Drop fleet default min resources (#3776)

* Drop fleet default min resources

* Fix memory display for <1GB cases

Author: r4victor

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

@@ -248,7 +248,7 @@ class InstanceGroupParams(CoreModel):
     resources: Annotated[
         Optional[ResourcesSpec],
         Field(description="The resources requirements"),
-    ] = ResourcesSpec()
+    ] = None
 
     blocks: Annotated[
         Union[Literal["auto"], int],

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

@@ -14,7 +14,7 @@
 from dstack._internal.core.models.envs import Env
 from dstack._internal.core.models.health import HealthStatus
 from dstack._internal.core.models.volumes import Volume
-from dstack._internal.utils.common import pretty_resources
+from dstack._internal.utils.common import format_mib_as_gb, pretty_resources
 from dstack._internal.utils.logging import get_logger
 
 logger = get_logger(__name__)
@@ -135,7 +135,7 @@ def _pretty_format(
                 "gpu_count": len(gpus),
             }
             if gpu.memory_mib > 0:
-                gpu_resources["gpu_memory"] = f"{gpu.memory_mib / 1024:.0f}GB"
+                gpu_resources["gpu_memory"] = format_mib_as_gb(gpu.memory_mib)
             output = pretty_resources(**gpu_resources)
             if include_spot and spot:
                 output += " (spot)"
@@ -146,15 +146,15 @@ def _pretty_format(
             resources["cpus"] = cpus
             resources["cpu_arch"] = cpu_arch
         if memory_mib > 0:
-            resources["memory"] = f"{memory_mib / 1024:.0f}GB"
+            resources["memory"] = format_mib_as_gb(memory_mib)
         if disk_size_mib > 0:
-            resources["disk_size"] = f"{disk_size_mib / 1024:.0f}GB"
+            resources["disk_size"] = format_mib_as_gb(disk_size_mib)
         if gpus:
             gpu = gpus[0]
             resources["gpu_name"] = gpu.name
             resources["gpu_count"] = len(gpus)
             if gpu.memory_mib > 0:
-                resources["gpu_memory"] = f"{gpu.memory_mib / 1024:.0f}GB"
+                resources["gpu_memory"] = format_mib_as_gb(gpu.memory_mib)
         output = pretty_resources(**resources)
         if include_spot and spot:
             output += " (spot)"

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

@@ -394,6 +394,16 @@ class ResourcesSpec(generate_dual_core_model(ResourcesSpecConfig)):
     """`gpu` is optional for backward compatibility."""
     disk: Annotated[Optional[DiskSpec], Field(description="The disk resources")] = DEFAULT_DISK
 
+    @classmethod
+    def unconstrained(cls) -> "ResourcesSpec":
+        """ResourcesSpec with no meaningful minimum constraints."""
+        return cls(
+            cpu=CPUSpec(count=Range[int](min=1, max=None)),
+            memory=Range[Memory](min=Memory.parse("0"), max=None),
+            gpu=DEFAULT_GPU_SPEC,
+            disk=None,
+        )
+
     def pretty_format(self) -> str:
         # TODO: Remove in 0.20. Use self.cpu directly
         cpu = parse_obj_as(CPUSpec, self.cpu)

src/dstack/_internal/server/services/fleets.py

@@ -937,8 +937,11 @@ def is_cloud_cluster(fleet_model: FleetModel) -> bool:
 
 def get_fleet_requirements(fleet_spec: FleetSpec) -> Requirements:
     profile = fleet_spec.merged_profile
+    resources = fleet_spec.configuration.resources
+    if resources is None:
+        resources = ResourcesSpec.unconstrained()
     requirements = Requirements(
-        resources=fleet_spec.configuration.resources or ResourcesSpec(),
+        resources=resources,
         max_price=profile.max_price,
         spot=get_policy_map(profile.spot_policy, default=SpotPolicy.ONDEMAND),
         reservation=fleet_spec.configuration.reservation,

src/dstack/_internal/utils/common.py

@@ -112,6 +112,11 @@ def pretty_date(time: datetime) -> str:
     return str(years) + " years ago"
 
 
+def format_mib_as_gb(mib: int) -> str:
+    """Format a MiB value as a human-readable GB string, e.g. 512 → '0.5GB', 8192 → '8GB'."""
+    return f"{round(mib / 1024, 1):g}GB"
+
+
 def pretty_resources(
     *,
     cpu_arch: Optional[Any] = None,

src/tests/_internal/server/routers/test_fleets.py

@@ -936,20 +936,7 @@ async def test_creates_fleet(self, test_db, session: AsyncSession, client: Async
                     "placement": None,
                     "env": {},
                     "ssh_config": None,
-                    "resources": {
-                        "cpu": {"min": 2, "max": None},
-                        "memory": {"min": 8.0, "max": None},
-                        "shm_size": None,
-                        "gpu": {
-                            "vendor": None,
-                            "name": None,
-                            "count": {"min": 0, "max": None},
-                            "memory": None,
-                            "total_memory": None,
-                            "compute_capability": None,
-                        },
-                        "disk": {"size": {"min": 100.0, "max": None}},
-                    },
+                    "resources": None,
                     "backends": None,
                     "regions": None,
                     "availability_zones": None,
@@ -1067,20 +1054,7 @@ async def test_creates_ssh_fleet(self, test_db, session: AsyncSession, client: A
                     },
                     "nodes": None,
                     "placement": None,
-                    "resources": {
-                        "cpu": {"min": 2, "max": None},
-                        "memory": {"min": 8.0, "max": None},
-                        "shm_size": None,
-                        "gpu": {
-                            "vendor": None,
-                            "name": None,
-                            "count": {"min": 0, "max": None},
-                            "memory": None,
-                            "total_memory": None,
-                            "compute_capability": None,
-                        },
-                        "disk": {"size": {"min": 100.0, "max": None}},
-                    },
+                    "resources": None,
                     "backends": None,
                     "regions": None,
                     "availability_zones": None,
@@ -1297,20 +1271,7 @@ async def test_updates_ssh_fleet(self, test_db, session: AsyncSession, client: A
                     },
                     "nodes": None,
                     "placement": None,
-                    "resources": {
-                        "cpu": {"min": 2, "max": None},
-                        "memory": {"min": 8.0, "max": None},
-                        "shm_size": None,
-                        "gpu": {
-                            "vendor": None,
-                            "name": None,
-                            "count": {"min": 0, "max": None},
-                            "memory": None,
-                            "total_memory": None,
-                            "compute_capability": None,
-                        },
-                        "disk": {"size": {"min": 100.0, "max": None}},
-                    },
+                    "resources": None,
                     "backends": None,
                     "regions": None,
                     "availability_zones": None,

src/tests/_internal/server/services/requirements/test_combine.py

@@ -164,6 +164,29 @@ def test_combines_requirements(
             == expected_requirements
         )
 
+    def test_unconstrained_fleet_resources_pass_through_run_requirements(self):
+        unconstrained_fleet = Requirements(
+            resources=ResourcesSpec.unconstrained(),
+        )
+        run = Requirements(
+            resources=ResourcesSpec(
+                cpu=CPUSpec(count=Range(min=2, max=None)),
+                memory=Range(min=Memory.parse("2GB"), max=None),
+                gpu=GPUSpec(count=Range(min=1, max=None)),
+                disk=DiskSpec(size=Range(min=Memory.parse("50GB"), max=None)),
+            ),
+        )
+        result = combine_fleet_and_run_requirements(unconstrained_fleet, run)
+        assert result is not None
+        combined_cpu = result.resources.cpu
+        assert isinstance(combined_cpu, CPUSpec)
+        assert combined_cpu.count.min == 2
+        assert result.resources.memory.min == Memory.parse("2GB")
+        assert result.resources.gpu is not None
+        assert result.resources.gpu.count.min == 1
+        assert result.resources.disk is not None
+        assert result.resources.disk.size.min == Memory.parse("50GB")
+
 
 class TestIntersectLists:
     def test_both_none_returns_none(self):