT_EXCEEDED",
+                "STOPPED_BY_USER",
+                "ABORTED_BY_USER",
+                "SERVER_ERROR",
+                name="runterminationreason",
+            ),
+            existing_nullable=True,
+            postgresql_using="termination_reason::VARCHAR::runterminationreason",
+        )
+        batch_op.alter_column(
+            "status",
+            existing_type=sa.String(length=100),
+            type_=postgresql.ENUM(
+                "PENDING",
+                "SUBMITTED",
+                "PROVISIONING",
+                "RUNNING",
+                "TERMINATING",
+                "TERMINATED",
+                "FAILED",
+                "DONE",
+                name="runstatus",
+            ),
+            existing_nullable=False,
+            postgresql_using="status::VARCHAR::runstatus",
+        )
+
+    with op.batch_alter_table("repos", schema=None) as batch_op:
+        batch_op.alter_column(
+            "type",
+            existing_type=sa.String(length=100),
+            type_=postgresql.ENUM("REMOTE", "LOCAL", "VIRTUAL", name="repotype"),
+            existing_nullable=False,
+            postgresql_using="type::VARCHAR::repotype",
+        )
+
+    with op.batch_alter_table("members", schema=None) as batch_op:
+        batch_op.alter_column(
+            "project_role",
+            existing_type=sa.String(length=100),
+            type_=postgresql.ENUM("ADMIN", "MANAGER", "USER", name="projectrole"),
+            existing_nullable=False,
+            postgresql_using="project_role::VARCHAR::projectrole",
+        )
+
+    with op.batch_alter_table("jobs", schema=None) as batch_op:
+        batch_op.alter_column(
+            "termination_reason",
+            existing_type=sa.String(length=100),
+            type_=postgresql.ENUM(
+                "FAILED_TO_START_DUE_TO_NO_CAPACITY",
+                "INTERRUPTED_BY_NO_CAPACITY",
+                "INSTANCE_UNREACHABLE",
+                "WAITING_INSTANCE_LIMIT_EXCEEDED",
+                "WAITING_RUNNER_LIMIT_EXCEEDED",
+                "TERMINATED_BY_USER",
+                "VOLUME_ERROR",
+                "GATEWAY_ERROR",
+                "SCALED_DOWN",
+                "DONE_BY_RUNNER",
+                "ABORTED_BY_USER",
+                "TERMINATED_BY_SERVER",
+                "INACTIVITY_DURATION_EXCEEDED",
+                "TERMINATED_DUE_TO_UTILIZATION_POLICY",
+                "CONTAINER_EXITED_WITH_ERROR",
+                "PORTS_BINDING_FAILED",
+                "CREATING_CONTAINER_ERROR",
+                "EXECUTOR_ERROR",
+                "MAX_DURATION_EXCEEDED",
+                name="jobterminationreason",
+            ),
+            existing_nullable=True,
+            postgresql_using="termination_reason::VARCHAR::jobterminationreason",
+        )
+        batch_op.alter_column(
+            "status",
+            existing_type=sa.String(length=100),
+            type_=postgresql.ENUM(
+                "SUBMITTED",
+                "PROVISIONING",
+                "PULLING",
+                "RUNNING",
+                "TERMINATING",
+                "TERMINATED",
+                "ABORTED",
+                "FAILED",
+                "DONE",
+                name="jobstatus",
+            ),
+            existing_nullable=False,
+            postgresql_using="status::VARCHAR::jobstatus",
+        )
+
+    with op.batch_alter_table("instances", schema=None) as batch_op:
+        batch_op.alter_column(
+            "status",
+            existing_type=sa.String(length=100),
+            type_=postgresql.ENUM(
+                "PENDING",
+                "PROVISIONING",
+                "IDLE",
+                "BUSY",
+                "TERMINATING",
+                "TERMINATED",
+                name="instancestatus",
+            ),
+            existing_nullable=False,
+            postgresql_using="status::VARCHAR::instancestatus",
+        )
+
+    with op.batch_alter_table("gateways", schema=None) as batch_op:
+        batch_op.alter_column(
+            "status",
+            existing_type=sa.String(length=100),
+            type_=postgresql.ENUM(
+                "SUBMITTED", "PROVISIONING", "RUNNING", "FAILED", name="gatewaystatus"
+            ),
+            existing_nullable=False,
+            postgresql_using="status::VARCHAR::gatewaystatus",
+        )
+
+    with op.batch_alter_table("fleets", schema=None) as batch_op:
+        batch_op.alter_column(
+            "status",
+            existing_type=sa.String(length=100),
+            type_=postgresql.ENUM(
+                "SUBMITTED", "ACTIVE", "TERMINATING", "TERMINATED", "FAILED", name="fleetstatus"
+            ),
+            existing_nullable=False,
+            postgresql_using="status::VARCHAR::fleetstatus",
+        )
+
+    # ### end Alembic commands ###

dstack/_internal/server/migrations/versions/e2d08cd1b8d9_add_jobmodel_fleet.py

@@ -0,0 +1,41 @@
+"""Add JobModel.fleet
+
+Revision ID: e2d08cd1b8d9
+Revises: 3d7f6c2ec000
+Create Date: 2025-08-15 11:26:05.670591
+
+"""
+
+import sqlalchemy as sa
+import sqlalchemy_utils
+from alembic import op
+
+# revision identifiers, used by Alembic.
+revision = "e2d08cd1b8d9"
+down_revision = "3d7f6c2ec000"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    # ### commands auto generated by Alembic - please adjust! ###
+    with op.batch_alter_table("jobs", schema=None) as batch_op:
+        batch_op.add_column(
+            sa.Column(
+                "fleet_id", sqlalchemy_utils.types.uuid.UUIDType(binary=False), nullable=True
+            )
+        )
+        batch_op.create_foreign_key(
+            batch_op.f("fk_jobs_fleet_id_fleets"), "fleets", ["fleet_id"], ["id"]
+        )
+
+    # ### end Alembic commands ###
+
+
+def downgrade() -> None:
+    # ### commands auto generated by Alembic - please adjust! ###
+    with op.batch_alter_table("jobs", schema=None) as batch_op:
+        batch_op.drop_constraint(batch_op.f("fk_jobs_fleet_id_fleets"), type_="foreignkey")
+        batch_op.drop_column("fleet_id")
+
+    # ### end Alembic commands ###

dstack/_internal/server/models.py

@@ -7,7 +7,6 @@ from sqlalchemy import (
     BigInteger,
     Boolean,
     DateTime,
-    Enum,
     Float,
     ForeignKey,
     Index,
@@ -85,7 +84,7 @@ class DecryptedString(CoreModel):
     decrypted: bool = True
     exc: Optional[Exception] = None
 
-    class Config:
+    class Config(CoreModel.Config):
         arbitrary_types_allowed = True
 
     def get_plaintext_or_error(self) -> str:
@@ -186,7 +185,7 @@ class UserModel(BaseModel):
     token: Mapped[DecryptedString] = mapped_column(EncryptedString(200), unique=True)
     # token_hash is needed for fast search by token when stored token is encrypted
     token_hash: Mapped[str] = mapped_column(String(2000), unique=True)
-    global_role: Mapped[GlobalRole] = mapped_column(Enum(GlobalRole))
+    global_role: Mapped[GlobalRole] = mapped_column(EnumAsString(GlobalRole, 100))
     # deactivated users cannot access API
     active: Mapped[bool] = mapped_column(Boolean, default=True)
 
@@ -247,7 +246,7 @@ class MemberModel(BaseModel):
     project: Mapped["ProjectModel"] = relationship()
     user_id: Mapped[uuid.UUID] = mapped_column(ForeignKey("users.id", ondelete="CASCADE"))
     user: Mapped[UserModel] = relationship(lazy="joined")
-    project_role: Mapped[ProjectRole] = mapped_column(Enum(ProjectRole))
+    project_role: Mapped[ProjectRole] = mapped_column(EnumAsString(ProjectRole, 100))
     # member_num defines members ordering
     member_num: Mapped[Optional[int]] = mapped_column(Integer)
 
@@ -279,7 +278,7 @@ class RepoModel(BaseModel):
     project: Mapped["ProjectModel"] = relationship()
     # RepoModel.name stores repo_id
     name: Mapped[str] = mapped_column(String(100))
-    type: Mapped[RepoType] = mapped_column(Enum(RepoType))
+    type: Mapped[RepoType] = mapped_column(EnumAsString(RepoType, 100))
 
     info: Mapped[str] = mapped_column(Text)
 
@@ -360,9 +359,9 @@ class RunModel(BaseModel):
     submitted_at: Mapped[datetime] = mapped_column(NaiveDateTime)
     last_processed_at: Mapped[datetime] = mapped_column(NaiveDateTime)
     next_triggered_at: Mapped[Optional[datetime]] = mapped_column(NaiveDateTime)
-    status: Mapped[RunStatus] = mapped_column(Enum(RunStatus), index=True)
+    status: Mapped[RunStatus] = mapped_column(EnumAsString(RunStatus, 100), index=True)
     termination_reason: Mapped[Optional[RunTerminationReason]] = mapped_column(
-        Enum(RunTerminationReason)
+        EnumAsString(RunTerminationReason, 100)
     )
     # resubmission_attempt counts consecutive transitions to pending without provisioning.
     # Can be used to choose retry delay depending on the attempt number.
@@ -391,19 +390,27 @@ class JobModel(BaseModel):
     id: Mapped[uuid.UUID] = mapped_column(
         UUIDType(binary=False), primary_key=True, default=uuid.uuid4
     )
+
     project_id: Mapped[uuid.UUID] = mapped_column(ForeignKey("projects.id", ondelete="CASCADE"))
     project: Mapped["ProjectModel"] = relationship()
+
     run_id: Mapped[uuid.UUID] = mapped_column(ForeignKey("runs.id", ondelete="CASCADE"))
     run: Mapped["RunModel"] = relationship()
+
+    # Jobs need to reference fleets because we may choose an optimal fleet for a master job
+    # but not yet create an instance for it.
+    fleet_id: Mapped[Optional[uuid.UUID]] = mapped_column(ForeignKey("fleets.id"))
+    fleet: Mapped[Optional["FleetModel"]] = relationship(back_populates="jobs")
+
     run_name: Mapped[str] = mapped_column(String(100))
     job_num: Mapped[int] = mapped_column(Integer)
     job_name: Mapped[str] = mapped_column(String(100))
     submission_num: Mapped[int] = mapped_column(Integer)
     submitted_at: Mapped[datetime] = mapped_column(NaiveDateTime)
     last_processed_at: Mapped[datetime] = mapped_column(NaiveDateTime)
-    status: Mapped[JobStatus] = mapped_column(Enum(JobStatus), index=True)
+    status: Mapped[JobStatus] = mapped_column(EnumAsString(JobStatus, 100), index=True)
     termination_reason: Mapped[Optional[JobTerminationReason]] = mapped_column(
-        Enum(JobTerminationReason)
+        EnumAsString(JobTerminationReason, 100)
     )
     termination_reason_message: Mapped[Optional[str]] = mapped_column(Text)
     # `disconnected_at` stores the first time of connectivity issues with the instance.
@@ -431,6 +438,9 @@ class JobModel(BaseModel):
     probes: Mapped[list["ProbeModel"]] = relationship(
         back_populates="job", order_by="ProbeModel.probe_num"
     )
+    # Whether the replica is registered to receive service requests.
+    # Always `False` for non-service runs.
+    registered: Mapped[bool] = mapped_column(Boolean, server_default=false())
 
 
 class GatewayModel(BaseModel):
@@ -446,7 +456,7 @@ class GatewayModel(BaseModel):
     # Use `get_gateway_configuration` to construct `configuration` for old gateways.
     configuration: Mapped[Optional[str]] = mapped_column(Text)
     created_at: Mapped[datetime] = mapped_column(NaiveDateTime, default=get_current_datetime)
-    status: Mapped[GatewayStatus] = mapped_column(Enum(GatewayStatus))
+    status: Mapped[GatewayStatus] = mapped_column(EnumAsString(GatewayStatus, 100))
     status_message: Mapped[Optional[str]] = mapped_column(Text)
     last_processed_at: Mapped[datetime] = mapped_column(NaiveDateTime)
 
@@ -532,12 +542,13 @@ class FleetModel(BaseModel):
     deleted: Mapped[bool] = mapped_column(Boolean, default=False)
     deleted_at: Mapped[Optional[datetime]] = mapped_column(NaiveDateTime)
 
-    status: Mapped[FleetStatus] = mapped_column(Enum(FleetStatus), index=True)
+    status: Mapped[FleetStatus] = mapped_column(EnumAsString(FleetStatus, 100), index=True)
     status_message: Mapped[Optional[str]] = mapped_column(Text)
 
     spec: Mapped[str] = mapped_column(Text)
 
     runs: Mapped[List["RunModel"]] = relationship(back_populates="fleet")
+    jobs: Mapped[List["JobModel"]] = relationship(back_populates="fleet")
     instances: Mapped[List["InstanceModel"]] = relationship(back_populates="fleet")
 
 
@@ -571,7 +582,7 @@ class InstanceModel(BaseModel):
     fleet_id: Mapped[Optional[uuid.UUID]] = mapped_column(ForeignKey("fleets.id"))
     fleet: Mapped[Optional["FleetModel"]] = relationship(back_populates="instances")
 
-    status: Mapped[InstanceStatus] = mapped_column(Enum(InstanceStatus), index=True)
+    status: Mapped[InstanceStatus] = mapped_column(EnumAsString(InstanceStatus, 100), index=True)
     unreachable: Mapped[bool] = mapped_column(Boolean)
 
     # VM
@@ -672,7 +683,7 @@ class VolumeModel(BaseModel):
     deleted: Mapped[bool] = mapped_column(Boolean, default=False)
     deleted_at: Mapped[Optional[datetime]] = mapped_column(NaiveDateTime)
 
-    status: Mapped[VolumeStatus] = mapped_column(Enum(VolumeStatus), index=True)
+    status: Mapped[VolumeStatus] = mapped_column(EnumAsString(VolumeStatus, 100), index=True)
     status_message: Mapped[Optional[str]] = mapped_column(Text)
 
     configuration: Mapped[str] = mapped_column(Text)

dstack/_internal/server/routers/gpus.py

@@ -0,0 +1,29 @@
+from typing import Tuple
+
+from fastapi import APIRouter, Depends
+from sqlalchemy.ext.asyncio import AsyncSession
+
+from dstack._internal.server.db import get_session
+from dstack._internal.server.models import ProjectModel, UserModel
+from dstack._internal.server.schemas.gpus import ListGpusRequest, ListGpusResponse
+from dstack._internal.server.security.permissions import ProjectMember
+from dstack._internal.server.services.gpus import list_gpus_grouped
+from dstack._internal.server.utils.routers import get_base_api_additional_responses
+
+project_router = APIRouter(
+    prefix="/api/project/{project_name}/gpus",
+    tags=["gpus"],
+    responses=get_base_api_additional_responses(),
+)
+
+
+@project_router.post("/list", response_model=ListGpusResponse, response_model_exclude_none=True)
+async def list_gpus(
+    body: ListGpusRequest,
+    session: AsyncSession = Depends(get_session),
+    user_project: Tuple[UserModel, ProjectModel] = Depends(ProjectMember()),
+) -> ListGpusResponse:
+    _, project = user_project
+    return await list_gpus_grouped(
+        session=session, project=project, run_spec=body.run_spec, group_by=body.group_by
+    )