feat(collector): add Lenovo XCC profile to skip noisy snapshot paths

Lenovo ThinkSystem SR650 V3 (and similar XCC-based servers) caused
collection runs of 23+ minutes because the BMC exposes two large high-
error-rate subtrees in the snapshot BFS:

  - Chassis/1/Sensors: 315 individual sensor members, 282/315 failing,
    ~3.7s per request → ~19 minutes wasted. These documents are never
    read by any LOGPile parser (thermal/power data comes from aggregate
    Chassis/*/Thermal and Chassis/*/Power endpoints).

  - Chassis/1/Oem/Lenovo: 75 requests (LEDs×47, Slots×26, etc.),
    68/75 failing → 8+ minutes wasted on non-inventory data.

Add a Lenovo profile (matched on SystemManufacturer/OEMNamespace "Lenovo")
that sets SnapshotExcludeContains to block individual sensor documents and
non-inventory Lenovo OEM subtrees from the snapshot BFS queue. Also sets
rate policy thresholds appropriate for XCC BMC latency (p95 often 3-5s).

Add SnapshotExcludeContains []string to AcquisitionTuning and check it
in the snapshot enqueue closure in redfish.go.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

bible-local/03-api.md

@@ -58,6 +58,7 @@ Responses:
 
 Optional request field:
 - `power_on_if_host_off`: when `true`, Redfish collection may power on the host before collection if preflight found it powered off
+- `debug_payloads`: when `true`, collector keeps extra diagnostic payloads and enables extended plan-B retries for slow HGX component inventory branches (`Assembly`, `Accelerators`, `Drives`, `NetworkAdapters`, `PCIeDevices`)
 
 ### `POST /api/collect/probe`
 

bible-local/05-collectors.md

@@ -27,6 +27,7 @@ Request fields passed from the server:
 - credential field (`password` or token)
 - `tls_mode`
 - optional `power_on_if_host_off`
+- optional `debug_payloads` for extended diagnostics
 
 ### Core rule
 
@@ -57,6 +58,17 @@ closes `skipCh` → goroutine in `Collect()` → `cancelCollect()`.
 
 The skip button is visible during `running` state and hidden once the job reaches a terminal state.
 
+### Extended diagnostics toggle
+
+The live collect form exposes a user-facing checkbox for extended diagnostics.
+
+- default collection prioritizes inventory completeness and bounded runtime
+- when extended diagnostics is off, heavy HGX component-chassis critical plan-B retries
+  (`Assembly`, `Accelerators`, `Drives`, `NetworkAdapters`, `PCIeDevices`) are skipped
+- when extended diagnostics is on, those retries are allowed and extra debug payloads are collected
+
+This toggle is intended for operator-driven deep diagnostics on problematic hosts, not for the default path.
+
 ### Discovery model
 
 The collector does not rely on one fixed vendor tree.

bible-local/10-decisions.md

@@ -1120,3 +1120,20 @@ incomplete for UI and Reanimator consumers.
 - System firmware such as BIOS and iBMC versions survives xFusion file exports.
 - xFusion archives participate more reliably in canonical device/export flows without special UI
   cases.
+
+---
+
+## ADL-043 — Extended HGX diagnostic plan-B is opt-in from the live collect form
+
+**Date:** 2026-04-13
+**Context:** Some Supermicro HGX Redfish targets expose slow or hanging component-chassis inventory
+collections during critical plan-B, especially under `Chassis/HGX_*` for `Assembly`,
+`Accelerators`, `Drives`, `NetworkAdapters`, and `PCIeDevices`. Default collection should not
+block operators on deep diagnostic retries that are useful mainly for troubleshooting.
+**Decision:** Keep the normal snapshot/replay path unchanged, but gate those heavy HGX
+component-chassis critical plan-B retries behind the existing live-collect `debug_payloads` flag,
+presented in the UI as "Сбор расширенных данных для диагностики".
+**Consequences:**
+- Default live collection skips those heavy diagnostic plan-B retries and reaches replay faster.
+- Operators can explicitly opt into the slower diagnostic path when they need deeper collection.
+- The same user-facing toggle continues to enable extra debug payload capture for troubleshooting.

internal/collector/redfish.go

@@ -345,8 +345,9 @@ func (c *RedfishConnector) Collect(ctx context.Context, req Request, emit Progre
 				"manager_critical_suffixes": acquisitionPlan.ScopedPaths.ManagerCriticalSuffixes,
 			},
 			"tuning": map[string]any{
-				"snapshot_max_documents": acquisitionPlan.Tuning.SnapshotMaxDocuments,
-				"snapshot_workers":       acquisitionPlan.Tuning.SnapshotWorkers,
+				"snapshot_max_documents":     acquisitionPlan.Tuning.SnapshotMaxDocuments,
+				"snapshot_workers":           acquisitionPlan.Tuning.SnapshotWorkers,
+				"snapshot_exclude_contains":  acquisitionPlan.Tuning.SnapshotExcludeContains,
 				"prefetch_workers":       acquisitionPlan.Tuning.PrefetchWorkers,
 				"prefetch_enabled":       boolPointerValue(acquisitionPlan.Tuning.PrefetchEnabled),
 				"nvme_post_probe":        boolPointerValue(acquisitionPlan.Tuning.NVMePostProbeEnabled),
@@ -496,7 +497,6 @@ func (c *RedfishConnector) Collect(ctx context.Context, req Request, emit Progre
 	return result, nil
 }
 
-
 // collectDebugPayloads fetches vendor-specific diagnostic endpoints on a best-effort basis.
 // Results are stored in rawPayloads["redfish_debug_payloads"] and exported with the bundle.
 // Enabled only when Request.DebugPayloads is true.
@@ -511,7 +511,6 @@ func (c *RedfishConnector) collectDebugPayloads(ctx context.Context, client *htt
 	return out
 }
 
-
 func firstNonEmptyPath(paths []string, fallback string) string {
 	for _, p := range paths {
 		if strings.TrimSpace(p) != "" {
@@ -543,7 +542,6 @@ func redfishSystemPowerState(systemDoc map[string]interface{}) string {
 	return ""
 }
 
-
 func (c *RedfishConnector) postJSON(ctx context.Context, client *http.Client, req Request, baseURL, resourcePath string, payload map[string]any) error {
 	body, err := json.Marshal(payload)
 	if err != nil {
@@ -1346,6 +1344,11 @@ func (c *RedfishConnector) collectRawRedfishTree(ctx context.Context, client *ht
 		if !shouldCrawlPath(path) {
 			return
 		}
+		for _, pattern := range tuning.SnapshotExcludeContains {
+			if pattern != "" && strings.Contains(path, pattern) {
+				return
+			}
+		}
 		mu.Lock()
 		if len(seen) >= maxDocuments {
 			mu.Unlock()
@@ -2299,7 +2302,6 @@ func redfishCriticalSlowGap() time.Duration {
 	return 1200 * time.Millisecond
 }
 
-
 func redfishSnapshotMemoryRequestTimeout() time.Duration {
 	if v := strings.TrimSpace(os.Getenv("LOGPILE_REDFISH_MEMORY_TIMEOUT")); v != "" {
 		if d, err := time.ParseDuration(v); err == nil && d > 0 {
@@ -2878,11 +2880,16 @@ func (c *RedfishConnector) recoverCriticalRedfishDocsPlanB(ctx context.Context,
 	timings := newRedfishPathTimingCollector(4)
 	var targets []string
 	seenTargets := make(map[string]struct{})
+	skippedDiagnosticTargets := 0
 	addTarget := func(path string) {
 		path = normalizeRedfishPath(path)
 		if path == "" {
 			return
 		}
+		if !shouldIncludeCriticalPlanBPath(req, path) {
+			skippedDiagnosticTargets++
+			return
+		}
 		if _, ok := seenTargets[path]; ok {
 			return
 		}
@@ -2968,6 +2975,13 @@ func (c *RedfishConnector) recoverCriticalRedfishDocsPlanB(ctx context.Context,
 		return 0
 	}
 	if emit != nil {
+		if skippedDiagnosticTargets > 0 {
+			emit(Progress{
+				Status:   "running",
+				Progress: 97,
+				Message:  fmt.Sprintf("Redfish: расширенная диагностика выключена, пропущено %d тяжелых diagnostic endpoint", skippedDiagnosticTargets),
+			})
+		}
 		totalETA := redfishCriticalCooldown() + estimatePlanBETA(len(targets))
 		emit(Progress{
 			Status:       "running",
@@ -3073,6 +3087,39 @@ func (c *RedfishConnector) recoverCriticalRedfishDocsPlanB(ctx context.Context,
 	return recovered
 }
 
+func shouldIncludeCriticalPlanBPath(req Request, path string) bool {
+	if req.DebugPayloads {
+		return true
+	}
+	return !isExtendedDiagnosticCriticalPlanBPath(path)
+}
+
+func isExtendedDiagnosticCriticalPlanBPath(path string) bool {
+	path = normalizeRedfishPath(path)
+	if path == "" {
+		return false
+	}
+	parts := strings.Split(strings.Trim(path, "/"), "/")
+	if len(parts) < 5 || parts[0] != "redfish" || parts[1] != "v1" || parts[2] != "Chassis" {
+		return false
+	}
+	if !strings.HasPrefix(parts[3], "HGX_") {
+		return false
+	}
+	for _, suffix := range []string{
+		"/Accelerators",
+		"/Assembly",
+		"/Drives",
+		"/NetworkAdapters",
+		"/PCIeDevices",
+	} {
+		if strings.HasSuffix(path, suffix) {
+			return true
+		}
+	}
+	return false
+}
+
 func (c *RedfishConnector) recoverProfilePlanBDocs(ctx context.Context, client *http.Client, req Request, baseURL string, plan redfishprofile.AcquisitionPlan, rawTree map[string]interface{}, emit ProgressFn) int {
 	if len(plan.PlanBPaths) == 0 || plan.Mode == redfishprofile.ModeFallback || !plan.Tuning.RecoveryPolicy.EnableProfilePlanB {
 		return 0

internal/collector/redfish_planb_test.go

@@ -0,0 +1,57 @@
+package collector
+
+import "testing"
+
+func TestShouldIncludeCriticalPlanBPath(t *testing.T) {
+	tests := []struct {
+		name string
+		req  Request
+		path string
+		want bool
+	}{
+		{
+			name: "skip hgx erot pcie without extended diagnostics",
+			req:  Request{},
+			path: "/redfish/v1/Chassis/HGX_ERoT_NVSwitch_0/PCIeDevices",
+			want: false,
+		},
+		{
+			name: "skip hgx chassis assembly without extended diagnostics",
+			req:  Request{},
+			path: "/redfish/v1/Chassis/HGX_Chassis_0/Assembly",
+			want: false,
+		},
+		{
+			name: "keep standard chassis inventory without extended diagnostics",
+			req:  Request{},
+			path: "/redfish/v1/Chassis/1/PCIeDevices",
+			want: true,
+		},
+		{
+			name: "keep nvme storage backplane drives without extended diagnostics",
+			req:  Request{},
+			path: "/redfish/v1/Chassis/NVMeSSD.0.Group.0.StorageBackplane/Drives",
+			want: true,
+		},
+		{
+			name: "keep system processors without extended diagnostics",
+			req:  Request{},
+			path: "/redfish/v1/Systems/HGX_Baseboard_0/Processors",
+			want: true,
+		},
+		{
+			name: "include hgx erot pcie when extended diagnostics enabled",
+			req:  Request{DebugPayloads: true},
+			path: "/redfish/v1/Chassis/HGX_ERoT_NVSwitch_0/PCIeDevices",
+			want: true,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			if got := shouldIncludeCriticalPlanBPath(tt.req, tt.path); got != tt.want {
+				t.Fatalf("shouldIncludeCriticalPlanBPath(%q) = %v, want %v", tt.path, got, tt.want)
+			}
+		})
+	}
+}

internal/collector/redfishprofile/matcher_test.go

@@ -326,6 +326,47 @@ func TestBuildAnalysisDirectives_SupermicroEnablesStorageRecovery(t *testing.T)
 	}
 }
 
+func TestMatchProfiles_LenovoXCCSelectsMatchedModeAndExcludesSensors(t *testing.T) {
+	match := MatchProfiles(MatchSignals{
+		SystemManufacturer:  "Lenovo",
+		ChassisManufacturer: "Lenovo",
+		OEMNamespaces:       []string{"Lenovo"},
+	})
+	if match.Mode != ModeMatched {
+		t.Fatalf("expected matched mode, got %q", match.Mode)
+	}
+	found := false
+	for _, profile := range match.Profiles {
+		if profile.Name() == "lenovo" {
+			found = true
+			break
+		}
+	}
+	if !found {
+		t.Fatal("expected lenovo profile to be selected")
+	}
+
+	// Verify the acquisition plan excludes noisy Lenovo-specific snapshot paths.
+	plan := BuildAcquisitionPlan(MatchSignals{
+		SystemManufacturer:  "Lenovo",
+		ChassisManufacturer: "Lenovo",
+		OEMNamespaces:       []string{"Lenovo"},
+	})
+	wantExcluded := []string{"/Sensors/", "/Oem/Lenovo/LEDs/", "/Oem/Lenovo/Slots/"}
+	for _, want := range wantExcluded {
+		found := false
+		for _, ex := range plan.Tuning.SnapshotExcludeContains {
+			if ex == want {
+				found = true
+				break
+			}
+		}
+		if !found {
+			t.Errorf("expected SnapshotExcludeContains to include %q, got %v", want, plan.Tuning.SnapshotExcludeContains)
+		}
+	}
+}
+
 func TestMatchProfiles_OrderingIsDeterministic(t *testing.T) {
 	signals := MatchSignals{
 		SystemManufacturer: "Micro-Star International Co., Ltd.",

internal/collector/redfishprofile/profile_lenovo.go

@@ -0,0 +1,65 @@
+package redfishprofile
+
+func lenovoProfile() Profile {
+	return staticProfile{
+		name:            "lenovo",
+		priority:        20,
+		safeForFallback: true,
+		matchFn: func(s MatchSignals) int {
+			score := 0
+			if containsFold(s.SystemManufacturer, "lenovo") ||
+				containsFold(s.ChassisManufacturer, "lenovo") {
+				score += 80
+			}
+			for _, ns := range s.OEMNamespaces {
+				if containsFold(ns, "lenovo") {
+					score += 30
+					break
+				}
+			}
+			// Lenovo XClarity Controller (XCC) is the BMC product line.
+			if containsFold(s.ServiceRootProduct, "xclarity") ||
+				containsFold(s.ServiceRootProduct, "xcc") {
+				score += 30
+			}
+			return min(score, 100)
+		},
+		extendAcquisition: func(plan *AcquisitionPlan, _ MatchSignals) {
+			// Lenovo XCC BMC exposes Chassis/1/Sensors with hundreds of individual
+			// sensor member documents (e.g. Chassis/1/Sensors/101L1). These are
+			// not used by any LOGPile parser — thermal/power data is read from
+			// the aggregate Chassis/*/Thermal and Chassis/*/Power endpoints. On
+			// a real server they largely return errors, wasting many minutes.
+			// Lenovo OEM subtrees under Oem/Lenovo/LEDs and Oem/Lenovo/Slots also
+			// enumerate dozens of individual documents not relevant to inventory.
+			ensureSnapshotExcludeContains(plan,
+				"/Sensors/",               // individual sensor docs (Chassis/1/Sensors/NNN)
+				"/Oem/Lenovo/LEDs/",       // individual LED status entries (~47 per server)
+				"/Oem/Lenovo/Slots/",      // individual slot detail entries (~26 per server)
+				"/Oem/Lenovo/Metrics/",    // operational metrics, not inventory
+				"/Oem/Lenovo/History",     // historical telemetry
+				"/Oem/Lenovo/ScheduledPower", // power scheduling config
+				"/Oem/Lenovo/BootSettings/BootOrder", // individual boot order lists
+				"/PortForwardingMap/",     // network port forwarding config
+			)
+			// Lenovo XCC BMC is typically slow (p95 latency often 3-5s even under
+			// normal load). Set rate thresholds that don't over-throttle on the
+			// first few requests, and give the ETA estimator a realistic baseline.
+			ensureRatePolicy(plan, AcquisitionRatePolicy{
+				TargetP95LatencyMS:      2000,
+				ThrottleP95LatencyMS:    4000,
+				MinSnapshotWorkers:      2,
+				MinPrefetchWorkers:      1,
+				DisablePrefetchOnErrors: true,
+			})
+			ensureETABaseline(plan, AcquisitionETABaseline{
+				DiscoverySeconds:     15,
+				SnapshotSeconds:      120,
+				PrefetchSeconds:      30,
+				CriticalPlanBSeconds: 40,
+				ProfilePlanBSeconds:  20,
+			})
+			addPlanNote(plan, "lenovo xcc acquisition extensions enabled: noisy sensor/oem paths excluded from snapshot")
+		},
+	}
+}

internal/collector/redfishprofile/profiles_common.go

@@ -56,6 +56,7 @@ func BuiltinProfiles() []Profile {
 		supermicroProfile(),
 		dellProfile(),
 		hpeProfile(),
+		lenovoProfile(),
 		inspurGroupOEMPlatformsProfile(),
 		hgxProfile(),
 		xfusionProfile(),
@@ -226,6 +227,10 @@ func ensurePrefetchPolicy(plan *AcquisitionPlan, policy AcquisitionPrefetchPolic
 	addPlanPaths(&plan.Tuning.PrefetchPolicy.ExcludeContains, policy.ExcludeContains...)
 }
 
+func ensureSnapshotExcludeContains(plan *AcquisitionPlan, patterns ...string) {
+	addPlanPaths(&plan.Tuning.SnapshotExcludeContains, patterns...)
+}
+
 func min(a, b int) int {
 	if a < b {
 		return a

internal/collector/redfishprofile/types.go

@@ -53,16 +53,17 @@ type AcquisitionScopedPathPolicy struct {
 }
 
 type AcquisitionTuning struct {
-	SnapshotMaxDocuments int
-	SnapshotWorkers      int
-	PrefetchEnabled      *bool
-	PrefetchWorkers      int
-	NVMePostProbeEnabled *bool
-	RatePolicy           AcquisitionRatePolicy
-	ETABaseline          AcquisitionETABaseline
-	PostProbePolicy      AcquisitionPostProbePolicy
-	RecoveryPolicy       AcquisitionRecoveryPolicy
-	PrefetchPolicy       AcquisitionPrefetchPolicy
+	SnapshotMaxDocuments    int
+	SnapshotWorkers         int
+	SnapshotExcludeContains []string
+	PrefetchEnabled         *bool
+	PrefetchWorkers         int
+	NVMePostProbeEnabled    *bool
+	RatePolicy              AcquisitionRatePolicy
+	ETABaseline             AcquisitionETABaseline
+	PostProbePolicy         AcquisitionPostProbePolicy
+	RecoveryPolicy          AcquisitionRecoveryPolicy
+	PrefetchPolicy          AcquisitionPrefetchPolicy
 }
 
 type AcquisitionRatePolicy struct {

internal/exporter/reanimator_converter.go

@@ -1961,7 +1961,10 @@ func pcieDedupKey(item ReanimatorPCIe) string {
 	slot := strings.ToLower(strings.TrimSpace(item.Slot))
 	serial := strings.ToLower(strings.TrimSpace(item.SerialNumber))
 	bdf := strings.ToLower(strings.TrimSpace(item.BDF))
-	if slot != "" {
+	// Generic slot names (e.g. "PCIe Device" from HGX BMC) are not unique
+	// hardware positions — multiple distinct devices share the same name.
+	// Fall through to serial/BDF so they are not incorrectly collapsed.
+	if slot != "" && !isGenericPCIeSlotName(slot) {
 		return "slot:" + slot
 	}
 	if serial != "" {
@@ -1970,9 +1973,22 @@ func pcieDedupKey(item ReanimatorPCIe) string {
 	if bdf != "" {
 		return "bdf:" + bdf
 	}
+	if slot != "" {
+		return "slot:" + slot
+	}
 	return strings.ToLower(strings.TrimSpace(item.DeviceClass)) + "|" + strings.ToLower(strings.TrimSpace(item.Model))
 }
 
+// isGenericPCIeSlotName reports whether slot is a generic device-type label
+// rather than a unique hardware position identifier.
+func isGenericPCIeSlotName(slot string) bool {
+	switch slot {
+	case "pcie device", "pcie slot", "pcie":
+		return true
+	}
+	return false
+}
+
 func pcieQualityScore(item ReanimatorPCIe) int {
 	score := 0
 	if strings.TrimSpace(item.SerialNumber) != "" {

internal/exporter/reanimator_converter_test.go

@@ -733,6 +733,42 @@ func TestConvertPCIeDevices_SkipsDisplayControllerDuplicates(t *testing.T) {
 	}
 }
 
+func TestConvertPCIeDevices_PreservesAllGPUsWithGenericSlot(t *testing.T) {
+	// Supermicro HGX BMC reports all GPU PCIe devices with Name "PCIe Device" —
+	// a generic label that is not a unique hardware position. All 8 GPUs must
+	// be preserved; dedup by generic slot name must not collapse them into one.
+	gpus := make([]models.GPU, 8)
+	serials := []string{
+		"1654925165720", "1654925166160", "1654925165942", "1654925165271",
+		"1654925165719", "1654925165252", "1654925165304", "1654925165587",
+	}
+	for i, sn := range serials {
+		gpus[i] = models.GPU{
+			Slot:         "PCIe Device",
+			Model:        "B200 180GB HBM3e",
+			Manufacturer: "NVIDIA",
+			SerialNumber: sn,
+			PartNumber:   "2901-886-A1",
+			Status:       "OK",
+		}
+	}
+	hw := &models.HardwareConfig{GPUs: gpus}
+	result := convertPCIeDevices(hw, "2026-04-13T10:00:00Z")
+	if len(result) != 8 {
+		t.Fatalf("expected 8 GPU entries (one per serial), got %d", len(result))
+	}
+	seen := make(map[string]bool)
+	for _, r := range result {
+		if seen[r.SerialNumber] {
+			t.Fatalf("duplicate serial %q in PCIe result", r.SerialNumber)
+		}
+		seen[r.SerialNumber] = true
+		if r.DeviceClass != "VideoController" {
+			t.Fatalf("expected VideoController device class, got %q", r.DeviceClass)
+		}
+	}
+}
+
 func TestConvertPCIeDevices_MapsGPUStatusHistory(t *testing.T) {
 	hw := &models.HardwareConfig{
 		GPUs: []models.GPU{

web/templates/index.html

@@ -85,7 +85,7 @@
                         </div>
                         <label class="api-form-checkbox" for="api-debug-payloads">
                             <input id="api-debug-payloads" name="debug_payloads" type="checkbox">
-                            <span>Сбор расширенных метрик для отладки</span>
+                            <span>Сбор расширенных данных для диагностики</span>
                         </label>
                         <div class="api-form-actions">
                             <button id="api-collect-btn" type="submit">Собрать</button>