feat: set machine labels directly on InfraMachineStatus

We do not need to set initial machine labels via META values: as the provider has access to Omni, it can write these labels directly to the resources in creates on Omni side.

Signed-off-by: Utku Ozdemir <[email protected]>

hack/test/integration.sh

@@ -183,6 +183,7 @@ docker run -d --network host \
   --api-power-mgmt-state-dir=/api-power-mgmt-state \
   --ipmi-pxe-boot-mode=bios \
   --min-reboot-interval=1m \
+  --machine-labels=a=b,c \
   --debug
 
 docker logs -f provider &

internal/provider/controllers/infra_machine_status.go

@@ -54,8 +54,8 @@ type APIPowerManager interface {
 type InfraMachineController = qtransform.QController[*infra.Machine, *infra.MachineStatus]
 
 // NewInfraMachineController initializes InfraMachineController.
-func NewInfraMachineController(agentService AgentService, apiPowerManager APIPowerManager, state state.State, pxeBootMode pxe.BootMode,
-	requeueInterval, minRebootInterval time.Duration,
+func NewInfraMachineController(agentService AgentService, apiPowerManager APIPowerManager, state state.State,
+	pxeBootMode pxe.BootMode, requeueInterval, minRebootInterval time.Duration, machineLabels map[string]string,
 ) *InfraMachineController {
 	helper := &infraMachineControllerHelper{
 		agentService:      agentService,
@@ -64,6 +64,7 @@ func NewInfraMachineController(agentService AgentService, apiPowerManager APIPow
 		pxeBootMode:       pxeBootMode,
 		requeueInterval:   requeueInterval,
 		minRebootInterval: minRebootInterval,
+		machineLabels:     machineLabels,
 	}
 
 	return qtransform.NewQController(
@@ -96,6 +97,7 @@ type infraMachineControllerHelper struct {
 	agentService      AgentService
 	apiPowerManager   APIPowerManager
 	state             state.State
+	machineLabels     map[string]string
 	pxeBootMode       pxe.BootMode
 	requeueInterval   time.Duration
 	minRebootInterval time.Duration
@@ -256,6 +258,11 @@ func (h *infraMachineControllerHelper) removeInternalStatus(ctx context.Context,
 func (h *infraMachineControllerHelper) populateInfraMachineStatus(status *baremetal.MachineStatus, infraMachineStatus *infra.MachineStatus) error {
 	infraMachineStatus.TypedSpec().Value.ReadyToUse = false
 
+	// set the labels
+	for k, v := range h.machineLabels {
+		infraMachineStatus.Metadata().Labels().Set(k, v)
+	}
+
 	// update power state
 	switch status.TypedSpec().Value.PowerState {
 	case specs.PowerState_POWER_STATE_ON:

internal/provider/imagefactory/client.go

@@ -7,13 +7,10 @@ package imagefactory
 import (
 	"context"
 	"fmt"
-	"strings"
 	"time"
 
 	"github.com/siderolabs/image-factory/pkg/client"
 	"github.com/siderolabs/image-factory/pkg/schematic"
-	"github.com/siderolabs/omni/client/pkg/meta"
-	"gopkg.in/yaml.v3"
 )
 
 var agentModeExtensions = []string{
@@ -36,16 +33,10 @@ type Client struct {
 	factoryClient         *client.Client
 	pxeBaseURL            string
 	agentModeTalosVersion string
-	machineLabelsMeta     string
 }
 
 // NewClient creates a new image factory client.
-func NewClient(baseURL, pxeBaseURL, agentModeTalosVersion string, machineLabels []string) (*Client, error) {
-	labelsMeta, err := parseLabels(machineLabels)
-	if err != nil {
-		return nil, err
-	}
-
+func NewClient(baseURL, pxeBaseURL, agentModeTalosVersion string) (*Client, error) {
 	factoryClient, err := client.New(baseURL)
 	if err != nil {
 		return nil, err
@@ -54,9 +45,7 @@ func NewClient(baseURL, pxeBaseURL, agentModeTalosVersion string, machineLabels
 	return &Client{
 		pxeBaseURL:            pxeBaseURL,
 		agentModeTalosVersion: agentModeTalosVersion,
-		machineLabelsMeta:     labelsMeta,
-
-		factoryClient: factoryClient,
+		factoryClient:         factoryClient,
 	}, nil
 }
 
@@ -66,13 +55,6 @@ func NewClient(baseURL, pxeBaseURL, agentModeTalosVersion string, machineLabels
 func (c *Client) SchematicIPXEURL(ctx context.Context, agentMode bool, talosVersion, arch string, extensions, extraKernelArgs []string) (string, error) {
 	var metaValues []schematic.MetaValue
 
-	if c.machineLabelsMeta != "" {
-		metaValues = append(metaValues, schematic.MetaValue{
-			Key:   meta.LabelsMeta,
-			Value: c.machineLabelsMeta,
-		})
-	}
-
 	if !agentMode && talosVersion == "" {
 		return "", fmt.Errorf("talosVersion is required when not booting into agent mode")
 	}
@@ -105,31 +87,3 @@ func (c *Client) SchematicIPXEURL(ctx context.Context, agentMode bool, talosVers
 
 	return ipxeURL, err
 }
-
-func parseLabels(machineLabels []string) (string, error) {
-	labels := map[string]string{}
-
-	for _, l := range machineLabels {
-		parts := strings.Split(l, "=")
-		if len(parts) > 2 {
-			return "", fmt.Errorf("malformed label %s", l)
-		}
-
-		value := ""
-
-		if len(parts) > 1 {
-			value = parts[1]
-		}
-
-		labels[parts[0]] = value
-	}
-
-	data, err := yaml.Marshal(meta.ImageLabels{
-		Labels: labels,
-	})
-	if err != nil {
-		return "", err
-	}
-
-	return string(data), nil
-}

internal/provider/provider.go

@@ -10,6 +10,7 @@ import (
 	_ "embed"
 	"fmt"
 	"os"
+	"strings"
 	"time"
 
 	"github.com/cosi-project/runtime/pkg/controller/runtime"
@@ -110,7 +111,7 @@ func (p *Provider) Run(ctx context.Context) error {
 		return fmt.Errorf("failed to create/update provider status: %w", err)
 	}
 
-	imageFactoryClient, err := imagefactory.NewClient(p.options.ImageFactoryBaseURL, p.options.ImageFactoryPXEBaseURL, p.options.AgentModeTalosVersion, p.options.MachineLabels)
+	imageFactoryClient, err := imagefactory.NewClient(p.options.ImageFactoryBaseURL, p.options.ImageFactoryPXEBaseURL, p.options.AgentModeTalosVersion)
 	if err != nil {
 		return fmt.Errorf("failed to create image factory client: %w", err)
 	}
@@ -131,6 +132,11 @@ func (p *Provider) Run(ctx context.Context) error {
 		return fmt.Errorf("failed to create config handler: %w", err)
 	}
 
+	parsedMachineLabels, err := p.parseLabels()
+	if err != nil {
+		return fmt.Errorf("failed to parse machine labels: %w", err)
+	}
+
 	srvr := server.New(ctx, p.options.APIListenAddress, p.options.APIPort, p.options.UseLocalBootAssets, configHandler, ipxeHandler, p.logger.With(zap.String("component", "server")))
 	agentService := agent.NewService(srvr, omniState, p.options.WipeWithZeroes, p.logger.With(zap.String("component", "agent_service"))) //nolint:contextcheck // false positive
 	machineStatusPoller := machinestatus.NewPoller(agentService, omniState, p.logger.With(zap.String("component", "machine_status_poller")))
@@ -141,7 +147,7 @@ func (p *Provider) Run(ctx context.Context) error {
 	// todo: enable if we re-enable reverse tunnel on Omni: https://github.com/siderolabs/omni/pull/746
 	// reverseTunnel := tunnel.New(omniState, omniAPIClient, p.logger.With(zap.String("component", "reverse_tunnel")))
 
-	infraMachineController := controllers.NewInfraMachineController(agentService, apiPowerManager, omniState, pxeBootMode, 1*time.Minute, p.options.MinRebootInterval)
+	infraMachineController := controllers.NewInfraMachineController(agentService, apiPowerManager, omniState, pxeBootMode, 1*time.Minute, p.options.MinRebootInterval, parsedMachineLabels)
 
 	if err = cosiRuntime.RegisterQController(infraMachineController); err != nil {
 		return fmt.Errorf("failed to register controller: %w", err)
@@ -275,3 +281,24 @@ func (p *Provider) clearState(ctx context.Context, st state.State) error {
 
 	return errs
 }
+
+func (p *Provider) parseLabels() (map[string]string, error) {
+	labels := make(map[string]string, len(p.options.MachineLabels))
+
+	for _, l := range p.options.MachineLabels {
+		parts := strings.Split(l, "=")
+		if len(parts) > 2 {
+			return nil, fmt.Errorf("malformed label %s", l)
+		}
+
+		value := ""
+
+		if len(parts) > 1 {
+			value = parts[1]
+		}
+
+		labels[parts[0]] = value
+	}
+
+	return labels, nil
+}