dstack 0.19.8__py3-none-any.whl → 0.19.10__py3-none-any.whl

dstack/_internal/server/background/tasks/process_submitted_jobs.py

@@ -93,11 +93,20 @@ async def _process_next_submitted_job():
         async with lock:
             res = await session.execute(
                 select(JobModel)
+                .join(JobModel.run)
                 .where(
                     JobModel.status == JobStatus.SUBMITTED,
                     JobModel.id.not_in(lockset),
                 )
-                .order_by(JobModel.last_processed_at.asc())
+                # Jobs are process in FIFO sorted by priority globally,
+                # thus runs from different project can "overtake" each other by using higher priorities.
+                # That's not a big problem as long as projects do not compete for the same compute resources.
+                # Jobs with lower priorities from other projects will be processed without major lag
+                # as long as new higher priority runs are not constantly submitted.
+                # TODO: Consider processing jobs from different projects fairly/round-robin
+                # Fully fair processing can be tricky to implement via the current DB queue as
+                # there can be many projects and we are limited by the max DB connections.
+                .order_by(RunModel.priority.desc(), JobModel.last_processed_at.asc())
                 .limit(1)
                 .with_for_update(skip_locked=True)
             )
@@ -360,16 +369,16 @@ async def _assign_job_to_pool_instance(
         (instance, common_utils.get_or_error(get_instance_offer(instance)))
         for instance in nonshared_instances
     ]
-    if not multinode:
-        shared_instances_with_offers = get_shared_pool_instances_with_offers(
-            pool_instances=pool_instances,
-            profile=profile,
-            requirements=job.job_spec.requirements,
-            idle_only=True,
-            fleet_model=fleet_model,
-            volumes=volumes,
-        )
-        instances_with_offers.extend(shared_instances_with_offers)
+    shared_instances_with_offers = get_shared_pool_instances_with_offers(
+        pool_instances=pool_instances,
+        profile=profile,
+        requirements=job.job_spec.requirements,
+        idle_only=True,
+        fleet_model=fleet_model,
+        multinode=multinode,
+        volumes=volumes,
+    )
+    instances_with_offers.extend(shared_instances_with_offers)
 
     if len(instances_with_offers) == 0:
         return None
@@ -572,7 +581,7 @@ def _create_instance_model_for_job(
 
 
 def _prepare_job_runtime_data(offer: InstanceOfferWithAvailability) -> JobRuntimeData:
-    if offer.total_blocks == 1:
+    if offer.blocks == offer.total_blocks:
         if env_utils.get_bool("DSTACK_FORCE_BRIDGE_NETWORK"):
             network_mode = NetworkMode.BRIDGE
         else:

dstack/_internal/server/migrations/versions/20166748b60c_add_jobmodel_disconnected_at.py

@@ -0,0 +1,100 @@
+"""Add JobModel.disconnected_at
+
+Revision ID: 20166748b60c
+Revises: 6c1a9d6530ee
+Create Date: 2025-05-13 16:24:32.496578
+
+"""
+
+import sqlalchemy as sa
+from alembic import op
+from alembic_postgresql_enum import TableReference
+
+import dstack._internal.server.models
+
+# revision identifiers, used by Alembic.
+revision = "20166748b60c"
+down_revision = "6c1a9d6530ee"
+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(
+                "disconnected_at", dstack._internal.server.models.NaiveDateTime(), nullable=True
+            )
+        )
+
+    op.sync_enum_values(
+        enum_schema="public",
+        enum_name="jobterminationreason",
+        new_values=[
+            "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",
+        ],
+        affected_columns=[
+            TableReference(
+                table_schema="public", table_name="jobs", column_name="termination_reason"
+            )
+        ],
+        enum_values_to_rename=[],
+    )
+    # ### end Alembic commands ###
+
+
+def downgrade() -> None:
+    # ### commands auto generated by Alembic - please adjust! ###
+    op.sync_enum_values(
+        enum_schema="public",
+        enum_name="jobterminationreason",
+        new_values=[
+            "FAILED_TO_START_DUE_TO_NO_CAPACITY",
+            "INTERRUPTED_BY_NO_CAPACITY",
+            "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",
+        ],
+        affected_columns=[
+            TableReference(
+                table_schema="public", table_name="jobs", column_name="termination_reason"
+            )
+        ],
+        enum_values_to_rename=[],
+    )
+    with op.batch_alter_table("jobs", schema=None) as batch_op:
+        batch_op.drop_column("disconnected_at")
+
+    # ### end Alembic commands ###

dstack/_internal/server/migrations/versions/6c1a9d6530ee_add_jobmodel_exit_status.py

@@ -0,0 +1,26 @@
+"""Add JobModel.exit_status
+
+Revision ID: 6c1a9d6530ee
+Revises: 7ba3b59d7ca6
+Create Date: 2025-05-09 10:25:19.715852
+
+"""
+
+import sqlalchemy as sa
+from alembic import op
+
+# revision identifiers, used by Alembic.
+revision = "6c1a9d6530ee"
+down_revision = "7ba3b59d7ca6"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    with op.batch_alter_table("jobs", schema=None) as batch_op:
+        batch_op.add_column(sa.Column("exit_status", sa.Integer(), nullable=True))
+
+
+def downgrade() -> None:
+    with op.batch_alter_table("jobs", schema=None) as batch_op:
+        batch_op.drop_column("exit_status")

dstack/_internal/server/migrations/versions/bca2fdf130bf_add_runmodel_priority.py

@@ -0,0 +1,34 @@
+"""Add RunModel.priority
+
+Revision ID: bca2fdf130bf
+Revises: 20166748b60c
+Create Date: 2025-05-14 15:24:21.269775
+
+"""
+
+import sqlalchemy as sa
+from alembic import op
+
+# revision identifiers, used by Alembic.
+revision = "bca2fdf130bf"
+down_revision = "20166748b60c"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    # ### commands auto generated by Alembic - please adjust! ###
+    with op.batch_alter_table("runs", schema=None) as batch_op:
+        batch_op.add_column(sa.Column("priority", sa.Integer(), nullable=True))
+    batch_op.execute("UPDATE runs SET priority = 0")
+    with op.batch_alter_table("runs", schema=None) as batch_op:
+        batch_op.alter_column("priority", nullable=False)
+    # ### end Alembic commands ###
+
+
+def downgrade() -> None:
+    # ### commands auto generated by Alembic - please adjust! ###
+    with op.batch_alter_table("runs", schema=None) as batch_op:
+        batch_op.drop_column("priority")
+
+    # ### end Alembic commands ###

dstack/_internal/server/models.py

@@ -348,6 +348,7 @@ class RunModel(BaseModel):
     resubmission_attempt: Mapped[int] = mapped_column(Integer, default=0)
     run_spec: Mapped[str] = mapped_column(Text)
     service_spec: Mapped[Optional[str]] = mapped_column(Text)
+    priority: Mapped[int] = mapped_column(Integer, default=0)
 
     jobs: Mapped[List["JobModel"]] = relationship(
         back_populates="run", lazy="selectin", order_by="[JobModel.replica_num, JobModel.job_num]"
@@ -382,6 +383,10 @@ class JobModel(BaseModel):
         Enum(JobTerminationReason)
     )
     termination_reason_message: Mapped[Optional[str]] = mapped_column(Text)
+    # `disconnected_at` stores the first time of connectivity issues with the instance.
+    # Resets every time connectivity is restored.
+    disconnected_at: Mapped[Optional[datetime]] = mapped_column(NaiveDateTime)
+    exit_status: Mapped[Optional[int]] = mapped_column(Integer)
     job_spec_data: Mapped[str] = mapped_column(Text)
     job_provisioning_data: Mapped[Optional[str]] = mapped_column(Text)
     runner_timestamp: Mapped[Optional[int]] = mapped_column(BigInteger)
@@ -390,7 +395,7 @@ class JobModel(BaseModel):
     remove_at: Mapped[Optional[datetime]] = mapped_column(NaiveDateTime)
     volumes_detached_at: Mapped[Optional[datetime]] = mapped_column(NaiveDateTime)
     # `instance_assigned` means instance assignment was done.
-    # if `instance_assigned` is True and `instance` is None, no instance was assiged.
+    # if `instance_assigned` is True and `instance` is None, no instance was assigned.
     instance_assigned: Mapped[bool] = mapped_column(Boolean, default=False)
     instance_id: Mapped[Optional[uuid.UUID]] = mapped_column(
         ForeignKey("instances.id", ondelete="CASCADE")

dstack/_internal/server/routers/repos.py

@@ -1,6 +1,7 @@
 from typing import List, Tuple
 
 from fastapi import APIRouter, Depends, Request, UploadFile
+from humanize import naturalsize
 from sqlalchemy.ext.asyncio import AsyncSession
 
 from dstack._internal.core.errors import ResourceNotExistsError, ServerClientError
@@ -14,9 +15,10 @@ from dstack._internal.server.schemas.repos import (
 )
 from dstack._internal.server.security.permissions import ProjectMember
 from dstack._internal.server.services import repos
+from dstack._internal.server.settings import SERVER_CODE_UPLOAD_LIMIT
 from dstack._internal.server.utils.routers import (
     get_base_api_additional_responses,
-    request_size_exceeded,
+    get_request_size,
 )
 
 router = APIRouter(
@@ -94,10 +96,12 @@ async def upload_code(
     session: AsyncSession = Depends(get_session),
     user_project: Tuple[UserModel, ProjectModel] = Depends(ProjectMember()),
 ):
-    if request_size_exceeded(request, limit=2 * 2**20):
+    request_size = get_request_size(request)
+    if SERVER_CODE_UPLOAD_LIMIT > 0 and request_size > SERVER_CODE_UPLOAD_LIMIT:
         raise ServerClientError(
-            "Repo diff size exceeds the limit of 2MB. "
-            "Use .gitignore to exclude large files from the repo."
+            f"Repo diff size is {naturalsize(request_size)}, which exceeds the limit of "
+            f"{naturalsize(SERVER_CODE_UPLOAD_LIMIT)}. Use .gitignore to exclude large files from the repo. This "
+            f"limit can be modified by setting the DSTACK_SERVER_CODE_UPLOAD_LIMIT_BYTES environment variable"
         )
     _, project = user_project
     await repos.upload_code(

dstack/_internal/server/schemas/runner.py

@@ -7,7 +7,14 @@ from typing_extensions import Annotated
 
 from dstack._internal.core.models.common import CoreModel, NetworkMode
 from dstack._internal.core.models.repos.remote import RemoteRepoCreds
-from dstack._internal.core.models.runs import ClusterInfo, JobSpec, JobStatus, RunSpec
+from dstack._internal.core.models.runs import (
+    ClusterInfo,
+    JobSpec,
+    JobStatus,
+    JobSubmission,
+    Run,
+    RunSpec,
+)
 from dstack._internal.core.models.volumes import InstanceMountPoint, VolumeMountPoint
 
 
@@ -16,6 +23,7 @@ class JobStateEvent(CoreModel):
     state: JobStatus
     termination_reason: Optional[str] = None
     termination_message: Optional[str] = None
+    exit_status: Optional[int] = None
 
 
 class LogEvent(CoreModel):
@@ -38,15 +46,18 @@ class PullResponse(CoreModel):
 
 
 class SubmitBody(CoreModel):
-    run_spec: Annotated[
-        RunSpec,
+    run: Annotated[
+        Run,
         Field(
             include={
-                "run_name",
-                "repo_id",
-                "repo_data",
-                "configuration",
-                "configuration_path",
+                "id": True,
+                "run_spec": {
+                    "run_name",
+                    "repo_id",
+                    "repo_data",
+                    "configuration",
+                    "configuration_path",
+                },
             }
         ),
     ]
@@ -69,9 +80,31 @@ class SubmitBody(CoreModel):
             }
         ),
     ]
+    job_submission: Annotated[
+        JobSubmission,
+        Field(
+            include={
+                "id",
+            }
+        ),
+    ]
     cluster_info: Annotated[Optional[ClusterInfo], Field(include=True)]
     secrets: Annotated[Optional[Dict[str, str]], Field(include=True)]
     repo_credentials: Annotated[Optional[RemoteRepoCreds], Field(include=True)]
+    # run_spec is deprecated in favor of run.run_spec
+    # TODO: Remove once we no longer support instances deployed with 0.19.8 or earlier.
+    run_spec: Annotated[
+        RunSpec,
+        Field(
+            include={
+                "run_name",
+                "repo_id",
+                "repo_data",
+                "configuration",
+                "configuration_path",
+            },
+        ),
+    ]
 
 
 class HealthcheckResponse(CoreModel):

dstack/_internal/server/services/instances.py

@@ -235,6 +235,7 @@ def get_shared_pool_instances_with_offers(
     *,
     idle_only: bool = False,
     fleet_model: Optional[FleetModel] = None,
+    multinode: bool = False,
     volumes: Optional[List[List[Volume]]] = None,
 ) -> list[tuple[InstanceModel, InstanceOfferWithAvailability]]:
     instances_with_offers: list[tuple[InstanceModel, InstanceOfferWithAvailability]] = []
@@ -243,19 +244,22 @@ def get_shared_pool_instances_with_offers(
         pool_instances=pool_instances,
         profile=profile,
         fleet_model=fleet_model,
-        multinode=False,
+        multinode=multinode,
         volumes=volumes,
         shared=True,
     )
     for instance in filtered_instances:
         if idle_only and instance.status not in [InstanceStatus.IDLE, InstanceStatus.BUSY]:
             continue
+        if multinode and instance.busy_blocks > 0:
+            continue
         offer = get_instance_offer(instance)
         if offer is None:
             continue
         total_blocks = common_utils.get_or_error(instance.total_blocks)
         idle_blocks = total_blocks - instance.busy_blocks
-        for blocks in range(1, total_blocks + 1):
+        min_blocks = total_blocks if multinode else 1
+        for blocks in range(min_blocks, total_blocks + 1):
             shared_offer = generate_shared_offer(offer, blocks, total_blocks)
             catalog_item = offer_to_catalog_item(shared_offer)
             if gpuhunt.matches(catalog_item, query_filter):

dstack/_internal/server/services/jobs/__init__.py

@@ -135,6 +135,7 @@ def job_model_to_job_submission(job_model: JobModel) -> JobSubmission:
         status=job_model.status,
         termination_reason=job_model.termination_reason,
         termination_reason_message=job_model.termination_reason_message,
+        exit_status=job_model.exit_status,
         job_provisioning_data=job_provisioning_data,
         job_runtime_data=get_job_runtime_data(job_model),
     )

dstack/_internal/server/services/jobs/configurators/base.py

@@ -6,7 +6,7 @@ from typing import Dict, List, Optional, Union
 
 from cachetools import TTLCache, cached
 
-import dstack.version as version
+from dstack._internal import settings
 from dstack._internal.core.errors import DockerRegistryError, ServerClientError
 from dstack._internal.core.models.common import RegistryAuth
 from dstack._internal.core.models.configurations import (
@@ -53,14 +53,14 @@ def get_default_image(python_version: str, nvcc: bool = False) -> str:
     suffix = ""
     if nvcc:
         suffix = "-devel"
-    return f"dstackai/base:py{python_version}-{version.base_image}-cuda-12.1{suffix}"
+    return f"{settings.DSTACK_BASE_IMAGE}:py{python_version}-{settings.DSTACK_BASE_IMAGE_VERSION}-cuda-12.1{suffix}"
 
 
 class JobConfigurator(ABC):
     TYPE: RunConfigurationType
 
     _image_config: Optional[ImageConfig] = None
-    # JobSSHKey should be shared for all jobs in a replica for inter-node communitation.
+    # JobSSHKey should be shared for all jobs in a replica for inter-node communication.
     _job_ssh_key: Optional[JobSSHKey] = None
 
     def __init__(self, run_spec: RunSpec):

dstack/_internal/server/services/runner/client.py

@@ -12,7 +12,7 @@ from dstack._internal.core.models.common import CoreModel, NetworkMode
 from dstack._internal.core.models.envs import Env
 from dstack._internal.core.models.repos.remote import RemoteRepoCreds
 from dstack._internal.core.models.resources import Memory
-from dstack._internal.core.models.runs import ClusterInfo, JobSpec, RunSpec
+from dstack._internal.core.models.runs import ClusterInfo, Job, Run
 from dstack._internal.core.models.volumes import InstanceMountPoint, Volume, VolumeMountPoint
 from dstack._internal.server.schemas.runner import (
     GPUDevice,
@@ -72,8 +72,8 @@ class RunnerClient:
 
     def submit_job(
         self,
-        run_spec: RunSpec,
-        job_spec: JobSpec,
+        run: Run,
+        job: Job,
         cluster_info: ClusterInfo,
         secrets: Dict[str, str],
         repo_credentials: Optional[RemoteRepoCreds],
@@ -81,6 +81,7 @@ class RunnerClient:
     ):
         # XXX: This is a quick-and-dirty hack to deliver InstanceModel-specific environment
         # variables to the runner without runner API modification.
+        job_spec = job.job_spec
         if instance_env is not None:
             if isinstance(instance_env, Env):
                 merged_env = instance_env.as_dict()
@@ -90,11 +91,13 @@ class RunnerClient:
             job_spec = job_spec.copy(deep=True)
             job_spec.env = merged_env
         body = SubmitBody(
-            run_spec=run_spec,
+            run=run,
             job_spec=job_spec,
+            job_submission=job.job_submissions[-1],
             cluster_info=cluster_info,
             secrets=secrets,
             repo_credentials=repo_credentials,
+            run_spec=run.run_spec,
         )
         resp = requests.post(
             # use .json() to encode enums

dstack/_internal/server/services/runs.py

@@ -16,7 +16,7 @@ from dstack._internal.core.errors import (
     ServerClientError,
 )
 from dstack._internal.core.models.common import ApplyAction
-from dstack._internal.core.models.configurations import AnyRunConfiguration
+from dstack._internal.core.models.configurations import RUN_PRIORITY_DEFAULT, AnyRunConfiguration
 from dstack._internal.core.models.instances import (
     InstanceAvailability,
     InstanceOfferWithAvailability,
@@ -434,7 +434,12 @@ async def apply_plan(
     # FIXME: potentially long write transaction
     # Avoid getting run_model after update
     await session.execute(
-        update(RunModel).where(RunModel.id == current_resource.id).values(run_spec=run_spec.json())
+        update(RunModel)
+        .where(RunModel.id == current_resource.id)
+        .values(
+            run_spec=run_spec.json(),
+            priority=run_spec.configuration.priority,
+        )
     )
     run = await get_run_by_name(
         session=session,
@@ -495,6 +500,7 @@ async def submit_run(
             status=RunStatus.SUBMITTED,
             run_spec=run_spec.json(),
             last_processed_at=submitted_at,
+            priority=run_spec.configuration.priority,
         )
         session.add(run_model)
 
@@ -721,15 +727,15 @@ async def _get_pool_offers(
     pool_instances = [i for i in pool_instances if i.id not in detaching_instances_ids]
     multinode = job.job_spec.jobs_per_replica > 1
 
-    if not multinode:
-        shared_instances_with_offers = get_shared_pool_instances_with_offers(
-            pool_instances=pool_instances,
-            profile=run_spec.merged_profile,
-            requirements=job.job_spec.requirements,
-            volumes=volumes,
-        )
-        for _, offer in shared_instances_with_offers:
-            pool_offers.append(offer)
+    shared_instances_with_offers = get_shared_pool_instances_with_offers(
+        pool_instances=pool_instances,
+        profile=run_spec.merged_profile,
+        requirements=job.job_spec.requirements,
+        volumes=volumes,
+        multinode=multinode,
+    )
+    for _, offer in shared_instances_with_offers:
+        pool_offers.append(offer)
 
     nonshared_instances = filter_pool_instances(
         pool_instances=pool_instances,
@@ -852,6 +858,13 @@ def _get_job_submission_cost(job_submission: JobSubmission) -> float:
 
 
 def _validate_run_spec_and_set_defaults(run_spec: RunSpec):
+    # This function may set defaults for null run_spec values,
+    # although most defaults are resolved when building job_spec
+    # so that we can keep both the original user-supplied value (null in run_spec)
+    # and the default in job_spec.
+    # If a property is stored in job_spec - resolve the default there.
+    # Server defaults are preferable over client defaults so that
+    # the defaults depend on the server version, not the client version.
     if run_spec.run_name is not None:
         validate_dstack_resource_name(run_spec.run_name)
     for mount_point in run_spec.configuration.volumes:
@@ -870,16 +883,19 @@ def _validate_run_spec_and_set_defaults(run_spec: RunSpec):
     if (
         run_spec.merged_profile.utilization_policy is not None
         and run_spec.merged_profile.utilization_policy.time_window
-        > settings.SERVER_METRICS_TTL_SECONDS
+        > settings.SERVER_METRICS_RUNNING_TTL_SECONDS
     ):
         raise ServerClientError(
-            f"Maximum utilization_policy.time_window is {settings.SERVER_METRICS_TTL_SECONDS}s"
+            f"Maximum utilization_policy.time_window is {settings.SERVER_METRICS_RUNNING_TTL_SECONDS}s"
         )
+    if run_spec.configuration.priority is None:
+        run_spec.configuration.priority = RUN_PRIORITY_DEFAULT
     set_resources_defaults(run_spec.configuration.resources)
 
 
 _UPDATABLE_SPEC_FIELDS = ["repo_code_hash", "configuration"]
-_CONF_TYPE_TO_UPDATABLE_FIELDS = {
+_CONF_UPDATABLE_FIELDS = ["priority"]
+_TYPE_SPECIFIC_CONF_UPDATABLE_FIELDS = {
     "dev-environment": ["inactivity_duration"],
     # Most service fields can be updated via replica redeployment.
     # TODO: Allow updating other fields when rolling deployment is supported.
@@ -915,12 +931,9 @@ def _check_can_update_configuration(
         raise ServerClientError(
             f"Configuration type changed from {current.type} to {new.type}, cannot update"
         )
-    updatable_fields = _CONF_TYPE_TO_UPDATABLE_FIELDS.get(new.type)
-    if updatable_fields is None:
-        raise ServerClientError(
-            f"Can only update {', '.join(_CONF_TYPE_TO_UPDATABLE_FIELDS)} configurations."
-            f" Not {new.type}"
-        )
+    updatable_fields = _CONF_UPDATABLE_FIELDS + _TYPE_SPECIFIC_CONF_UPDATABLE_FIELDS.get(
+        new.type, []
+    )
     diff = diff_models(current, new)
     changed_fields = list(diff.keys())
     for key in changed_fields:

dstack/_internal/server/settings.py

@@ -1,4 +1,5 @@
 import os
+import warnings
 from pathlib import Path
 
 DSTACK_DIR_PATH = Path("~/.dstack/").expanduser()
@@ -45,7 +46,25 @@ SERVER_CLOUDWATCH_LOG_REGION = os.getenv("DSTACK_SERVER_CLOUDWATCH_LOG_REGION")
 
 SERVER_GCP_LOGGING_PROJECT = os.getenv("DSTACK_SERVER_GCP_LOGGING_PROJECT")
 
-SERVER_METRICS_TTL_SECONDS = int(os.getenv("DSTACK_SERVER_METRICS_TTL_SECONDS", 3600))
+SERVER_METRICS_RUNNING_TTL_SECONDS: int
+_SERVER_METRICS_RUNNING_TTL_SECONDS = os.getenv("DSTACK_SERVER_METRICS_RUNNING_TTL_SECONDS")
+if _SERVER_METRICS_RUNNING_TTL_SECONDS is None:
+    _SERVER_METRICS_RUNNING_TTL_SECONDS = os.getenv("DSTACK_SERVER_METRICS_TTL_SECONDS")
+    if _SERVER_METRICS_RUNNING_TTL_SECONDS is not None:
+        warnings.warn(
+            (
+                "DSTACK_SERVER_METRICS_TTL_SECONDS is deprecated,"
+                " use DSTACK_SERVER_METRICS_RUNNING_TTL_SECONDS instead"
+            ),
+            DeprecationWarning,
+        )
+    else:
+        _SERVER_METRICS_RUNNING_TTL_SECONDS = 3600
+SERVER_METRICS_RUNNING_TTL_SECONDS = int(_SERVER_METRICS_RUNNING_TTL_SECONDS)
+del _SERVER_METRICS_RUNNING_TTL_SECONDS
+SERVER_METRICS_FINISHED_TTL_SECONDS = int(
+    os.getenv("DSTACK_SERVER_METRICS_FINISHED_TTL_SECONDS", 7 * 24 * 3600)
+)
 
 DEFAULT_PROJECT_NAME = "main"
 
@@ -66,6 +85,7 @@ DEFAULT_SERVICE_CLIENT_MAX_BODY_SIZE = int(
 USER_PROJECT_DEFAULT_QUOTA = int(os.getenv("DSTACK_USER_PROJECT_DEFAULT_QUOTA", 10))
 FORBID_SERVICES_WITHOUT_GATEWAY = os.getenv("DSTACK_FORBID_SERVICES_WITHOUT_GATEWAY") is not None
 
+SERVER_CODE_UPLOAD_LIMIT = int(os.getenv("DSTACK_SERVER_CODE_UPLOAD_LIMIT", 2 * 2**20))
 
 # Development settings