feat: add configuration for EPHEMERAL volume

Fixes #9261

Signed-off-by: Andrey Smirnov <andrey.smirnov@siderolabs.com>

.github/workflows/ci.yaml

@@ -1,6 +1,6 @@
 # THIS FILE WAS AUTOMATICALLY GENERATED, PLEASE DO NOT EDIT.
 #
-# Generated on 2024-09-03T14:18:03Z by kres b5ca957.
+# Generated on 2024-09-05T10:32:02Z by kres b5ca957.
 
 name: default
 concurrency:
@@ -2556,6 +2556,10 @@ jobs:
       - name: e2e-qemu
         env:
           IMAGE_REGISTRY: registry.dev.siderolabs.io
+          QEMU_EXTRA_DISKS: "2"
+          QEMU_EXTRA_DISKS_DRIVERS: ide,nvme
+          QEMU_EXTRA_DISKS_SIZE: "10240"
+          WITH_CONFIG_PATCH_WORKER: '@hack/test/patches/ephemeral-nvme.yaml'
         run: |
           sudo -E make e2e-qemu
       - name: save artifacts
@@ -2855,6 +2859,10 @@ jobs:
       - name: e2e-qemu
         env:
           IMAGE_REGISTRY: registry.dev.siderolabs.io
+          QEMU_EXTRA_DISKS: "2"
+          QEMU_EXTRA_DISKS_DRIVERS: ide,nvme
+          QEMU_EXTRA_DISKS_SIZE: "10240"
+          WITH_CONFIG_PATCH_WORKER: '@hack/test/patches/ephemeral-nvme.yaml'
           WITH_DISK_ENCRYPTION: "true"
           WITH_KUBESPAN: "true"
           WITH_VIRTUAL_IP: "true"

.github/workflows/integration-qemu-cron.yaml

@@ -1,6 +1,6 @@
 # THIS FILE WAS AUTOMATICALLY GENERATED, PLEASE DO NOT EDIT.
 #
-# Generated on 2024-05-27T16:20:10Z by kres bcb280a.
+# Generated on 2024-09-04T16:05:13Z by kres b5ca957.
 
 name: integration-qemu-cron
 concurrency:
@@ -77,6 +77,10 @@ jobs:
       - name: e2e-qemu
         env:
           IMAGE_REGISTRY: registry.dev.siderolabs.io
+          QEMU_EXTRA_DISKS: "2"
+          QEMU_EXTRA_DISKS_DRIVERS: ide,nvme
+          QEMU_EXTRA_DISKS_SIZE: "10240"
+          WITH_CONFIG_PATCH_WORKER: '@hack/test/patches/ephemeral-nvme.yaml'
         run: |
           sudo -E make e2e-qemu
       - name: save artifacts

.github/workflows/integration-qemu-encrypted-vip-cron.yaml

@@ -1,6 +1,6 @@
 # THIS FILE WAS AUTOMATICALLY GENERATED, PLEASE DO NOT EDIT.
 #
-# Generated on 2024-05-27T16:20:10Z by kres bcb280a.
+# Generated on 2024-09-05T10:32:02Z by kres b5ca957.
 
 name: integration-qemu-encrypted-vip-cron
 concurrency:
@@ -77,6 +77,10 @@ jobs:
       - name: e2e-qemu
         env:
           IMAGE_REGISTRY: registry.dev.siderolabs.io
+          QEMU_EXTRA_DISKS: "2"
+          QEMU_EXTRA_DISKS_DRIVERS: ide,nvme
+          QEMU_EXTRA_DISKS_SIZE: "10240"
+          WITH_CONFIG_PATCH_WORKER: '@hack/test/patches/ephemeral-nvme.yaml'
           WITH_DISK_ENCRYPTION: "true"
           WITH_KUBESPAN: "true"
           WITH_VIRTUAL_IP: "true"

.kres.yaml

@@ -325,6 +325,10 @@ spec:
           withSudo: true
           environment:
             IMAGE_REGISTRY: registry.dev.siderolabs.io
+            QEMU_EXTRA_DISKS: "2"
+            QEMU_EXTRA_DISKS_SIZE: "10240"
+            QEMU_EXTRA_DISKS_DRIVERS: "ide,nvme"
+            WITH_CONFIG_PATCH_WORKER: "@hack/test/patches/ephemeral-nvme.yaml"
         - name: save-talos-logs
           conditions:
             - always
@@ -1098,6 +1102,10 @@ spec:
             WITH_VIRTUAL_IP: true
             WITH_KUBESPAN: true
             IMAGE_REGISTRY: registry.dev.siderolabs.io
+            QEMU_EXTRA_DISKS: "2"
+            QEMU_EXTRA_DISKS_SIZE: "10240"
+            QEMU_EXTRA_DISKS_DRIVERS: "ide,nvme"
+            WITH_CONFIG_PATCH_WORKER: "@hack/test/patches/ephemeral-nvme.yaml"
         - name: save-talos-logs
           conditions:
             - always

api/resource/definitions/block/block.proto

@@ -164,5 +164,6 @@ message VolumeStatusSpec {
   talos.resource.definitions.enums.BlockFilesystemType filesystem = 10;
   string mount_location = 11;
   talos.resource.definitions.enums.BlockEncryptionProviderType encryption_provider = 12;
+  string pretty_size = 13;
 }
 

cmd/talosctl/cmd/docs.go

@@ -19,6 +19,7 @@ import (
 	"gopkg.in/yaml.v3"
 
 	"github.com/siderolabs/talos/pkg/machinery/config/encoder"
+	"github.com/siderolabs/talos/pkg/machinery/config/types/block"
 	"github.com/siderolabs/talos/pkg/machinery/config/types/network"
 	"github.com/siderolabs/talos/pkg/machinery/config/types/runtime"
 	"github.com/siderolabs/talos/pkg/machinery/config/types/runtime/extensions"
@@ -130,6 +131,10 @@ var docsCmd = &cobra.Command{
 					name:    "security",
 					fileDoc: security.GetFileDoc(),
 				},
+				{
+					name:    "block",
+					fileDoc: block.GetFileDoc(),
+				},
 			} {
 				path := filepath.Join(dir, pkg.name)
 

hack/release.toml

@@ -223,6 +223,12 @@ This can be also explicitly enabled by setting `talos.halt_if_installed=1` in ke
         description = """\
 Talos Linux installer now never wipes the system disk on upgrades, which means that the flag
 `--preserve` is always set for `talosctl upgrade`.
+"""
+
+    [notes.disk-management]
+        title = "Disk Management"
+        description = """\
+Talos Linux now supports [configuration](https://www.talos.dev/v1.8/talos-guides/configuration/disk-management/#machine-configuration) for the `EPHEMERAL` volume.
 """
 
 [make_deps]

hack/test/e2e-qemu.sh

@@ -196,6 +196,7 @@ function create_cluster {
     --disk=15360 \
     --extra-disks="${QEMU_EXTRA_DISKS:-0}" \
     --extra-disks-size="${QEMU_EXTRA_DISKS_SIZE:-5120}" \
+    --extra-disks-drivers="${QEMU_EXTRA_DISKS_DRIVERS:-}" \
     --mtu=1430 \
     --memory=2048 \
     --memory-workers="${QEMU_MEMORY_WORKERS:-2048}" \

hack/test/patches/ephemeral-nvme.yaml

@@ -0,0 +1,8 @@
+apiVersion: v1alpha1
+kind: VolumeConfig
+name: EPHEMERAL
+provisioning:
+  diskSelector:
+    match: disk.transport == 'nvme'
+  minSize: 4GB
+  maxSize: 8GB

internal/app/machined/pkg/controllers/block/discovery.go

@@ -14,7 +14,6 @@ import (
 	"github.com/cosi-project/runtime/pkg/controller"
 	"github.com/cosi-project/runtime/pkg/safe"
 	"github.com/cosi-project/runtime/pkg/state"
-	"github.com/dustin/go-humanize"
 	"github.com/siderolabs/gen/maps"
 	"github.com/siderolabs/go-blockdevice/v2/blkid"
 	"github.com/siderolabs/go-blockdevice/v2/partitioning"
@@ -208,8 +207,7 @@ func (ctrl *DiscoveryController) rescan(ctx context.Context, r controller.Runtim
 			dv.TypedSpec().Parent = device.TypedSpec().Parent
 			dv.TypedSpec().ParentDevPath = filepath.Join("/dev", device.TypedSpec().Parent)
 
-			dv.TypedSpec().Size = info.Size
-			dv.TypedSpec().PrettySize = humanize.Bytes(info.Size)
+			dv.TypedSpec().SetSize(info.Size)
 			dv.TypedSpec().SectorSize = info.SectorSize
 			dv.TypedSpec().IOSize = info.IOSize
 
@@ -232,14 +230,12 @@ func (ctrl *DiscoveryController) rescan(ctx context.Context, r controller.Runtim
 				dv.TypedSpec().Parent = id
 				dv.TypedSpec().ParentDevPath = filepath.Join("/dev", id)
 
-				dv.TypedSpec().Size = nested.ProbedSize
-
-				if dv.TypedSpec().Size == 0 {
-					dv.TypedSpec().Size = nested.PartitionSize
+				if nested.ProbedSize != 0 {
+					dv.TypedSpec().SetSize(nested.ProbedSize)
+				} else {
+					dv.TypedSpec().SetSize(nested.PartitionSize)
 				}
 
-				dv.TypedSpec().PrettySize = humanize.Bytes(dv.TypedSpec().Size)
-
 				dv.TypedSpec().SectorSize = info.SectorSize
 				dv.TypedSpec().IOSize = info.IOSize
 

internal/app/machined/pkg/controllers/block/disks.go

@@ -11,7 +11,6 @@ import (
 
 	"github.com/cosi-project/runtime/pkg/controller"
 	"github.com/cosi-project/runtime/pkg/safe"
-	"github.com/dustin/go-humanize"
 	blkdev "github.com/siderolabs/go-blockdevice/v2/block"
 	"go.uber.org/zap"
 
@@ -157,9 +156,9 @@ func (ctrl *DisksController) analyzeBlockDevice(ctx context.Context, r controlle
 	touchedDisks[device.Metadata().ID()] = struct{}{}
 
 	return safe.WriterModify(ctx, r, block.NewDisk(block.NamespaceName, device.Metadata().ID()), func(d *block.Disk) error {
+		d.TypedSpec().SetSize(size)
+
 		d.TypedSpec().DevPath = filepath.Join("/dev", device.Metadata().ID())
-		d.TypedSpec().Size = size
-		d.TypedSpec().PrettySize = humanize.Bytes(size)
 		d.TypedSpec().IOSize = ioSize
 		d.TypedSpec().SectorSize = sectorSize
 		d.TypedSpec().Readonly = readOnly

internal/app/machined/pkg/controllers/block/internal/volumes/grow.go

@@ -84,7 +84,7 @@ func Grow(ctx context.Context, logger *zap.Logger, volumeContext ManagerContext)
 	}
 
 	volumeContext.Status.Phase = block.VolumePhaseProvisioned
-	volumeContext.Status.Size += availableGrowth
+	volumeContext.Status.SetSize(volumeContext.Status.Size + availableGrowth)
 
 	return nil
 }

internal/app/machined/pkg/controllers/block/internal/volumes/locate.go

@@ -55,7 +55,7 @@ func LocateAndProvision(ctx context.Context, logger *zap.Logger, volumeContext M
 
 			volumeContext.Status.UUID = dv.Uuid
 			volumeContext.Status.PartitionUUID = dv.PartitionUuid
-			volumeContext.Status.Size = dv.Size
+			volumeContext.Status.SetSize(dv.Size)
 
 			return nil
 		}
@@ -138,7 +138,7 @@ func LocateAndProvision(ctx context.Context, logger *zap.Logger, volumeContext M
 		volumeContext.Status.Phase = block.VolumePhaseProvisioned
 		volumeContext.Status.Location = pickedDisk
 		volumeContext.Status.ParentLocation = ""
-		volumeContext.Status.Size = diskCheckResult.DiskSize
+		volumeContext.Status.SetSize(diskCheckResult.DiskSize)
 	case block.VolumeTypePartition:
 		// we need to create a partition on the disk
 		partitionCreateResult, err := CreatePartition(ctx, logger, pickedDisk, volumeContext.Cfg, diskCheckResult.HasGPT)
@@ -151,7 +151,7 @@ func LocateAndProvision(ctx context.Context, logger *zap.Logger, volumeContext M
 		volumeContext.Status.PartitionIndex = partitionCreateResult.PartitionIdx
 		volumeContext.Status.ParentLocation = pickedDisk
 		volumeContext.Status.PartitionUUID = partitionCreateResult.Partition.PartGUID.String()
-		volumeContext.Status.Size = partitionCreateResult.Size
+		volumeContext.Status.SetSize(partitionCreateResult.Size)
 	default:
 		panic("unexpected volume type")
 	}

internal/app/machined/pkg/controllers/block/volume_config.go

@@ -204,16 +204,19 @@ func (ctrl *VolumeConfigController) Run(ctx context.Context, r controller.Runtim
 
 func (ctrl *VolumeConfigController) manageEphemeral(config cfg.Config) func(vc *block.VolumeConfig) error {
 	return func(vc *block.VolumeConfig) error {
+		extraVolumeConfig := config.Volumes().ByName(constants.EphemeralPartitionLabel)
+
 		vc.TypedSpec().Type = block.VolumeTypePartition
 
 		vc.TypedSpec().Provisioning = block.ProvisioningSpec{
 			Wave: block.WaveSystemDisk,
 			DiskSelector: block.DiskSelector{
-				Match: systemDiskMatch(),
+				Match: extraVolumeConfig.Provisioning().DiskSelector().ValueOr(systemDiskMatch()),
 			},
 			PartitionSpec: block.PartitionSpec{
-				MinSize:  partition.EphemeralMinSize,
-				Grow:     true,
+				MinSize:  extraVolumeConfig.Provisioning().MinSize().ValueOr(partition.EphemeralMinSize),
+				MaxSize:  extraVolumeConfig.Provisioning().MaxSize().ValueOr(0),
+				Grow:     extraVolumeConfig.Provisioning().Grow().ValueOr(true),
 				Label:    constants.EphemeralPartitionLabel,
 				TypeUUID: partition.LinuxFilesystemData,
 			},

internal/app/machined/pkg/controllers/block/volume_config_test.go

@@ -10,12 +10,17 @@ import (
 	"testing"
 	"time"
 
+	"github.com/siderolabs/go-pointer"
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/suite"
 
 	blockctrls "github.com/siderolabs/talos/internal/app/machined/pkg/controllers/block"
 	"github.com/siderolabs/talos/internal/app/machined/pkg/controllers/ctest"
+	"github.com/siderolabs/talos/internal/pkg/partition"
+	"github.com/siderolabs/talos/pkg/machinery/cel"
+	"github.com/siderolabs/talos/pkg/machinery/cel/celenv"
 	"github.com/siderolabs/talos/pkg/machinery/config/container"
+	blockcfg "github.com/siderolabs/talos/pkg/machinery/config/types/block"
 	"github.com/siderolabs/talos/pkg/machinery/config/types/v1alpha1"
 	"github.com/siderolabs/talos/pkg/machinery/constants"
 	"github.com/siderolabs/talos/pkg/machinery/meta"
@@ -105,6 +110,14 @@ func (suite *VolumeConfigSuite) TestReconcileDefaults() {
 		asrt.NoError(err)
 		asrt.Equal(`volume.partition_label == "EPHEMERAL"`, string(locator))
 
+		locator, err = r.TypedSpec().Provisioning.DiskSelector.Match.MarshalText()
+		asrt.NoError(err)
+		asrt.Equal(`system_disk`, string(locator))
+
+		asrt.True(r.TypedSpec().Provisioning.PartitionSpec.Grow)
+		asrt.EqualValues(0, r.TypedSpec().Provisioning.PartitionSpec.MaxSize)
+		asrt.EqualValues(partition.EphemeralMinSize, r.TypedSpec().Provisioning.PartitionSpec.MinSize)
+
 		asrt.Equal(constants.EphemeralMountPoint, r.TypedSpec().Mount.TargetPath)
 	})
 }
@@ -198,3 +211,52 @@ func (suite *VolumeConfigSuite) TestReconcileEncryptedSTATE() {
 		asrt.Empty(r.TypedSpec().Encryption)
 	})
 }
+
+func (suite *VolumeConfigSuite) TestReconcileExtraEPHEMERALConfig() {
+	ctest.AssertNoResource[*block.VolumeConfig](suite, constants.EphemeralPartitionLabel)
+
+	u, err := url.Parse("https://foo:6443")
+	suite.Require().NoError(err)
+
+	ctr, err := container.New(
+		&v1alpha1.Config{
+			ConfigVersion: "v1alpha1",
+			MachineConfig: &v1alpha1.MachineConfig{},
+			ClusterConfig: &v1alpha1.ClusterConfig{
+				ControlPlane: &v1alpha1.ControlPlaneConfig{
+					Endpoint: &v1alpha1.Endpoint{
+						URL: u,
+					},
+				},
+			},
+		},
+		&blockcfg.VolumeConfigV1Alpha1{
+			MetaName: constants.EphemeralPartitionLabel,
+			ProvisioningSpec: blockcfg.ProvisioningSpec{
+				DiskSelectorSpec: blockcfg.DiskSelector{
+					Match: cel.MustExpression(cel.ParseBooleanExpression(`disk.transport == "nvme"`, celenv.DiskLocator())),
+				},
+				ProvisioningGrow:    pointer.To(false),
+				ProvisioningMaxSize: blockcfg.MustByteSize("2.5TiB"),
+			},
+		},
+	)
+	suite.Require().NoError(err)
+
+	cfg := config.NewMachineConfig(ctr)
+	suite.Require().NoError(suite.State().Create(suite.Ctx(), cfg))
+
+	// now the volume config should be created
+	ctest.AssertResource(suite, constants.EphemeralPartitionLabel, func(r *block.VolumeConfig, asrt *assert.Assertions) {
+		asrt.NotEmpty(r.TypedSpec().Provisioning)
+		asrt.Empty(r.TypedSpec().Encryption)
+
+		locator, err := r.TypedSpec().Provisioning.DiskSelector.Match.MarshalText()
+		asrt.NoError(err)
+		asrt.Equal(`disk.transport == "nvme"`, string(locator))
+
+		asrt.False(r.TypedSpec().Provisioning.PartitionSpec.Grow)
+		asrt.EqualValues(2.5*1024*1024*1024*1024, r.TypedSpec().Provisioning.PartitionSpec.MaxSize)
+		asrt.EqualValues(partition.EphemeralMinSize, r.TypedSpec().Provisioning.PartitionSpec.MinSize)
+	})
+}