Drop pools (#2401)

* Remove pools from the CLI and Python API

* Remove pools from HTTP API and profiles/configurations

* Do not require pools for instances

Author: r4victor

src/dstack/_internal/cli/commands/pool.py

@@ -13,7 +13,6 @@
 from dstack._internal.cli.commands.gateway import GatewayCommand
 from dstack._internal.cli.commands.init import InitCommand
 from dstack._internal.cli.commands.logs import LogsCommand
-from dstack._internal.cli.commands.pool import PoolCommand
 from dstack._internal.cli.commands.ps import PsCommand
 from dstack._internal.cli.commands.run import RunCommand
 from dstack._internal.cli.commands.server import ServerCommand
@@ -65,7 +64,6 @@ def main():
     DeleteCommand.register(subparsers)
     FleetCommand.register(subparsers)
     GatewayCommand.register(subparsers)
-    PoolCommand.register(subparsers)
     InitCommand.register(subparsers)
     LogsCommand.register(subparsers)
     PsCommand.register(subparsers)

src/dstack/_internal/cli/main.py

@@ -101,8 +101,6 @@ def apply_configuration(
                 max_price=profile.max_price,
                 working_dir=conf.working_dir,
                 run_name=conf.name,
-                pool_name=profile.pool_name,
-                instance_name=profile.instance_name,
                 creation_policy=profile.creation_policy,
                 termination_policy=profile.termination_policy,
                 termination_policy_idle=profile.termination_idle_time,

src/dstack/_internal/cli/services/configurators/run.py

@@ -4,8 +4,6 @@
 
 from dstack._internal.core.models.configurations import AnyRunConfiguration
 from dstack._internal.core.models.profiles import (
-    DEFAULT_INSTANCE_RETRY_DURATION,
-    DEFAULT_POOL_TERMINATION_IDLE_TIME,
     CreationPolicy,
     Profile,
     ProfileRetryPolicy,
@@ -16,9 +14,9 @@
 )
 
 
-def register_profile_args(parser: argparse.ArgumentParser, pool_add: bool = False):
+def register_profile_args(parser: argparse.ArgumentParser):
     """
-    Registers `parser` with `dstack run` and `dstack pool add`
+    Registers `parser` with `dstack apply` run configuration
     CLI arguments that override `profiles.yml` settings.
     """
     profile_group = parser.add_argument_group("Profile")
@@ -36,14 +34,13 @@ def register_profile_args(parser: argparse.ArgumentParser, pool_add: bool = Fals
         help="The maximum price per hour, in dollars",
         dest="max_price",
     )
-    if not pool_add:
-        profile_group.add_argument(
-            "--max-duration",
-            type=max_duration,
-            dest="max_duration",
-            help="The maximum duration of the run",
-            metavar="DURATION",
-        )
+    profile_group.add_argument(
+        "--max-duration",
+        type=max_duration,
+        dest="max_duration",
+        help="The maximum duration of the run",
+        metavar="DURATION",
+    )
     profile_group.add_argument(
         "-b",
         "--backend",
@@ -67,42 +64,28 @@ def register_profile_args(parser: argparse.ArgumentParser, pool_add: bool = Fals
         dest="instance_types",
         help="The cloud-specific instance types that will be tried for provisioning",
     )
-    if pool_add:
-        pools_group_exc = parser
-    else:
-        pools_group = parser.add_argument_group("Pools")
-        pools_group_exc = pools_group.add_mutually_exclusive_group()
-    pools_group_exc.add_argument(
-        "--pool",
-        dest="pool_name",
-        help="The name of the pool. If not set, the default pool will be used",
-    )
-    pools_group_exc.add_argument(
+
+    fleets_group = parser.add_argument_group("Fleets")
+    fleets_group_exc = fleets_group.add_mutually_exclusive_group()
+    fleets_group_exc.add_argument(
         "-R",
         "--reuse",
         dest="creation_policy_reuse",
         action="store_true",
-        help="Reuse instance from pool",
+        help="Reuse an existing instance from fleet (do not provision a new one)",
     )
-    pools_group_exc.add_argument(
+    fleets_group_exc.add_argument(
         "--dont-destroy",
         dest="dont_destroy",
         action="store_true",
-        help="Do not destroy instance after the run is finished",
+        help="Do not destroy instance after the run is finished (if the run provisions a new instance)",
     )
-    pools_group_exc.add_argument(
+    fleets_group_exc.add_argument(
         "--idle-duration",
         dest="idle_duration",
         type=str,
-        help="Time to wait before destroying the idle instance",
+        help="Time to wait before destroying the idle instance (if the run provisions a new instance)",
     )
-    if not pool_add:
-        pools_group_exc.add_argument(
-            "--instance",
-            dest="instance_name",
-            metavar="NAME",
-            help="Reuse instance from pool with name [code]NAME[/]",
-        )
 
     spot_group = parser.add_argument_group("Spot policy")
     spot_group_exc = spot_group.add_mutually_exclusive_group()
@@ -149,7 +132,6 @@ def register_profile_args(parser: argparse.ArgumentParser, pool_add: bool = Fals
 def apply_profile_args(
     args: argparse.Namespace,
     profile_settings: Union[Profile, AnyRunConfiguration],
-    pool_add: bool = False,
 ):
     """
     Overrides `profile_settings` settings with arguments registered by `register_profile_args()`.
@@ -165,53 +147,29 @@ def apply_profile_args(
         profile_settings.instance_types = args.instance_types
     if args.max_price is not None:
         profile_settings.max_price = args.max_price
-    if not pool_add:
-        if args.max_duration is not None:
-            profile_settings.max_duration = args.max_duration
-
-    if args.pool_name:
-        profile_settings.pool_name = args.pool_name
+    if args.max_duration is not None:
+        profile_settings.max_duration = args.max_duration
 
     if args.idle_duration is not None:
         profile_settings.termination_idle_time = args.idle_duration
-    if pool_add and args.idle_duration is None:
-        profile_settings.termination_idle_time = DEFAULT_POOL_TERMINATION_IDLE_TIME
 
     if args.dont_destroy:
         profile_settings.termination_policy = TerminationPolicy.DONT_DESTROY
-    if not pool_add:
-        if args.instance_name:
-            profile_settings.instance_name = args.instance_name
-        if args.creation_policy_reuse:
-            profile_settings.creation_policy = CreationPolicy.REUSE
+    if args.creation_policy_reuse:
+        profile_settings.creation_policy = CreationPolicy.REUSE
 
     if args.spot_policy is not None:
         profile_settings.spot_policy = args.spot_policy
-    if pool_add and args.spot_policy is None:  # ONDEMAND by default for `dstack pool add`
-        profile_settings.spot_policy = SpotPolicy.ONDEMAND
-
-    if not pool_add:
-        if args.retry_policy is not None:
-            if not profile_settings.retry_policy:
-                profile_settings.retry_policy = ProfileRetryPolicy()
-            profile_settings.retry_policy.retry = args.retry_policy
-        elif args.retry_duration is not None:
-            if not profile_settings.retry_policy:
-                profile_settings.retry_policy = ProfileRetryPolicy()
-            profile_settings.retry_policy.retry = True
-            profile_settings.retry_policy.duration = args.retry_duration
-    else:
-        if args.retry_policy is not None:
-            if not profile_settings.retry_policy:
-                profile_settings.retry_policy = ProfileRetryPolicy()
-            profile_settings.retry_policy.retry = args.retry_policy
-            if profile_settings.retry_policy.retry:
-                profile_settings.retry_policy.duration = DEFAULT_INSTANCE_RETRY_DURATION
-        elif args.retry_duration is not None:
-            if not profile_settings.retry_policy:
-                profile_settings.retry_policy = ProfileRetryPolicy()
-            profile_settings.retry_policy.retry = True
-            profile_settings.retry_policy.duration = args.retry_duration  # --retry-duration
+
+    if args.retry_policy is not None:
+        if not profile_settings.retry_policy:
+            profile_settings.retry_policy = ProfileRetryPolicy()
+        profile_settings.retry_policy.retry = args.retry_policy
+    elif args.retry_duration is not None:
+        if not profile_settings.retry_policy:
+            profile_settings.retry_policy = ProfileRetryPolicy()
+        profile_settings.retry_policy.retry = True
+        profile_settings.retry_policy.duration = args.retry_duration
 
 
 def max_duration(v: str) -> int:

src/dstack/_internal/cli/services/profile.py

@@ -10,8 +10,7 @@
 from dstack._internal.core.models.backends.base import BackendType
 from dstack._internal.core.models.common import CoreModel
 from dstack._internal.core.models.envs import Env
-from dstack._internal.core.models.instances import InstanceOfferWithAvailability, SSHKey
-from dstack._internal.core.models.pools import Instance
+from dstack._internal.core.models.instances import Instance, InstanceOfferWithAvailability, SSHKey
 from dstack._internal.core.models.profiles import (
     Profile,
     ProfileParams,

src/dstack/_internal/core/models/fleets.py

@@ -1,5 +1,7 @@
+import datetime
 from enum import Enum
 from typing import List, Optional
+from uuid import UUID
 
 import gpuhunt
 from pydantic import root_validator
@@ -167,3 +169,25 @@ def is_active(self) -> bool:
     @classmethod
     def finished_statuses(cls) -> List["InstanceStatus"]:
         return [cls.TERMINATING, cls.TERMINATED]
+
+
+class Instance(CoreModel):
+    id: UUID
+    project_name: str
+    backend: Optional[BackendType] = None
+    instance_type: Optional[InstanceType] = None
+    name: str
+    fleet_id: Optional[UUID] = None
+    fleet_name: Optional[str] = None
+    instance_num: int
+    job_name: Optional[str] = None  # deprecated, always None (instance can have more than one job)
+    hostname: Optional[str] = None
+    status: InstanceStatus
+    unreachable: bool = False
+    termination_reason: Optional[str] = None
+    created: datetime.datetime
+    region: Optional[str] = None
+    availability_zone: Optional[str] = None
+    price: Optional[float] = None
+    total_blocks: Optional[int] = None
+    busy_blocks: int = 0

src/dstack/_internal/core/models/instances.py

@@ -8,12 +8,9 @@
 from dstack._internal.core.models.common import CoreModel, Duration
 
 DEFAULT_RETRY_DURATION = 3600
-DEFAULT_POOL_NAME = "default-pool"
 
 DEFAULT_RUN_TERMINATION_IDLE_TIME = 5 * 60  # 5 minutes
-DEFAULT_POOL_TERMINATION_IDLE_TIME = 72 * 60 * 60  # 3 days
-
-DEFAULT_INSTANCE_RETRY_DURATION = 60 * 60 * 24  # 24h
+DEFAULT_FLEET_TERMINATION_IDLE_TIME = 72 * 60 * 60  # 3 days
 
 DEFAULT_STOP_DURATION = 300
 
@@ -223,7 +220,7 @@ class ProfileParams(CoreModel):
     creation_policy: Annotated[
         Optional[CreationPolicy],
         Field(
-            description="The policy for using instances from the pool. Defaults to `reuse-or-create`"
+            description="The policy for using instances from fleets. Defaults to `reuse-or-create`"
         ),
     ]
     idle_duration: Annotated[
@@ -252,10 +249,6 @@ class ProfileParams(CoreModel):
             description="Deprecated in favor of `idle_duration`",
         ),
     ]
-    # The name of the pool. If not set, dstack will use the default name
-    pool_name: Optional[str]
-    # The name of the instance
-    instance_name: Optional[str]
     # The policy for resubmitting the run. Deprecated in favor of `retry`
     retry_policy: Optional[ProfileRetryPolicy]
 

src/dstack/_internal/core/models/pools.py

@@ -476,11 +476,6 @@ class ApplyRunPlanInput(CoreModel):
     ] = None
 
 
-class PoolInstanceOffers(CoreModel):
-    pool_name: str
-    instances: List[InstanceOfferWithAvailability]
-
-
 def get_policy_map(spot_policy: Optional[SpotPolicy], default: SpotPolicy) -> Optional[bool]:
     """
     Map profile.spot_policy[SpotPolicy|None] to requirements.spot[bool|None]

src/dstack/_internal/core/models/profiles.py

@@ -27,7 +27,6 @@
     instances,
     logs,
     metrics,
-    pools,
     projects,
     prometheus,
     repos,
@@ -184,8 +183,6 @@ def register_routes(app: FastAPI, ui: bool = True):
     app.include_router(volumes.project_router)
     app.include_router(service_proxy.router, prefix="/proxy/services", tags=["service-proxy"])
     app.include_router(model_proxy.router, prefix="/proxy/models", tags=["model-proxy"])
-    app.include_router(pools.root_router)
-    app.include_router(pools.router)
     app.include_router(prometheus.router)
 
     @app.exception_handler(ForbiddenError)