feat: Gate Flow power/firmware ops on host assignment state

rest-api/flow/docs/component-manager-architecture.md

@@ -101,7 +101,7 @@ type FirmwareStatusReader interface {
 
 type BringUpController interface {
     // CapabilityBringUpControl
-    BringUpControl(ctx, target) error
+    BringUpControl(ctx, target, info) error
 }
 
 type BringUpStatusReader interface {

rest-api/flow/internal/converter/protobuf/converter.go

@@ -1163,7 +1163,8 @@ func ScheduledOperationFrom(
 		}
 
 		return &operations.PowerControlTaskInfo{
-			Operation: operations.PowerOperationPowerOn,
+			Operation:               operations.PowerOperationPowerOn,
+			OverrideAssignmentCheck: r.PowerOn.GetOverrideAssignmentCheck(),
 		}, ts, r.PowerOn.GetQueueOptions(), r.PowerOn.GetRuleId(), nil
 
 	case *pb.ScheduledOperation_PowerOff:
@@ -1180,8 +1181,9 @@ func ScheduledOperationFrom(
 		}
 
 		return &operations.PowerControlTaskInfo{
-			Operation: powerOp,
-			Forced:    r.PowerOff.GetForced(),
+			Operation:               powerOp,
+			Forced:                  r.PowerOff.GetForced(),
+			OverrideAssignmentCheck: r.PowerOff.GetOverrideAssignmentCheck(),
 		}, ts, r.PowerOff.GetQueueOptions(), r.PowerOff.GetRuleId(), nil
 
 	case *pb.ScheduledOperation_PowerReset:
@@ -1198,8 +1200,9 @@ func ScheduledOperationFrom(
 		}
 
 		return &operations.PowerControlTaskInfo{
-			Operation: powerOp,
-			Forced:    r.PowerReset.GetForced(),
+			Operation:               powerOp,
+			Forced:                  r.PowerReset.GetForced(),
+			OverrideAssignmentCheck: r.PowerReset.GetOverrideAssignmentCheck(),
 		}, ts, r.PowerReset.GetQueueOptions(), r.PowerReset.GetRuleId(), nil
 
 	case *pb.ScheduledOperation_BringUp:
@@ -1210,7 +1213,9 @@ func ScheduledOperationFrom(
 			)
 		}
 
-		return &operations.BringUpTaskInfo{}, ts, nil, r.BringUp.GetRuleId(), nil
+		return &operations.BringUpTaskInfo{
+			OverrideAssignmentCheck: r.BringUp.GetOverrideAssignmentCheck(),
+		}, ts, nil, r.BringUp.GetRuleId(), nil
 
 	case *pb.ScheduledOperation_Ingest:
 		ts, err := TargetSpecFrom(r.Ingest.GetTargetSpec())
@@ -1224,9 +1229,10 @@ func ScheduledOperationFrom(
 
 	case *pb.ScheduledOperation_UpgradeFirmware:
 		info := &operations.FirmwareControlTaskInfo{
-			Operation:     operations.FirmwareOperationUpgrade,
-			TargetVersion: r.UpgradeFirmware.GetTargetVersion(),
-			SubTargets:    r.UpgradeFirmware.GetSubTargets(),
+			Operation:               operations.FirmwareOperationUpgrade,
+			TargetVersion:           r.UpgradeFirmware.GetTargetVersion(),
+			SubTargets:              r.UpgradeFirmware.GetSubTargets(),
+			OverrideAssignmentCheck: r.UpgradeFirmware.GetOverrideAssignmentCheck(),
 		}
 
 		if r.UpgradeFirmware.GetStartTime() != nil {

rest-api/flow/internal/nicoapi/grpc.go

@@ -276,6 +276,104 @@ func (c *grpcClient) FindMachinesByIds(ctx context.Context, machineIds []string)
 	return result, nil
 }
 
+// FindHostMachineIdsByRack queries Core for host machines (DPUs excluded) on
+// the given rack and returns their machine IDs.
+func (c *grpcClient) FindHostMachineIdsByRack(ctx context.Context, rackID string) ([]string, error) {
+	if rackID == "" {
+		return nil, errors.New("rack ID is required")
+	}
+
+	ctx, cancel := context.WithTimeout(ctx, c.grpcTimeout)
+	defer cancel()
+
+	cfg := &pb.MachineSearchConfig{
+		RackId: &pb.RackId{Id: rackID},
+		// include_dpus defaults to false; exclude_hosts defaults to false.
+		// We want hosts only because Assigned is a host-only state.
+	}
+
+	res, err := c.gclient.FindMachineIds(ctx, cfg)
+	if err != nil {
+		return nil, fmt.Errorf("FindMachineIds for rack %s: %w", rackID, err)
+	}
+
+	ids := make([]string, 0, len(res.GetMachineIds()))
+	for _, mid := range res.GetMachineIds() {
+		if id := mid.GetId(); id != "" {
+			ids = append(ids, id)
+		}
+	}
+	return ids, nil
+}
+
+// FindSwitchRackIDs returns the rack assignment of each given switch.
+func (c *grpcClient) FindSwitchRackIDs(ctx context.Context, switchIds []string) (map[string]string, error) {
+	if len(switchIds) == 0 {
+		return nil, nil
+	}
+
+	ctx, cancel := context.WithTimeout(ctx, c.grpcTimeout)
+	defer cancel()
+
+	req := &pb.SwitchesByIdsRequest{
+		SwitchIds: make([]*pb.SwitchId, 0, len(switchIds)),
+	}
+	for _, id := range switchIds {
+		req.SwitchIds = append(req.SwitchIds, &pb.SwitchId{Id: id})
+	}
+
+	resp, err := c.gclient.FindSwitchesByIds(ctx, req)
+	if err != nil {
+		return nil, fmt.Errorf("FindSwitchesByIds: %w", err)
+	}
+
+	result := make(map[string]string, len(resp.GetSwitches()))
+	for _, sw := range resp.GetSwitches() {
+		sid := sw.GetId().GetId()
+		if sid == "" {
+			continue
+		}
+		if rid := sw.GetRackId().GetId(); rid != "" {
+			result[sid] = rid
+		}
+	}
+	return result, nil
+}
+
+// FindPowerShelfRackIDs returns the rack assignment of each given power shelf.
+func (c *grpcClient) FindPowerShelfRackIDs(ctx context.Context, shelfIds []string) (map[string]string, error) {
+	if len(shelfIds) == 0 {
+		return nil, nil
+	}
+
+	ctx, cancel := context.WithTimeout(ctx, c.grpcTimeout)
+	defer cancel()
+
+	req := &pb.PowerShelvesByIdsRequest{
+		PowerShelfIds: make([]*pb.PowerShelfId, 0, len(shelfIds)),
+	}
+	for _, id := range shelfIds {
+		req.PowerShelfIds = append(req.PowerShelfIds, &pb.PowerShelfId{Id: id})
+	}
+
+	resp, err := c.gclient.FindPowerShelvesByIds(ctx, req)
+	if err != nil {
+		return nil, fmt.Errorf("FindPowerShelvesByIds: %w", err)
+	}
+
+	result := make(map[string]string, len(resp.GetPowerShelves()))
+	for _, ps := range resp.GetPowerShelves() {
+		pid := ps.GetId().GetId()
+		if pid == "" {
+			continue
+		}
+		if rid := ps.GetRackId().GetId(); rid != "" {
+			result[pid] = rid
+		}
+	}
+	return result, nil
+}
+
 // GetMachinePositionInfo returns position information for the given machine IDs
 func (c *grpcClient) GetMachinePositionInfo(ctx context.Context, machineIds []string) ([]MachinePosition, error) {
 	ctx, cancel := context.WithTimeout(ctx, c.grpcTimeout)
@@ -648,3 +746,15 @@ func (c *grpcClient) SetLeakingMachineIds(ids []string) {
 func (c *grpcClient) SetLeakingSwitchIds(ids []string) {
 	panic("Not a unit test")
 }
+
+func (c *grpcClient) SetSwitchRackID(switchID, rackID string) {
+	panic("Not a unit test")
+}
+
+func (c *grpcClient) SetPowerShelfRackID(shelfID, rackID string) {
+	panic("Not a unit test")
+}
+
+func (c *grpcClient) SetRackHostMachineIDs(rackID string, machineIDs []string) {
+	panic("Not a unit test")
+}

rest-api/flow/internal/nicoapi/mock.go

@@ -5,6 +5,7 @@ package nicoapi
 
 import (
 	"context"
+	"errors"
 	"time"
 
 	"github.com/NVIDIA/infra-controller-rest/flow/internal/common/utils"
@@ -21,15 +22,23 @@ type mockClient struct {
 	firmwareUpdateTimeWindowErr error // If set, SetFirmwareUpdateTimeWindow will return this error
 	adminPowerControlErr        error // If set, AdminPowerControl will return this error
 	desiredFirmwareVersions     []*pb.DesiredFirmwareVersionEntry
+	// Topology lookups exercised by the rack-assignment safety check. Tests
+	// populate these via Set...RackId / Set...HostMachineIds helpers.
+	switchRackIDs        map[string]string // switch ID → rack ID
+	powerShelfRackIDs    map[string]string // power shelf ID → rack ID
+	hostMachinesByRackID map[string][]string
 }
 
 // NewMockClient returns a "GRPC" client that returns mock values so it can be used in unit tests.
 func NewMockClient() Client {
 	return &mockClient{
-		machines:          map[string]MachineDetail{},
-		powerStates:       map[string]PowerState{},
-		machineInterfaces: map[string]MachineInterface{},
-		expectedSwitches:  map[string]ExpectedSwitchInfo{},
+		machines:             map[string]MachineDetail{},
+		powerStates:          map[string]PowerState{},
+		machineInterfaces:    map[string]MachineInterface{},
+		expectedSwitches:     map[string]ExpectedSwitchInfo{},
+		switchRackIDs:        map[string]string{},
+		powerShelfRackIDs:    map[string]string{},
+		hostMachinesByRackID: map[string][]string{},
 	}
 }
 
@@ -121,6 +130,62 @@ func (c *mockClient) FindMachinesByIds(ctx context.Context, machineIds []string)
 	return result, nil
 }
 
+func (c *mockClient) FindHostMachineIdsByRack(_ context.Context, rackID string) ([]string, error) {
+	if rackID == "" {
+		return nil, errors.New("rack ID is required")
+	}
+	ids := c.hostMachinesByRackID[rackID]
+	if len(ids) == 0 {
+		return nil, nil
+	}
+	out := make([]string, len(ids))
+	copy(out, ids)
+	return out, nil
+}
+
+func (c *mockClient) FindSwitchRackIDs(_ context.Context, switchIds []string) (map[string]string, error) {
+	if len(switchIds) == 0 {
+		return nil, nil
+	}
+	out := make(map[string]string, len(switchIds))
+	for _, id := range switchIds {
+		if rid, ok := c.switchRackIDs[id]; ok && rid != "" {
+			out[id] = rid
+		}
+	}
+	return out, nil
+}
+
+func (c *mockClient) FindPowerShelfRackIDs(_ context.Context, shelfIds []string) (map[string]string, error) {
+	if len(shelfIds) == 0 {
+		return nil, nil
+	}
+	out := make(map[string]string, len(shelfIds))
+	for _, id := range shelfIds {
+		if rid, ok := c.powerShelfRackIDs[id]; ok && rid != "" {
+			out[id] = rid
+		}
+	}
+	return out, nil
+}
+
+// SetSwitchRackID records the rack assignment for a switch (mock only).
+func (c *mockClient) SetSwitchRackID(switchID, rackID string) {
+	c.switchRackIDs[switchID] = rackID
+}
+
+// SetPowerShelfRackID records the rack assignment for a power shelf (mock only).
+func (c *mockClient) SetPowerShelfRackID(shelfID, rackID string) {
+	c.powerShelfRackIDs[shelfID] = rackID
+}
+
+// SetRackHostMachineIDs records which host machines a rack contains (mock only).
+func (c *mockClient) SetRackHostMachineIDs(rackID string, machineIDs []string) {
+	out := make([]string, len(machineIDs))
+	copy(out, machineIDs)
+	c.hostMachinesByRackID[rackID] = out
+}
+
 func (c *mockClient) GetMachinePositionInfo(ctx context.Context, machineIds []string) ([]MachinePosition, error) {
 	// Mock implementation returns empty for now
 	return nil, nil

rest-api/flow/internal/nicoapi/mod.go

@@ -35,6 +35,21 @@ type Client interface {
 	// FindMachinesByIds returns detailed machine information for the given machine IDs
 	FindMachinesByIds(ctx context.Context, machineIds []string) ([]MachineDetail, error)
 
+	// FindHostMachineIdsByRack returns the IDs of host (non-DPU) machines that
+	// belong to the given rack. Empty rackID is rejected. Returns nil when the
+	// rack has no host machines.
+	FindHostMachineIdsByRack(ctx context.Context, rackID string) ([]string, error)
+
+	// FindSwitchRackIDs returns the mapping from switch ID to rack ID for the
+	// given switches. A switch without a rack assignment is omitted from the
+	// result rather than reported as an empty string.
+	FindSwitchRackIDs(ctx context.Context, switchIds []string) (map[string]string, error)
+
+	// FindPowerShelfRackIDs returns the mapping from power-shelf ID to rack ID
+	// for the given shelves. A shelf without a rack assignment is omitted from
+	// the result rather than reported as an empty string.
+	FindPowerShelfRackIDs(ctx context.Context, shelfIds []string) (map[string]string, error)
+
 	// GetMachinePositionInfo returns position information for the given machine IDs
 	GetMachinePositionInfo(ctx context.Context, machineIds []string) ([]MachinePosition, error)
 
@@ -114,4 +129,7 @@ type Client interface {
 	AddExpectedSwitchInfo(info ExpectedSwitchInfo)
 	SetLeakingMachineIds(ids []string)
 	SetLeakingSwitchIds([]string)
+	SetSwitchRackID(switchID, rackID string)
+	SetPowerShelfRackID(shelfID, rackID string)
+	SetRackHostMachineIDs(rackID string, machineIDs []string)
 }

rest-api/flow/internal/service/server_impl.go

@@ -650,7 +650,8 @@ func (rs *FlowServerImpl) PowerOnRack(
 		req.GetQueueOptions(),
 		req.GetRuleId(),
 		&operations.PowerControlTaskInfo{
-			Operation: operations.PowerOperationPowerOn,
+			Operation:               operations.PowerOperationPowerOn,
+			OverrideAssignmentCheck: req.GetOverrideAssignmentCheck(),
 		},
 	)
 }
@@ -670,8 +671,9 @@ func (rs *FlowServerImpl) PowerOffRack(
 		req.GetQueueOptions(),
 		req.GetRuleId(),
 		&operations.PowerControlTaskInfo{
-			Operation: op,
-			Forced:    req.GetForced(),
+			Operation:               op,
+			Forced:                  req.GetForced(),
+			OverrideAssignmentCheck: req.GetOverrideAssignmentCheck(),
 		},
 	)
 }
@@ -691,8 +693,9 @@ func (rs *FlowServerImpl) PowerResetRack(
 		req.GetQueueOptions(),
 		req.GetRuleId(),
 		&operations.PowerControlTaskInfo{
-			Operation: op,
-			Forced:    req.GetForced(),
+			Operation:               op,
+			Forced:                  req.GetForced(),
+			OverrideAssignmentCheck: req.GetOverrideAssignmentCheck(),
 		},
 	)
 }
@@ -715,7 +718,8 @@ func (rs *FlowServerImpl) BringUpRack(
 	}
 
 	info := &operations.BringUpTaskInfo{
-		RuleID: protobuf.UUIDStringFrom(req.GetRuleId()),
+		RuleID:                  protobuf.UUIDStringFrom(req.GetRuleId()),
+		OverrideAssignmentCheck: req.GetOverrideAssignmentCheck(),
 	}
 	opReq, err := rs.convertTargetSpecToOperationRequest(
 		targetSpec, req.GetDescription(), info,
@@ -1233,10 +1237,11 @@ func (rs *FlowServerImpl) UpgradeFirmware(
 
 	// Build FirmwareControlTaskInfo
 	info := &operations.FirmwareControlTaskInfo{
-		Operation:     operations.FirmwareOperationUpgrade,
-		TargetVersion: req.GetTargetVersion(),
-		RuleID:        protobuf.UUIDStringFrom(req.GetRuleId()),
-		SubTargets:    req.GetSubTargets(),
+		Operation:               operations.FirmwareOperationUpgrade,
+		TargetVersion:           req.GetTargetVersion(),
+		RuleID:                  protobuf.UUIDStringFrom(req.GetRuleId()),
+		SubTargets:              req.GetSubTargets(),
+		OverrideAssignmentCheck: req.GetOverrideAssignmentCheck(),
 	}
 
 	// Parse optional time parameters for scheduled upgrade

rest-api/flow/internal/task/componentmanager/builtin/builtin_test.go

@@ -445,7 +445,7 @@ func TestServiceCatalog(t *testing.T) {
 			name:              "nvswitch nvswitchmanager",
 			componentType:     devicetypes.ComponentTypeNVSwitch,
 			implementation:    nvswitchnsm.ImplementationName,
-			requiredProviders: []string{nsmprovider.ProviderName},
+			requiredProviders: []string{nsmprovider.ProviderName, nicoprovider.ProviderName},
 			capabilities: capability.CapabilitySet{
 				capability.CapabilityFirmwareControl,
 				capability.CapabilityFirmwareStatus,