Fix: Solved compilation issue and fixed gpu logic.

packaging/Makefile

@@ -16,7 +16,7 @@ debian-package-code:
 	cp -r ../examples/data ./aleph-vm/opt/aleph-vm/examples/data
 	mkdir -p ./aleph-vm/opt/aleph-vm/examples/volumes
 	# Fixing this protobuf dependency version to avoid getting CI errors as version 5.29.0 have this compilation issue
-	pip3 install --progress-bar off --target ./aleph-vm/opt/aleph-vm/ 'aleph-message==0.5.0' 'eth-account==0.10' 'sentry-sdk==1.31.0' 'qmp==1.1.0' 'aleph-superfluid~=0.2.1' 'sqlalchemy[asyncio]>=2.0' 'aiosqlite==0.19.0' 'alembic==1.13.1' 'aiohttp_cors==0.7.0' 'pyroute2==0.7.12' 'python-cpuid==0.1.0' 'solathon==1.0.2' 'protobuf==5.28.3'
+	pip3 install --progress-bar off --target ./aleph-vm/opt/aleph-vm/ 'aleph-message@git+https://github.com/aleph-im/aleph-message@andres-feature-add_gpu_requirement' 'eth-account==0.10' 'sentry-sdk==1.31.0' 'qmp==1.1.0' 'aleph-superfluid~=0.2.1' 'sqlalchemy[asyncio]>=2.0' 'aiosqlite==0.19.0' 'alembic==1.13.1' 'aiohttp_cors==0.7.0' 'pyroute2==0.7.12' 'python-cpuid==0.1.0' 'solathon==1.0.2' 'protobuf==5.28.3'
 	python3 -m compileall ./aleph-vm/opt/aleph-vm/
 
 debian-package-resources: firecracker-bins vmlinux download-ipfs-kubo  target/bin/sevctl

src/aleph/vm/models.py

@@ -13,7 +13,7 @@
     ItemHash,
     ProgramContent,
 )
-from aleph_message.models.execution.environment import HypervisorType
+from aleph_message.models.execution.environment import HypervisorType, GpuProperties
 
 from aleph.vm.conf import settings
 from aleph.vm.controllers.firecracker.executable import AlephFirecrackerExecutable
@@ -74,7 +74,7 @@ class VmExecution:
         AlephProgramResources | AlephInstanceResources | AlephQemuResources | AlephQemuConfidentialInstance | None
     ) = None
     vm: AlephFirecrackerExecutable | AlephQemuInstance | AlephQemuConfidentialInstance | None = None
-    gpus: List[HostGPU]
+    gpus: List[HostGPU] = []
 
     times: VmExecutionTimes
 
@@ -223,11 +223,13 @@ async def prepare(self) -> None:
 
     def prepare_gpus(self, available_gpus: List[GpuDevice]) -> None:
         gpus = []
-        for gpu in self.message.requirements.gpu:
-            for available_gpu in available_gpus:
-                if available_gpu.device_id == gpu.device_id:
-                    gpus.append(HostGPU(pci_host=available_gpu.pci_host))
-                    break
+        if self.message.requirements and self.message.requirements.gpu:
+            for gpu in self.message.requirements.gpu:
+                gpu = GpuProperties.parse_obj(gpu)
+                for available_gpu in available_gpus:
+                    if available_gpu.device_id == gpu.device_id:
+                        gpus.append(HostGPU(pci_host=available_gpu.pci_host))
+                        break
         self.gpus = gpus
 
     def uses_gpu(self, pci_host: str) -> bool:

src/aleph/vm/pool.py

@@ -299,9 +299,13 @@ def get_instance_executions(self) -> Iterable[VmExecution]:
     def get_available_gpus(self) -> List[GpuDevice]:
         available_gpus = []
         for gpu in self.gpus:
+            used = False
             for _, execution in self.executions.items():
-                if not execution.uses_gpu(gpu.pci_host):
-                    available_gpus.append(gpu)
+                if execution.uses_gpu(gpu.pci_host):
+                   used = True
+                   break
+            if not used:
+                available_gpus.append(gpu)
         return available_gpus
 
     def get_executions_by_sender(self, payment_type: PaymentType) -> dict[str, dict[str, list[VmExecution]]]: