refactor(gpu): centralize driver request validation

Signed-off-by: Evan Lezar <elezar@nvidia.com>

Author: elezar

crates/openshell-driver-kubernetes/src/driver.rs

@@ -205,6 +205,10 @@ impl KubernetesComputeDriver {
 
     pub async fn validate_sandbox_create(&self, sandbox: &Sandbox) -> Result<(), tonic::Status> {
         let gpu_requested = sandbox.spec.as_ref().is_some_and(|spec| spec.gpu);
+        self.validate_gpu_request(gpu_requested).await
+    }
+
+    async fn validate_gpu_request(&self, gpu_requested: bool) -> Result<(), tonic::Status> {
         if gpu_requested
             && !self.has_gpu_capacity().await.map_err(|err| {
                 tonic::Status::internal(format!("check GPU node capacity failed: {err}"))

crates/openshell-driver-vm/src/driver.rs

@@ -2572,15 +2572,7 @@ fn validate_vm_sandbox(sandbox: &Sandbox, gpu_enabled: bool) -> Result<(), Statu
         .as_ref()
         .ok_or_else(|| Status::invalid_argument("sandbox spec is required"))?;
 
-    if spec.gpu && !gpu_enabled {
-        return Err(Status::failed_precondition(
-            "GPU support is not enabled on this driver; start with --gpu",
-        ));
-    }
-
-    if !spec.gpu && !spec.gpu_device.is_empty() {
-        return Err(Status::invalid_argument("gpu_device requires gpu=true"));
-    }
+    validate_gpu_request(spec.gpu, &spec.gpu_device, gpu_enabled)?;
 
     if let Some(template) = spec.template.as_ref() {
         if !template.agent_socket_path.is_empty() {
@@ -2597,7 +2589,6 @@ fn validate_vm_sandbox(sandbox: &Sandbox, gpu_enabled: bool) -> Result<(), Statu
     Ok(())
 }
 
-#[allow(clippy::result_large_err)]
 fn validate_sandbox_id(sandbox_id: &str) -> Result<(), Status> {
     if sandbox_id.is_empty() {
         return Err(Status::invalid_argument("sandbox id is required"));
@@ -2623,6 +2614,20 @@ fn validate_sandbox_id(sandbox_id: &str) -> Result<(), Status> {
     Ok(())
 }
 
+#[allow(clippy::result_large_err)]
+fn validate_gpu_request(gpu: bool, gpu_device: &str, gpu_enabled: bool) -> Result<(), Status> {
+    if gpu && !gpu_enabled {
+        return Err(Status::failed_precondition(
+            "GPU support is not enabled on this VM driver; start the VM compute driver with GPU support enabled",
+        ));
+    }
+
+    if !gpu && !gpu_device.is_empty() {
+        return Err(Status::invalid_argument("gpu_device requires gpu=true"));
+    }
+    Ok(())
+}
+
 #[allow(clippy::result_large_err)]
 fn parse_registry_reference(image_ref: &str) -> Result<Reference, Status> {
     Reference::try_from(image_ref).map_err(|err| {
@@ -4468,7 +4473,10 @@ mod tests {
         let err = validate_vm_sandbox(&sandbox, false)
             .expect_err("gpu should be rejected when not enabled");
         assert_eq!(err.code(), Code::FailedPrecondition);
-        assert!(err.message().contains("GPU support is not enabled"));
+        assert!(
+            err.message()
+                .contains("GPU support is not enabled on this VM driver")
+        );
     }
 
     #[test]