fix: talosctl support and race tests

.github/workflows/ci.yaml

@@ -1,6 +1,6 @@
 # THIS FILE WAS AUTOMATICALLY GENERATED, PLEASE DO NOT EDIT.
 #
-# Generated on 2024-12-24T15:00:58Z by kres fcff05e.
+# Generated on 2024-12-25T15:13:54Z by kres fcff05e.
 
 name: default
 concurrency:
@@ -3367,6 +3367,8 @@ jobs:
           QEMU_EXTRA_DISKS: "3"
           QEMU_EXTRA_DISKS_DRIVERS: ide,nvme
           QEMU_EXTRA_DISKS_SIZE: "10240"
+          QEMU_MEMORY_CONTROLPLANES: "4096"
+          QEMU_MEMORY_WORKERS: "4096"
           TAG_SUFFIX: -race
           WITH_CONFIG_PATCH_WORKER: '@hack/test/patches/ephemeral-nvme.yaml:@hack/test/patches/dm-raid-module.yaml'
         run: |

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

@@ -1,6 +1,6 @@
 # THIS FILE WAS AUTOMATICALLY GENERATED, PLEASE DO NOT EDIT.
 #
-# Generated on 2024-11-28T13:53:18Z by kres 232fe63.
+# Generated on 2024-12-25T15:13:54Z by kres fcff05e.
 
 name: integration-qemu-race-cron
 concurrency:
@@ -94,6 +94,8 @@ jobs:
           QEMU_EXTRA_DISKS: "3"
           QEMU_EXTRA_DISKS_DRIVERS: ide,nvme
           QEMU_EXTRA_DISKS_SIZE: "10240"
+          QEMU_MEMORY_CONTROLPLANES: "4096"
+          QEMU_MEMORY_WORKERS: "4096"
           TAG_SUFFIX: -race
           WITH_CONFIG_PATCH_WORKER: '@hack/test/patches/ephemeral-nvme.yaml:@hack/test/patches/dm-raid-module.yaml'
         run: |

.kres.yaml

@@ -1283,6 +1283,8 @@ spec:
             QEMU_EXTRA_DISKS_SIZE: "10240"
             QEMU_EXTRA_DISKS_DRIVERS: "ide,nvme"
             WITH_CONFIG_PATCH_WORKER: "@hack/test/patches/ephemeral-nvme.yaml:@hack/test/patches/dm-raid-module.yaml"
+            QEMU_MEMORY_CONTROLPLANES: 4096 # race-enabled Talos consumes lots of RAM
+            QEMU_MEMORY_WORKERS: 4096
             TAG_SUFFIX: -race
             IMAGE_REGISTRY: registry.dev.siderolabs.io
         - name: save-talos-logs

cmd/talosctl/cmd/talos/support.go

@@ -12,6 +12,7 @@ import (
 	"io"
 	"os"
 	"strings"
+	"sync"
 	"text/tabwriter"
 
 	"github.com/cosi-project/runtime/pkg/resource"
@@ -111,7 +112,7 @@ var supportCmd = &cobra.Command{
 }
 
 func collectData(dest *os.File, progress chan bundle.Progress) error {
-	return WithClient(func(ctx context.Context, c *client.Client) error {
+	return WithClientNoNodes(func(ctx context.Context, c *client.Client) error {
 		clientset, err := getKubernetesClient(ctx, c)
 		if err != nil {
 			fmt.Fprintf(os.Stderr, "Failed to create kubernetes client %s\n", err)
@@ -142,11 +143,7 @@ func collectData(dest *os.File, progress chan bundle.Progress) error {
 }
 
 func getKubernetesClient(ctx context.Context, c *client.Client) (*k8s.Clientset, error) {
-	if len(GlobalArgs.Endpoints) == 0 {
-		fmt.Fprintln(os.Stderr, "No endpoints set for the cluster, the command might not be able to get kubeconfig")
-	}
-
-	kubeconfig, err := c.Kubeconfig(client.WithNodes(ctx, GlobalArgs.Endpoints...))
+	kubeconfig, err := c.Kubeconfig(ctx)
 	if err != nil {
 		return nil, err
 	}
@@ -284,6 +281,7 @@ func showProgress(progress <-chan bundle.Progress, errors *supportBundleErrors)
 	uiprogress.Start()
 
 	type nodeProgress struct {
+		mu    sync.Mutex
 		state string
 		bar   *uiprogress.Bar
 	}
@@ -298,29 +296,39 @@ func showProgress(progress <-chan bundle.Progress, errors *supportBundleErrors)
 			ok bool
 		)
 
-		if np, ok = nodes[p.Source]; !ok {
+		src := p.Source
+
+		if _, ok = nodes[p.Source]; !ok {
 			bar := uiprogress.AddBar(p.Total)
 			bar = bar.AppendCompleted().PrependElapsed()
 
-			src := p.Source
-
 			np = &nodeProgress{
 				state: "initializing...",
 				bar:   bar,
 			}
 
-			bar.AppendFunc(func(b *uiprogress.Bar) string {
-				return fmt.Sprintf("%s: %s", src, np.state)
-			})
+			bar.AppendFunc(
+				func(src string, np *nodeProgress) func(b *uiprogress.Bar) string {
+					return func(b *uiprogress.Bar) string {
+						np.mu.Lock()
+						defer np.mu.Unlock()
+
+						return fmt.Sprintf("%s: %s", src, np.state)
+					}
+				}(src, np),
+			)
 
 			bar.Width = 20
 
 			nodes[src] = np
 		} else {
-			np = nodes[p.Source]
+			np = nodes[src]
 		}
 
+		np.mu.Lock()
 		np.state = p.State
+		np.mu.Unlock()
+
 		np.bar.Incr()
 	}
 

go.mod

@@ -157,7 +157,7 @@ require (
 	github.com/siderolabs/go-retry v0.3.3
 	github.com/siderolabs/go-smbios v0.3.3
 	github.com/siderolabs/go-tail v0.1.1
-	github.com/siderolabs/go-talos-support v0.1.1
+	github.com/siderolabs/go-talos-support v0.1.2
 	github.com/siderolabs/grpc-proxy v0.5.1
 	github.com/siderolabs/kms-client v0.1.0
 	github.com/siderolabs/net v0.4.0

go.sum

@@ -675,8 +675,8 @@ github.com/siderolabs/go-smbios v0.3.3 h1:rM3UKHQ8in1mqNRkpV75Ls3Wnk6rAhQJVYKUsK
 github.com/siderolabs/go-smbios v0.3.3/go.mod h1:kScnr0XSyzLfkRo/ChjITgI0rPRQnIi6PdgbxVCwA9U=
 github.com/siderolabs/go-tail v0.1.1 h1:3XeJgd97OHyFAIE7nQEMcRhOfnv7DvXbu0BRKbtT6u8=
 github.com/siderolabs/go-tail v0.1.1/go.mod h1:IihAL39acadXHfb5fEAOKK2DaDFIrG2+VD3b2H/ziZ0=
-github.com/siderolabs/go-talos-support v0.1.1 h1:g51J0WQssQAycU/0cDliC2l4uX2H02yUs2+fa5pCvHg=
-github.com/siderolabs/go-talos-support v0.1.1/go.mod h1:o4woiYS+2J3djCQgyHZRVZQm8XpazQr+XPcTXAZvamo=
+github.com/siderolabs/go-talos-support v0.1.2 h1:xKFwT8emzxpmamIe3W35QlmadC54OaPNO9/Y+fL7WwM=
+github.com/siderolabs/go-talos-support v0.1.2/go.mod h1:o9zRfWJQhW5j3PQxs7v0jmG4igD4peDatqbAGQFe4oo=
 github.com/siderolabs/grpc-proxy v0.5.1 h1:WTZYLMPTZPt43BzEJ02LT9kYA9qAfquWwCezc6NPPYE=
 github.com/siderolabs/grpc-proxy v0.5.1/go.mod h1:EQwE87LiWxhiIUPBeWmpjJb9DIWxWID8R6ARtdTC+8A=
 github.com/siderolabs/kms-client v0.1.0 h1:rCDWzcDDsNlp6zdyLngOuuhchVILn+vwUQy3tk6rQps=

internal/app/machined/pkg/startup/cgroups.go

@@ -14,6 +14,7 @@ import (
 	"github.com/containerd/cgroups/v3/cgroup1"
 	"github.com/containerd/cgroups/v3/cgroup2"
 	"github.com/opencontainers/runtime-spec/specs-go"
+	"github.com/siderolabs/go-debug"
 	"github.com/siderolabs/go-pointer"
 	"go.uber.org/zap"
 
@@ -22,6 +23,16 @@ import (
 	"github.com/siderolabs/talos/pkg/machinery/constants"
 )
 
+func zeroIfRace[T any](v T) T {
+	if debug.RaceEnabled {
+		var zeroT T
+
+		return zeroT
+	}
+
+	return v
+}
+
 // CreateSystemCgroups creates system cgroups.
 //
 //nolint:gocyclo
@@ -130,7 +141,7 @@ func CreateSystemCgroups(ctx context.Context, log *zap.Logger, rt runtime.Runtim
 			name: constants.CgroupDashboard,
 			resources: &cgroup2.Resources{
 				Memory: &cgroup2.Memory{
-					Max: pointer.To[int64](constants.CgroupDashboardMaxMemory),
+					Max: zeroIfRace(pointer.To[int64](constants.CgroupDashboardMaxMemory)),
 				},
 				CPU: &cgroup2.CPU{
 					Weight: pointer.To[uint64](cgroup.MillicoresToCPUWeight(cgroup.MilliCores(constants.CgroupDashboardMillicores))),
@@ -143,7 +154,7 @@ func CreateSystemCgroups(ctx context.Context, log *zap.Logger, rt runtime.Runtim
 				Memory: &cgroup2.Memory{
 					Min: pointer.To[int64](constants.CgroupApidReservedMemory),
 					Low: pointer.To[int64](constants.CgroupApidReservedMemory * 2),
-					Max: pointer.To[int64](constants.CgroupApidMaxMemory),
+					Max: zeroIfRace(pointer.To[int64](constants.CgroupApidMaxMemory)),
 				},
 				CPU: &cgroup2.CPU{
 					Weight: pointer.To[uint64](cgroup.MillicoresToCPUWeight(cgroup.MilliCores(constants.CgroupApidMillicores))),
@@ -156,7 +167,7 @@ func CreateSystemCgroups(ctx context.Context, log *zap.Logger, rt runtime.Runtim
 				Memory: &cgroup2.Memory{
 					Min: pointer.To[int64](constants.CgroupTrustdReservedMemory),
 					Low: pointer.To[int64](constants.CgroupTrustdReservedMemory * 2),
-					Max: pointer.To[int64](constants.CgroupTrustdMaxMemory),
+					Max: zeroIfRace(pointer.To[int64](constants.CgroupTrustdMaxMemory)),
 				},
 				CPU: &cgroup2.CPU{
 					Weight: pointer.To[uint64](cgroup.MillicoresToCPUWeight(cgroup.MilliCores(constants.CgroupTrustdMillicores))),

pkg/cluster/crashdump.go

@@ -9,6 +9,7 @@ import (
 	"fmt"
 	"io"
 	"os"
+	"time"
 
 	"github.com/siderolabs/gen/xslices"
 	"github.com/siderolabs/go-talos-support/support"
@@ -33,6 +34,10 @@ func Crashdump(ctx context.Context, cluster provision.Cluster, logWriter io.Writ
 
 	defer supportFile.Close() //nolint:errcheck
 
+	// limit support bundle generation time
+	ctx, cancel := context.WithTimeout(ctx, 5*time.Minute)
+	defer cancel()
+
 	c, err := client.New(ctx, client.WithDefaultConfig())
 	if err != nil {
 		fmt.Fprintf(logWriter, "error creating crashdump: %s\n", err)
@@ -50,7 +55,8 @@ func Crashdump(ctx context.Context, cluster provision.Cluster, logWriter io.Writ
 		bundle.WithArchiveOutput(supportFile),
 		bundle.WithTalosClient(c),
 		bundle.WithNodes(nodes...),
-		bundle.WithNumWorkers(1),
+		bundle.WithNumWorkers(4),
+		bundle.WithLogOutput(io.Discard),
 	}
 
 	kubeclient, err := getKubernetesClient(ctx, c, controlplane)