fix(redfish): gate hgx diagnostic plan-b by debug toggle

parseGPUWithSupplementalDocs did not read PCIeInterface from the device
doc, only from function docs. xFusion GPU PCIeCard entries carry link
width/speed in PCIeInterface (LanesInUse/Maxlanes/PCIeType/MaxPCIeType)
so GPU link width was always empty for xFusion servers.

Also apply the xFusion OEM function-level fallback for GPU function docs,
consistent with the NIC and PCIeDevice paths.

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

@@ -496,7 +496,6 @@ 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 +510,6 @@ 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 +541,6 @@ 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 {
@@ -2299,7 +2296,6 @@ 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 +2874,16 @@ 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 +2969,13 @@ 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 +3081,39 @@ 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
@@ -3592,7 +3633,7 @@ func parseNIC(doc map[string]interface{}) models.NetworkAdapter {
 			}
 			if pcieIf, ok := ctrl["PCIeInterface"].(map[string]interface{}); ok && linkWidth == 0 && maxLinkWidth == 0 && linkSpeed == "" && maxLinkSpeed == "" {
 				linkWidth = asInt(pcieIf["LanesInUse"])
-				maxLinkWidth = asInt(pcieIf["MaxLanes"])
+				maxLinkWidth = firstNonZeroInt(asInt(pcieIf["MaxLanes"]), asInt(pcieIf["Maxlanes"]))
 				linkSpeed = firstNonEmpty(asString(pcieIf["PCIeType"]), asString(pcieIf["CurrentLinkSpeedGTs"]), asString(pcieIf["CurrentLinkSpeed"]))
 				maxLinkSpeed = firstNonEmpty(asString(pcieIf["MaxPCIeType"]), asString(pcieIf["MaxLinkSpeedGTs"]), asString(pcieIf["MaxLinkSpeed"]))
 			}
@@ -3705,6 +3746,9 @@ func enrichNICFromPCIe(nic *models.NetworkAdapter, pcieDoc map[string]interface{
 	if strings.TrimSpace(nic.MaxLinkSpeed) == "" {
 		nic.MaxLinkSpeed = firstNonEmpty(asString(pcieDoc["MaxLinkSpeedGTs"]), asString(pcieDoc["MaxLinkSpeed"]))
 	}
+	if nic.LinkWidth == 0 || nic.MaxLinkWidth == 0 || nic.LinkSpeed == "" || nic.MaxLinkSpeed == "" {
+		redfishEnrichFromOEMxFusionPCIeLink(pcieDoc, &nic.LinkWidth, &nic.MaxLinkWidth, &nic.LinkSpeed, &nic.MaxLinkSpeed)
+	}
 	if normalizeRedfishIdentityField(nic.SerialNumber) == "" {
 		nic.SerialNumber = findFirstNormalizedStringByKeys(pcieDoc, "SerialNumber")
 	}
@@ -3736,6 +3780,9 @@ func enrichNICFromPCIe(nic *models.NetworkAdapter, pcieDoc map[string]interface{
 		if strings.TrimSpace(nic.MaxLinkSpeed) == "" {
 			nic.MaxLinkSpeed = firstNonEmpty(asString(fn["MaxLinkSpeedGTs"]), asString(fn["MaxLinkSpeed"]))
 		}
+		if nic.LinkWidth == 0 || nic.MaxLinkWidth == 0 || nic.LinkSpeed == "" || nic.MaxLinkSpeed == "" {
+			redfishEnrichFromOEMxFusionPCIeLink(fn, &nic.LinkWidth, &nic.MaxLinkWidth, &nic.LinkSpeed, &nic.MaxLinkSpeed)
+		}
 		if normalizeRedfishIdentityField(nic.SerialNumber) == "" {
 			nic.SerialNumber = findFirstNormalizedStringByKeys(fn, "SerialNumber")
 		}
@@ -4302,6 +4349,21 @@ func parseGPUWithSupplementalDocs(doc map[string]interface{}, functionDocs []map
 		gpu.DeviceID = asHexOrInt(doc["DeviceId"])
 	}
 
+	if pcieIf, ok := doc["PCIeInterface"].(map[string]interface{}); ok {
+		if gpu.CurrentLinkWidth == 0 {
+			gpu.CurrentLinkWidth = asInt(pcieIf["LanesInUse"])
+		}
+		if gpu.MaxLinkWidth == 0 {
+			gpu.MaxLinkWidth = firstNonZeroInt(asInt(pcieIf["MaxLanes"]), asInt(pcieIf["Maxlanes"]))
+		}
+		if gpu.CurrentLinkSpeed == "" {
+			gpu.CurrentLinkSpeed = firstNonEmpty(asString(pcieIf["PCIeType"]), asString(pcieIf["CurrentLinkSpeedGTs"]), asString(pcieIf["CurrentLinkSpeed"]))
+		}
+		if gpu.MaxLinkSpeed == "" {
+			gpu.MaxLinkSpeed = firstNonEmpty(asString(pcieIf["MaxPCIeType"]), asString(pcieIf["MaxLinkSpeedGTs"]), asString(pcieIf["MaxLinkSpeed"]))
+		}
+	}
+
 	for _, fn := range functionDocs {
 		if gpu.BDF == "" {
 			gpu.BDF = sanitizeRedfishBDF(asString(fn["FunctionId"]))
@@ -4324,6 +4386,9 @@ func parseGPUWithSupplementalDocs(doc map[string]interface{}, functionDocs []map
 		if gpu.CurrentLinkSpeed == "" {
 			gpu.CurrentLinkSpeed = firstNonEmpty(asString(fn["CurrentLinkSpeedGTs"]), asString(fn["CurrentLinkSpeed"]))
 		}
+		if gpu.CurrentLinkWidth == 0 || gpu.MaxLinkWidth == 0 || gpu.CurrentLinkSpeed == "" || gpu.MaxLinkSpeed == "" {
+			redfishEnrichFromOEMxFusionPCIeLink(fn, &gpu.CurrentLinkWidth, &gpu.MaxLinkWidth, &gpu.CurrentLinkSpeed, &gpu.MaxLinkSpeed)
+		}
 	}
 
 	if isMissingOrRawPCIModel(gpu.Model) {
@@ -4384,6 +4449,9 @@ func parsePCIeDeviceWithSupplementalDocs(doc map[string]interface{}, functionDoc
 		if dev.MaxLinkSpeed == "" {
 			dev.MaxLinkSpeed = firstNonEmpty(asString(fn["MaxLinkSpeedGTs"]), asString(fn["MaxLinkSpeed"]))
 		}
+		if dev.LinkWidth == 0 || dev.MaxLinkWidth == 0 || dev.LinkSpeed == "" || dev.MaxLinkSpeed == "" {
+			redfishEnrichFromOEMxFusionPCIeLink(fn, &dev.LinkWidth, &dev.MaxLinkWidth, &dev.LinkSpeed, &dev.MaxLinkSpeed)
+		}
 	}
 	if dev.DeviceClass == "" || isGenericPCIeClassLabel(dev.DeviceClass) {
 		dev.DeviceClass = firstNonEmpty(redfishFirstStringAcrossDocs(supplementalDocs, "DeviceType"), dev.DeviceClass)
@@ -4633,6 +4701,59 @@ func buildBDFfromOemPublic(doc map[string]interface{}) string {
 	return fmt.Sprintf("%04x:%02x:%02x.%x", segment, bus, dev, fn)
 }
 
+// redfishEnrichFromOEMxFusionPCIeLink fills in missing PCIe link width/speed
+// from the xFusion OEM namespace. xFusion reports link width as a string like
+// "X8" in Oem.xFusion.LinkWidth / Oem.xFusion.LinkWidthAbility, and link speed
+// as a string like "Gen4 (16.0GT/s)" in Oem.xFusion.LinkSpeed /
+// Oem.xFusion.LinkSpeedAbility. These fields appear on PCIeFunction docs.
+func redfishEnrichFromOEMxFusionPCIeLink(doc map[string]interface{}, linkWidth, maxLinkWidth *int, linkSpeed, maxLinkSpeed *string) {
+	oem, _ := doc["Oem"].(map[string]interface{})
+	if oem == nil {
+		return
+	}
+	xf, _ := oem["xFusion"].(map[string]interface{})
+	if xf == nil {
+		return
+	}
+	if *linkWidth == 0 {
+		*linkWidth = parseXFusionLinkWidth(asString(xf["LinkWidth"]))
+	}
+	if *maxLinkWidth == 0 {
+		*maxLinkWidth = parseXFusionLinkWidth(asString(xf["LinkWidthAbility"]))
+	}
+	if strings.TrimSpace(*linkSpeed) == "" {
+		*linkSpeed = strings.TrimSpace(asString(xf["LinkSpeed"]))
+	}
+	if strings.TrimSpace(*maxLinkSpeed) == "" {
+		*maxLinkSpeed = strings.TrimSpace(asString(xf["LinkSpeedAbility"]))
+	}
+}
+
+// parseXFusionLinkWidth converts an xFusion link-width string like "X8" or
+// "x16" to the integer lane count. Returns 0 for unrecognised values.
+func parseXFusionLinkWidth(s string) int {
+	s = strings.TrimSpace(s)
+	if s == "" {
+		return 0
+	}
+	s = strings.TrimPrefix(strings.ToUpper(s), "X")
+	v := asInt(s)
+	if v <= 0 {
+		return 0
+	}
+	return v
+}
+
+// firstNonZeroInt returns the first argument that is non-zero.
+func firstNonZeroInt(vals ...int) int {
+	for _, v := range vals {
+		if v != 0 {
+			return v
+		}
+	}
+	return 0
+}
+
 func normalizeRedfishIdentityField(v string) string {
 	v = strings.TrimSpace(v)
 	if v == "" {

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/redfish_test.go

@@ -1341,6 +1341,48 @@ func TestParseNIC_PrefersControllerSlotLabelAndPCIeInterface(t *testing.T) {
 	}
 }
 
+func TestParseNIC_xFusionMaxlanesAndOEMLinkWidth(t *testing.T) {
+	// xFusion uses "Maxlanes" (lowercase 'l') in PCIeInterface, not "MaxLanes".
+	// xFusion also stores per-function link width as Oem.xFusion.LinkWidth = "X8".
+	nic := parseNIC(map[string]interface{}{
+		"Id":    "OCPCard1",
+		"Model": "ConnectX-6 Lx",
+		"Controllers": []interface{}{
+			map[string]interface{}{
+				"PCIeInterface": map[string]interface{}{
+					"LanesInUse":  8,
+					"Maxlanes":    8, // xFusion uses lowercase 'l'
+					"PCIeType":    "Gen4",
+					"MaxPCIeType": "Gen4",
+				},
+			},
+		},
+	})
+	if nic.LinkWidth != 8 || nic.MaxLinkWidth != 8 {
+		t.Fatalf("expected link widths 8/8 from xFusion Maxlanes, got current=%d max=%d", nic.LinkWidth, nic.MaxLinkWidth)
+	}
+
+	// enrichNICFromPCIe: OEM xFusion LinkWidth on a PCIeFunction doc.
+	nic2 := models.NetworkAdapter{}
+	fnDoc := map[string]interface{}{
+		"Oem": map[string]interface{}{
+			"xFusion": map[string]interface{}{
+				"LinkWidth":        "X8",
+				"LinkWidthAbility": "X8",
+				"LinkSpeed":        "Gen4 (16.0GT/s)",
+				"LinkSpeedAbility": "Gen4 (16.0GT/s)",
+			},
+		},
+	}
+	enrichNICFromPCIe(&nic2, map[string]interface{}{}, []map[string]interface{}{fnDoc}, nil)
+	if nic2.LinkWidth != 8 || nic2.MaxLinkWidth != 8 {
+		t.Fatalf("expected link width 8 from xFusion OEM LinkWidth, got current=%d max=%d", nic2.LinkWidth, nic2.MaxLinkWidth)
+	}
+	if nic2.LinkSpeed != "Gen4 (16.0GT/s)" || nic2.MaxLinkSpeed != "Gen4 (16.0GT/s)" {
+		t.Fatalf("expected link speed from xFusion OEM LinkSpeed, got current=%q max=%q", nic2.LinkSpeed, nic2.MaxLinkSpeed)
+	}
+}
+
 func TestParseNIC_DropsUnrealisticPortCount(t *testing.T) {
 	nic := parseNIC(map[string]interface{}{
 		"Id": "1",
@@ -2773,6 +2815,28 @@ func TestReplayCollectGPUs_DedupUsesRedfishPathBeforeHeuristics(t *testing.T) {
 	}
 }
 
+func TestParseGPU_xFusionPCIeInterfaceMaxlanes(t *testing.T) {
+	// xFusion GPU PCIeDevices (PCIeCard1..N) carry link width in PCIeInterface
+	// with "Maxlanes" (lowercase 'l') rather than "MaxLanes".
+	doc := map[string]interface{}{
+		"Id":    "PCIeCard1",
+		"Model": "RTX PRO 6000",
+		"PCIeInterface": map[string]interface{}{
+			"LanesInUse":  16,
+			"Maxlanes":    16,
+			"PCIeType":    "Gen5",
+			"MaxPCIeType": "Gen5",
+		},
+	}
+	gpu := parseGPU(doc, nil, 1)
+	if gpu.CurrentLinkWidth != 16 || gpu.MaxLinkWidth != 16 {
+		t.Fatalf("expected link widths 16/16 from PCIeInterface, got current=%d max=%d", gpu.CurrentLinkWidth, gpu.MaxLinkWidth)
+	}
+	if gpu.CurrentLinkSpeed != "Gen5" || gpu.MaxLinkSpeed != "Gen5" {
+		t.Fatalf("expected link speeds Gen5/Gen5 from PCIeInterface, got current=%q max=%q", gpu.CurrentLinkSpeed, gpu.MaxLinkSpeed)
+	}
+}
+
 func TestParseGPU_UsesNestedOemSerialNumber(t *testing.T) {
 	doc := map[string]interface{}{
 		"Id":           "GPU4",

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>