Fix full-history metrics range loading

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

audit/internal/collector/nvidia.go

@@ -13,14 +13,18 @@ import (
 const nvidiaVendorID = 0x10de
 
 type nvidiaGPUInfo struct {
-	BDF            string
+	BDF                string
-	Serial         string
+	Serial             string
-	VBIOS          string
+	VBIOS              string
-	TemperatureC   *float64
+	TemperatureC       *float64
-	PowerW         *float64
+	PowerW             *float64
-	ECCUncorrected *int64
+	ECCUncorrected     *int64
-	ECCCorrected   *int64
+	ECCCorrected       *int64
-	HWSlowdown     *bool
+	HWSlowdown         *bool
+	PCIeLinkGenCurrent *int
+	PCIeLinkGenMax     *int
+	PCIeLinkWidthCur   *int
+	PCIeLinkWidthMax   *int
 }
 
 // enrichPCIeWithNVIDIA enriches NVIDIA PCIe devices with data from nvidia-smi.
@@ -94,7 +98,7 @@ func enrichPCIeWithNVIDIAData(devs []schema.HardwarePCIeDevice, gpuByBDF map[str
 func queryNVIDIAGPUs() (map[string]nvidiaGPUInfo, error) {
 	out, err := exec.Command(
 		"nvidia-smi",
-		"--query-gpu=index,pci.bus_id,serial,vbios_version,temperature.gpu,power.draw,ecc.errors.uncorrected.aggregate.total,ecc.errors.corrected.aggregate.total,clocks_throttle_reasons.hw_slowdown",
+		"--query-gpu=index,pci.bus_id,serial,vbios_version,temperature.gpu,power.draw,ecc.errors.uncorrected.aggregate.total,ecc.errors.corrected.aggregate.total,clocks_throttle_reasons.hw_slowdown,pcie.link.gen.current,pcie.link.gen.max,pcie.link.width.current,pcie.link.width.max",
 		"--format=csv,noheader,nounits",
 	).Output()
 	if err != nil {
@@ -118,8 +122,8 @@ func parseNVIDIASMIQuery(raw string) (map[string]nvidiaGPUInfo, error) {
 		if len(rec) == 0 {
 			continue
 		}
-		if len(rec) < 9 {
+		if len(rec) < 13 {
-			return nil, fmt.Errorf("unexpected nvidia-smi columns: got %d, want 9", len(rec))
+			return nil, fmt.Errorf("unexpected nvidia-smi columns: got %d, want 13", len(rec))
 		}
 
 		bdf := normalizePCIeBDF(rec[1])
@@ -128,14 +132,18 @@ func parseNVIDIASMIQuery(raw string) (map[string]nvidiaGPUInfo, error) {
 		}
 
 		info := nvidiaGPUInfo{
-			BDF:            bdf,
+			BDF:                bdf,
-			Serial:         strings.TrimSpace(rec[2]),
+			Serial:             strings.TrimSpace(rec[2]),
-			VBIOS:          strings.TrimSpace(rec[3]),
+			VBIOS:              strings.TrimSpace(rec[3]),
-			TemperatureC:   parseMaybeFloat(rec[4]),
+			TemperatureC:       parseMaybeFloat(rec[4]),
-			PowerW:         parseMaybeFloat(rec[5]),
+			PowerW:             parseMaybeFloat(rec[5]),
-			ECCUncorrected: parseMaybeInt64(rec[6]),
+			ECCUncorrected:     parseMaybeInt64(rec[6]),
-			ECCCorrected:   parseMaybeInt64(rec[7]),
+			ECCCorrected:       parseMaybeInt64(rec[7]),
-			HWSlowdown:     parseMaybeBool(rec[8]),
+			HWSlowdown:         parseMaybeBool(rec[8]),
+			PCIeLinkGenCurrent: parseMaybeInt(rec[9]),
+			PCIeLinkGenMax:     parseMaybeInt(rec[10]),
+			PCIeLinkWidthCur:   parseMaybeInt(rec[11]),
+			PCIeLinkWidthMax:   parseMaybeInt(rec[12]),
 		}
 		result[bdf] = info
 	}
@@ -167,6 +175,22 @@ func parseMaybeInt64(v string) *int64 {
 	return &n
 }
 
+func parseMaybeInt(v string) *int {
+	v = strings.TrimSpace(v)
+	if v == "" || strings.EqualFold(v, "n/a") || strings.EqualFold(v, "not supported") || strings.EqualFold(v, "[not supported]") {
+		return nil
+	}
+	n, err := strconv.Atoi(v)
+	if err != nil {
+		return nil
+	}
+	return &n
+}
+
+func pcieLinkGenLabel(gen int) string {
+	return fmt.Sprintf("Gen%d", gen)
+}
+
 func parseMaybeBool(v string) *bool {
 	v = strings.TrimSpace(strings.ToLower(v))
 	switch v {
@@ -231,4 +255,22 @@ func injectNVIDIATelemetry(dev *schema.HardwarePCIeDevice, info nvidiaGPUInfo) {
 	if info.HWSlowdown != nil {
 		dev.HWSlowdown = info.HWSlowdown
 	}
+	// Override PCIe link speed/width with nvidia-smi driver values.
+	// sysfs current_link_speed reflects the instantaneous physical link state and
+	// can show Gen1 when the GPU is idle due to ASPM power management. The driver
+	// knows the negotiated speed regardless of the current power state.
+	if info.PCIeLinkGenCurrent != nil {
+		s := pcieLinkGenLabel(*info.PCIeLinkGenCurrent)
+		dev.LinkSpeed = &s
+	}
+	if info.PCIeLinkGenMax != nil {
+		s := pcieLinkGenLabel(*info.PCIeLinkGenMax)
+		dev.MaxLinkSpeed = &s
+	}
+	if info.PCIeLinkWidthCur != nil {
+		dev.LinkWidth = info.PCIeLinkWidthCur
+	}
+	if info.PCIeLinkWidthMax != nil {
+		dev.MaxLinkWidth = info.PCIeLinkWidthMax
+	}
 }

audit/internal/collector/nvidia_test.go

@@ -6,7 +6,7 @@ import (
 )
 
 func TestParseNVIDIASMIQuery(t *testing.T) {
-	raw := "0, 00000000:65:00.0, GPU-SERIAL-1, 96.00.1F.00.02, 54, 210.33, 0, 5, Not Active\n"
+	raw := "0, 00000000:65:00.0, GPU-SERIAL-1, 96.00.1F.00.02, 54, 210.33, 0, 5, Not Active, 4, 4, 16, 16\n"
 	byBDF, err := parseNVIDIASMIQuery(raw)
 	if err != nil {
 		t.Fatalf("parse failed: %v", err)
@@ -28,6 +28,12 @@ func TestParseNVIDIASMIQuery(t *testing.T) {
 	if gpu.HWSlowdown == nil || *gpu.HWSlowdown {
 		t.Fatalf("hw slowdown: got %v, want false", gpu.HWSlowdown)
 	}
+	if gpu.PCIeLinkGenCurrent == nil || *gpu.PCIeLinkGenCurrent != 4 {
+		t.Fatalf("pcie link gen current: got %v, want 4", gpu.PCIeLinkGenCurrent)
+	}
+	if gpu.PCIeLinkGenMax == nil || *gpu.PCIeLinkGenMax != 4 {
+		t.Fatalf("pcie link gen max: got %v, want 4", gpu.PCIeLinkGenMax)
+	}
 }
 
 func TestNormalizePCIeBDF(t *testing.T) {

audit/internal/collector/storage.go

@@ -77,11 +77,24 @@ func discoverStorageDevices() []lsblkDevice {
 		if dev.Type != "disk" {
 			continue
 		}
+		if isVirtualBMCDisk(dev) {
+			slog.Debug("storage: skipping BMC virtual disk", "name", dev.Name, "model", dev.Model)
+			continue
+		}
 		disks = append(disks, dev)
 	}
 	return disks
 }
 
+// isVirtualBMCDisk returns true for BMC/IPMI virtual USB mass storage devices
+// that appear as disks but are not real hardware (e.g. iDRAC Virtual HDisk*).
+// These have zero reported size, a generic fake serial, and a model name that
+// starts with "Virtual HDisk".
+func isVirtualBMCDisk(dev lsblkDevice) bool {
+	model := strings.ToLower(strings.TrimSpace(dev.Model))
+	return strings.HasPrefix(model, "virtual hdisk")
+}
+
 func lsblkDevices() []lsblkDevice {
 	out, err := exec.Command("lsblk", "-J", "-d",
 		"-o", "NAME,TYPE,SIZE,SERIAL,MODEL,TRAN,HCTL").Output()

audit/internal/platform/sat.go

@@ -12,6 +12,7 @@ import (
 	"os"
 	"os/exec"
 	"path/filepath"
+	"syscall"
 	"sort"
 	"strconv"
 	"strings"
@@ -531,6 +532,13 @@ func runSATCommandCtx(ctx context.Context, verboseLog, name string, cmd []string
 	}
 
 	c := exec.CommandContext(ctx, resolvedCmd[0], resolvedCmd[1:]...)
+	c.SysProcAttr = &syscall.SysProcAttr{Setpgid: true}
+	c.Cancel = func() error {
+		if c.Process != nil {
+			_ = syscall.Kill(-c.Process.Pid, syscall.SIGKILL)
+		}
+		return nil
+	}
 	if len(env) > 0 {
 		c.Env = append(os.Environ(), env...)
 	}

audit/internal/webui/api.go

@@ -346,8 +346,10 @@ func (h *handler) handleAPINetworkStatus(w http.ResponseWriter, r *http.Request)
 		return
 	}
 	writeJSON(w, map[string]any{
-		"interfaces":    ifaces,
+		"interfaces":     ifaces,
-		"default_route": h.opts.App.DefaultRoute(),
+		"default_route":  h.opts.App.DefaultRoute(),
+		"pending_change": h.hasPendingNetworkChange(),
+		"rollback_in":    h.pendingNetworkRollbackIn(),
 	})
 }
 
@@ -744,13 +746,7 @@ func (h *handler) feedRings(sample platform.LiveMetricSample) {
 	h.ringMemLoad.push(sample.MemLoadPct)
 
 	h.ringsMu.Lock()
-	for i, fan := range sample.Fans {
+	h.pushFanRings(sample.Fans)
-		for len(h.ringFans) <= i {
-			h.ringFans = append(h.ringFans, newMetricsRing(120))
-			h.fanNames = append(h.fanNames, fan.Name)
-		}
-		h.ringFans[i].push(float64(fan.RPM))
-	}
 	for _, gpu := range sample.GPUs {
 		idx := gpu.GPUIndex
 		for len(h.gpuRings) <= idx {
@@ -769,6 +765,51 @@ func (h *handler) feedRings(sample platform.LiveMetricSample) {
 	h.ringsMu.Unlock()
 }
 
+func (h *handler) pushFanRings(fans []platform.FanReading) {
+	if len(fans) == 0 && len(h.ringFans) == 0 {
+		return
+	}
+	fanValues := make(map[string]float64, len(fans))
+	for _, fan := range fans {
+		if fan.Name == "" {
+			continue
+		}
+		fanValues[fan.Name] = fan.RPM
+		found := false
+		for i, name := range h.fanNames {
+			if name == fan.Name {
+				found = true
+				if i >= len(h.ringFans) {
+					h.ringFans = append(h.ringFans, newMetricsRing(120))
+				}
+				break
+			}
+		}
+		if !found {
+			h.fanNames = append(h.fanNames, fan.Name)
+			h.ringFans = append(h.ringFans, newMetricsRing(120))
+		}
+	}
+	for i, ring := range h.ringFans {
+		if ring == nil {
+			continue
+		}
+		name := ""
+		if i < len(h.fanNames) {
+			name = h.fanNames[i]
+		}
+		if rpm, ok := fanValues[name]; ok {
+			ring.push(rpm)
+			continue
+		}
+		if last, ok := ring.latest(); ok {
+			ring.push(last)
+			continue
+		}
+		ring.push(0)
+	}
+}
+
 func (h *handler) pushNamedMetricRing(dst *[]*namedMetricsRing, name string, value float64) {
 	if name == "" {
 		return
@@ -847,7 +888,10 @@ func (h *handler) applyPendingNetworkChange(apply func() (app.ActionResult, erro
 		return result, err
 	}
 
-	pnc := &pendingNetChange{snapshot: snapshot}
+	pnc := &pendingNetChange{
+		snapshot: snapshot,
+		deadline: time.Now().Add(netRollbackTimeout),
+	}
 	pnc.timer = time.AfterFunc(netRollbackTimeout, func() {
 		_ = h.opts.App.RestoreNetworkSnapshot(snapshot)
 		h.pendingNetMu.Lock()
@@ -864,6 +908,25 @@ func (h *handler) applyPendingNetworkChange(apply func() (app.ActionResult, erro
 	return result, nil
 }
 
+func (h *handler) hasPendingNetworkChange() bool {
+	h.pendingNetMu.Lock()
+	defer h.pendingNetMu.Unlock()
+	return h.pendingNet != nil
+}
+
+func (h *handler) pendingNetworkRollbackIn() int {
+	h.pendingNetMu.Lock()
+	defer h.pendingNetMu.Unlock()
+	if h.pendingNet == nil {
+		return 0
+	}
+	remaining := int(time.Until(h.pendingNet.deadline).Seconds())
+	if remaining < 1 {
+		return 1
+	}
+	return remaining
+}
+
 func (h *handler) handleAPINetworkConfirm(w http.ResponseWriter, _ *http.Request) {
 	h.pendingNetMu.Lock()
 	pnc := h.pendingNet

audit/internal/webui/api_test.go

@@ -7,6 +7,7 @@ import (
 	"testing"
 
 	"bee/audit/internal/app"
+	"bee/audit/internal/platform"
 )
 
 func TestXrandrCommandAddsDefaultX11Env(t *testing.T) {
@@ -100,3 +101,29 @@ func TestHandleAPIExportBundleQueuesTask(t *testing.T) {
 		t.Fatalf("target=%q want support-bundle", got)
 	}
 }
+
+func TestPushFanRingsTracksByNameAndCarriesForwardMissingSamples(t *testing.T) {
+	h := &handler{}
+	h.pushFanRings([]platform.FanReading{
+		{Name: "FAN_A", RPM: 4200},
+		{Name: "FAN_B", RPM: 5100},
+	})
+	h.pushFanRings([]platform.FanReading{
+		{Name: "FAN_B", RPM: 5200},
+	})
+
+	if len(h.fanNames) != 2 || h.fanNames[0] != "FAN_A" || h.fanNames[1] != "FAN_B" {
+		t.Fatalf("fanNames=%v", h.fanNames)
+	}
+	aVals, _ := h.ringFans[0].snapshot()
+	bVals, _ := h.ringFans[1].snapshot()
+	if len(aVals) != 2 || len(bVals) != 2 {
+		t.Fatalf("fan ring lengths: A=%d B=%d", len(aVals), len(bVals))
+	}
+	if aVals[1] != 4200 {
+		t.Fatalf("FAN_A should carry forward last value, got %v", aVals)
+	}
+	if bVals[1] != 5200 {
+		t.Fatalf("FAN_B should use latest sampled value, got %v", bVals)
+	}
+}

audit/internal/webui/metricsdb.go

@@ -120,7 +120,7 @@ func (m *MetricsDB) Write(s platform.LiveMetricSample) error {
 
 // LoadRecent returns up to n samples in chronological order (oldest first).
 func (m *MetricsDB) LoadRecent(n int) ([]platform.LiveMetricSample, error) {
-	return m.loadSamples(`SELECT ts,cpu_load_pct,mem_load_pct,power_w FROM sys_metrics ORDER BY ts DESC LIMIT ?`, n)
+	return m.loadSamples(`SELECT ts,cpu_load_pct,mem_load_pct,power_w FROM (SELECT ts,cpu_load_pct,mem_load_pct,power_w FROM sys_metrics ORDER BY ts DESC LIMIT ?) ORDER BY ts`, n)
 }
 
 // LoadAll returns all persisted samples in chronological order (oldest first).
@@ -151,11 +151,6 @@ func (m *MetricsDB) loadSamples(query string, args ...any) ([]platform.LiveMetri
 	if len(sysRows) == 0 {
 		return nil, nil
 	}
-	// Reverse to chronological order
-	for i, j := 0, len(sysRows)-1; i < j; i, j = i+1, j-1 {
-		sysRows[i], sysRows[j] = sysRows[j], sysRows[i]
-	}
-
 	// Collect min/max ts for range query
 	minTS := sysRows[0].ts
 	maxTS := sysRows[len(sysRows)-1].ts

audit/internal/webui/metricsdb_test.go

@@ -0,0 +1,69 @@
+package webui
+
+import (
+	"path/filepath"
+	"testing"
+	"time"
+
+	"bee/audit/internal/platform"
+)
+
+func TestMetricsDBLoadSamplesKeepsChronologicalRangeForGPUs(t *testing.T) {
+	db, err := openMetricsDB(filepath.Join(t.TempDir(), "metrics.db"))
+	if err != nil {
+		t.Fatalf("openMetricsDB: %v", err)
+	}
+	defer db.Close()
+
+	base := time.Unix(1_700_000_000, 0).UTC()
+	for i := 0; i < 3; i++ {
+		err := db.Write(platform.LiveMetricSample{
+			Timestamp:  base.Add(time.Duration(i) * time.Second),
+			CPULoadPct: float64(10 + i),
+			MemLoadPct: float64(20 + i),
+			PowerW:     float64(300 + i),
+			GPUs: []platform.GPUMetricRow{
+				{GPUIndex: 0, PowerW: float64(100 + i)},
+				{GPUIndex: 2, PowerW: float64(200 + i)},
+			},
+		})
+		if err != nil {
+			t.Fatalf("Write(%d): %v", i, err)
+		}
+	}
+
+	all, err := db.LoadAll()
+	if err != nil {
+		t.Fatalf("LoadAll: %v", err)
+	}
+	if len(all) != 3 {
+		t.Fatalf("LoadAll len=%d want 3", len(all))
+	}
+	for i, sample := range all {
+		if len(sample.GPUs) != 2 {
+			t.Fatalf("LoadAll sample %d GPUs=%v want 2 rows", i, sample.GPUs)
+		}
+		if sample.GPUs[0].GPUIndex != 0 || sample.GPUs[0].PowerW != float64(100+i) {
+			t.Fatalf("LoadAll sample %d GPU0=%+v", i, sample.GPUs[0])
+		}
+		if sample.GPUs[1].GPUIndex != 2 || sample.GPUs[1].PowerW != float64(200+i) {
+			t.Fatalf("LoadAll sample %d GPU1=%+v", i, sample.GPUs[1])
+		}
+	}
+
+	recent, err := db.LoadRecent(2)
+	if err != nil {
+		t.Fatalf("LoadRecent: %v", err)
+	}
+	if len(recent) != 2 {
+		t.Fatalf("LoadRecent len=%d want 2", len(recent))
+	}
+	if !recent[0].Timestamp.Before(recent[1].Timestamp) {
+		t.Fatalf("LoadRecent timestamps not ascending: %v >= %v", recent[0].Timestamp, recent[1].Timestamp)
+	}
+	for i, sample := range recent {
+		if len(sample.GPUs) != 2 {
+			t.Fatalf("LoadRecent sample %d GPUs=%v want 2 rows", i, sample.GPUs)
+		}
+	}
+}

audit/internal/webui/pages.go

@@ -522,13 +522,30 @@ func renderMetrics() string {
 </div>
 
 <script>
+const chartIds = [
+  'chart-server-load','chart-server-temp-cpu','chart-server-temp-gpu','chart-server-temp-ambient','chart-server-power','chart-server-fans',
+  'chart-gpu-all-load','chart-gpu-all-memload','chart-gpu-all-power','chart-gpu-all-temp'
+];
+
+function refreshChartImage(el) {
+  if (!el || el.dataset.loading === '1') return;
+  const baseSrc = el.dataset.baseSrc || el.src.split('?')[0];
+  const nextSrc = baseSrc + '?t=' + Date.now();
+  const probe = new Image();
+  el.dataset.baseSrc = baseSrc;
+  el.dataset.loading = '1';
+  probe.onload = function() {
+    el.src = nextSrc;
+    el.dataset.loading = '0';
+  };
+  probe.onerror = function() {
+    el.dataset.loading = '0';
+  };
+  probe.src = nextSrc;
+}
+
 function refreshCharts() {
-  const t = '?t=' + Date.now();
+  chartIds.forEach(id => refreshChartImage(document.getElementById(id)));
-  ['chart-server-load','chart-server-temp-cpu','chart-server-temp-gpu','chart-server-temp-ambient','chart-server-power','chart-server-fans',
-   'chart-gpu-all-load','chart-gpu-all-memload','chart-gpu-all-power','chart-gpu-all-temp'].forEach(id => {
-    const el = document.getElementById(id);
-    if (el) el.src = el.src.split('?')[0] + t;
-  });
 }
 setInterval(refreshCharts, 3000);
 
@@ -892,6 +909,8 @@ func renderNetworkInline() string {
 </div>
 <script>
 var _netCountdownTimer = null;
+var _netRefreshTimer = null;
+const NET_ROLLBACK_SECS = 60;
 function loadNetwork() {
   fetch('/api/network').then(r=>r.json()).then(d => {
     const rows = (d.interfaces||[]).map(i =>
@@ -902,21 +921,33 @@ function loadNetwork() {
     document.getElementById('iface-table').innerHTML =
       '<table><tr><th>Interface</th><th>State (click to toggle)</th><th>Addresses</th></tr>'+rows+'</table>' +
       (d.default_route ? '<p style="font-size:12px;color:var(--muted);margin-top:8px">Default route: '+d.default_route+'</p>' : '');
-  });
+    if (d.pending_change) showNetPending(d.rollback_in || NET_ROLLBACK_SECS);
+    else hideNetPending();
+  }).catch(function() {});
 }
 function selectIface(iface) {
   document.getElementById('dhcp-iface').value = iface;
   document.getElementById('st-iface').value = iface;
 }
 function toggleIface(iface, currentState) {
+  showNetPending(NET_ROLLBACK_SECS);
   fetch('/api/network/toggle',{method:'POST',headers:{'Content-Type':'application/json'},body:JSON.stringify({iface:iface})})
     .then(r=>r.json()).then(d => {
-      if (d.error) { alert('Error: '+d.error); return; }
+      if (d.error) { hideNetPending(); alert('Error: '+d.error); return; }
       loadNetwork();
-      showNetPending(d.rollback_in || 60);
+      showNetPending(d.rollback_in || NET_ROLLBACK_SECS);
+    }).catch(function() {
+      setTimeout(loadNetwork, 1500);
     });
 }
+function hideNetPending() {
+  const el = document.getElementById('net-pending');
+  if (_netCountdownTimer) clearInterval(_netCountdownTimer);
+  _netCountdownTimer = null;
+  el.style.display = 'none';
+}
 function showNetPending(secs) {
+  if (!secs || secs < 1) { hideNetPending(); return; }
   const el = document.getElementById('net-pending');
   el.style.display = 'block';
   if (_netCountdownTimer) clearInterval(_netCountdownTimer);
@@ -925,30 +956,33 @@ function showNetPending(secs) {
   _netCountdownTimer = setInterval(function() {
     remaining--;
     document.getElementById('net-countdown').textContent = remaining;
-    if (remaining <= 0) { clearInterval(_netCountdownTimer); _netCountdownTimer=null; el.style.display='none'; loadNetwork(); }
+    if (remaining <= 0) { hideNetPending(); loadNetwork(); }
   }, 1000);
 }
 function confirmNetChange() {
-  if (_netCountdownTimer) { clearInterval(_netCountdownTimer); _netCountdownTimer=null; }
+  hideNetPending();
-  document.getElementById('net-pending').style.display='none';
+  fetch('/api/network/confirm',{method:'POST'}).then(()=>loadNetwork()).catch(()=>{});
-  fetch('/api/network/confirm',{method:'POST'});
 }
 function rollbackNetChange() {
-  if (_netCountdownTimer) { clearInterval(_netCountdownTimer); _netCountdownTimer=null; }
+  hideNetPending();
-  document.getElementById('net-pending').style.display='none';
+  fetch('/api/network/rollback',{method:'POST'}).then(()=>loadNetwork()).catch(()=>{});
-  fetch('/api/network/rollback',{method:'POST'}).then(()=>loadNetwork());
 }
 function runDHCP() {
   const iface = document.getElementById('dhcp-iface').value.trim();
+  showNetPending(NET_ROLLBACK_SECS);
   fetch('/api/network/dhcp',{method:'POST',headers:{'Content-Type':'application/json'},body:JSON.stringify({interface:iface||'all'})})
     .then(r=>r.json()).then(d => {
       document.getElementById('dhcp-out').textContent = d.output || d.error || 'Done.';
-      if (!d.error) showNetPending(d.rollback_in || 60);
+      if (d.error) { hideNetPending(); return; }
+      showNetPending(d.rollback_in || NET_ROLLBACK_SECS);
       loadNetwork();
+    }).catch(function() {
+      setTimeout(loadNetwork, 1500);
     });
 }
 function setStatic() {
   const dns = document.getElementById('st-dns').value.split(',').map(s=>s.trim()).filter(Boolean);
+  showNetPending(NET_ROLLBACK_SECS);
   fetch('/api/network/static',{method:'POST',headers:{'Content-Type':'application/json'},body:JSON.stringify({
     interface: document.getElementById('st-iface').value,
     address: document.getElementById('st-addr').value,
@@ -957,11 +991,16 @@ function setStatic() {
     dns: dns,
   })}).then(r=>r.json()).then(d => {
     document.getElementById('static-out').textContent = d.output || d.error || 'Done.';
-    if (!d.error) showNetPending(d.rollback_in || 60);
+    if (d.error) { hideNetPending(); return; }
+    showNetPending(d.rollback_in || NET_ROLLBACK_SECS);
     loadNetwork();
+  }).catch(function() {
+    setTimeout(loadNetwork, 1500);
   });
 }
 loadNetwork();
+if (_netRefreshTimer) clearInterval(_netRefreshTimer);
+_netRefreshTimer = setInterval(loadNetwork, 5000);
 </script>`
 }
 
@@ -1562,7 +1601,7 @@ function loadTasks() {
       return;
     }
     const rows = tasks.map(t => {
-      const dur = t.started_at ? formatDur(t.started_at, t.done_at) : '';
+      const dur = t.elapsed_sec ? formatDurSec(t.elapsed_sec) : '';
       const statusClass = {running:'badge-ok',pending:'badge-unknown',done:'badge-ok',failed:'badge-err',cancelled:'badge-unknown'}[t.status]||'badge-unknown';
       const statusLabel = {running:'&#9654; running',pending:'pending',done:'&#10003; done',failed:'&#10007; failed',cancelled:'cancelled'}[t.status]||t.status;
       let actions = '<button class="btn btn-sm btn-secondary" onclick="viewLog(\''+t.id+'\',\''+escHtml(t.name)+'\')">Logs</button>';
@@ -1587,14 +1626,11 @@ function loadTasks() {
 
 function escHtml(s) { return (s||'').replace(/&/g,'&amp;').replace(/</g,'&lt;').replace(/>/g,'&gt;').replace(/"/g,'&quot;'); }
 function fmtTime(s) { if (!s) return ''; try { return new Date(s).toLocaleTimeString(); } catch(e){ return s; } }
-function formatDur(start, end) {
+function formatDurSec(sec) {
-  try {
+  sec = Math.max(0, Math.round(sec||0));
-    const s = new Date(start), e = end ? new Date(end) : new Date();
+  if (sec < 60) return sec+'s';
-    const sec = Math.round((e-s)/1000);
+  const m = Math.floor(sec/60), ss = sec%60;
-    if (sec < 60) return sec+'s';
+  return m+'m '+ss+'s';
-    const m = Math.floor(sec/60), ss = sec%60;
-    return m+'m '+ss+'s';
-  } catch(e){ return ''; }
 }
 
 function cancelTask(id) {

audit/internal/webui/server.go

@@ -10,6 +10,7 @@ import (
 	"net/http"
 	"os"
 	"path/filepath"
+	"sort"
 	"strings"
 	"sync"
 	"time"
@@ -84,6 +85,15 @@ func (r *metricsRing) snapshot() ([]float64, []string) {
 	return v, labels
 }
 
+func (r *metricsRing) latest() (float64, bool) {
+	r.mu.Lock()
+	defer r.mu.Unlock()
+	if len(r.vals) == 0 {
+		return 0, false
+	}
+	return r.vals[len(r.vals)-1], true
+}
+
 func timestampsSameLocalDay(times []time.Time) bool {
 	if len(times) == 0 {
 		return true
@@ -118,9 +128,12 @@ type namedMetricsRing struct {
 	Ring *metricsRing
 }
 
+const metricsChartWindow = 120
+
 // pendingNetChange tracks a network state change awaiting confirmation.
 type pendingNetChange struct {
 	snapshot platform.NetworkSnapshot
+	deadline time.Time
 	timer    *time.Timer
 	mu       sync.Mutex
 }
@@ -171,7 +184,7 @@ func NewHandler(opts HandlerOptions) http.Handler {
 	// Open metrics DB and pre-fill ring buffers from history.
 	if db, err := openMetricsDB(metricsDBPath); err == nil {
 		h.metricsDB = db
-		if samples, err := db.LoadRecent(120); err == nil {
+		if samples, err := db.LoadRecent(metricsChartWindow); err == nil {
 			for _, s := range samples {
 				h.feedRings(s)
 			}
@@ -292,11 +305,11 @@ func (h *handler) startMetricsCollector() {
 		defer ticker.Stop()
 		for range ticker.C {
 			sample := platform.SampleLiveMetrics()
-			h.feedRings(sample)
-			h.setLatestMetric(sample)
 			if h.metricsDB != nil {
 				_ = h.metricsDB.Write(sample)
 			}
+			h.feedRings(sample)
+			h.setLatestMetric(sample)
 		}
 	}()
 }
@@ -448,222 +461,13 @@ func (h *handler) handleMetricsChartSVG(w http.ResponseWriter, r *http.Request)
 	path := strings.TrimPrefix(r.URL.Path, "/api/metrics/chart/")
 	path = strings.TrimSuffix(path, ".svg")
 
-	if h.metricsDB != nil {
+	if h.metricsDB == nil {
-		if datasets, names, labels, title, yMin, yMax, ok := h.chartDataFromDB(path); ok {
+		http.Error(w, "metrics database not available", http.StatusServiceUnavailable)
-			buf, err := renderChartSVG(title, datasets, names, labels, yMin, yMax)
+		return
-			if err != nil {
-				http.Error(w, err.Error(), http.StatusInternalServerError)
-				return
-			}
-			w.Header().Set("Content-Type", "image/svg+xml")
-			w.Header().Set("Cache-Control", "no-store")
-			_, _ = w.Write(buf)
-			return
-		}
 	}
-
+	datasets, names, labels, title, yMin, yMax, ok := h.chartDataFromDB(path)
-	var datasets [][]float64
+	if !ok {
-	var names []string
+		http.Error(w, "metrics history unavailable", http.StatusServiceUnavailable)
-	var labels []string
-	var title string
-	var yMin, yMax *float64 // nil = auto; for load charts fixed 0-100
-
-	switch {
-	// ── Server sub-charts ─────────────────────────────────────────────────
-	case path == "server-load":
-		title = "CPU / Memory Load"
-		vCPULoad, l := h.ringCPULoad.snapshot()
-		vMemLoad, _ := h.ringMemLoad.snapshot()
-		labels = l
-		datasets = [][]float64{vCPULoad, vMemLoad}
-		names = []string{"CPU Load %", "Mem Load %"}
-		yMin = floatPtr(0)
-		yMax = floatPtr(100)
-
-	case path == "server-temp", path == "server-temp-cpu":
-		title = "CPU Temperature"
-		h.ringsMu.Lock()
-		datasets, names, labels = snapshotNamedRings(h.cpuTempRings)
-		h.ringsMu.Unlock()
-		yMin = floatPtr(0)
-		yMax = autoMax120(datasets...)
-
-	case path == "server-temp-gpu":
-		title = "GPU Temperature"
-		h.ringsMu.Lock()
-		for idx, gr := range h.gpuRings {
-			if gr == nil {
-				continue
-			}
-			vTemp, l := gr.Temp.snapshot()
-			datasets = append(datasets, vTemp)
-			names = append(names, fmt.Sprintf("GPU %d", idx))
-			if len(labels) == 0 {
-				labels = l
-			}
-		}
-		h.ringsMu.Unlock()
-		yMin = floatPtr(0)
-		yMax = autoMax120(datasets...)
-
-	case path == "server-temp-ambient":
-		title = "Ambient / Other Sensors"
-		h.ringsMu.Lock()
-		datasets, names, labels = snapshotNamedRings(h.ambientTempRings)
-		h.ringsMu.Unlock()
-		yMin = floatPtr(0)
-		yMax = autoMax120(datasets...)
-
-	case path == "server-power":
-		title = "System Power"
-		vPower, l := h.ringPower.snapshot()
-		vPower = normalizePowerSeries(vPower)
-		labels = l
-		datasets = [][]float64{vPower}
-		names = []string{"Power W"}
-		yMin = floatPtr(0)
-		yMax = autoMax120(vPower)
-
-	case path == "server-fans":
-		title = "Fan RPM"
-		h.ringsMu.Lock()
-		for i, fr := range h.ringFans {
-			fv, _ := fr.snapshot()
-			datasets = append(datasets, fv)
-			name := "Fan"
-			if i < len(h.fanNames) {
-				name = h.fanNames[i]
-			}
-			names = append(names, name+" RPM")
-		}
-		h.ringsMu.Unlock()
-		yMin = floatPtr(0)
-		yMax = autoMax120(datasets...)
-
-	// ── Combined GPU charts (all GPUs on one chart) ───────────────────────
-	case path == "gpu-all-load":
-		title = "GPU Compute Load"
-		h.ringsMu.Lock()
-		for idx, gr := range h.gpuRings {
-			if gr == nil {
-				continue
-			}
-			vUtil, l := gr.Util.snapshot()
-			datasets = append(datasets, vUtil)
-			names = append(names, fmt.Sprintf("GPU %d", idx))
-			if len(labels) == 0 {
-				labels = l
-			}
-		}
-		h.ringsMu.Unlock()
-		yMin = floatPtr(0)
-		yMax = floatPtr(100)
-
-	case path == "gpu-all-memload":
-		title = "GPU Memory Load"
-		h.ringsMu.Lock()
-		for idx, gr := range h.gpuRings {
-			if gr == nil {
-				continue
-			}
-			vMem, l := gr.MemUtil.snapshot()
-			datasets = append(datasets, vMem)
-			names = append(names, fmt.Sprintf("GPU %d", idx))
-			if len(labels) == 0 {
-				labels = l
-			}
-		}
-		h.ringsMu.Unlock()
-		yMin = floatPtr(0)
-		yMax = floatPtr(100)
-
-	case path == "gpu-all-power":
-		title = "GPU Power"
-		h.ringsMu.Lock()
-		for idx, gr := range h.gpuRings {
-			if gr == nil {
-				continue
-			}
-			vPow, l := gr.Power.snapshot()
-			datasets = append(datasets, vPow)
-			names = append(names, fmt.Sprintf("GPU %d", idx))
-			if len(labels) == 0 {
-				labels = l
-			}
-		}
-		h.ringsMu.Unlock()
-		yMin = floatPtr(0)
-		yMax = autoMax120(datasets...)
-
-	case path == "gpu-all-temp":
-		title = "GPU Temperature"
-		h.ringsMu.Lock()
-		for idx, gr := range h.gpuRings {
-			if gr == nil {
-				continue
-			}
-			vTemp, l := gr.Temp.snapshot()
-			datasets = append(datasets, vTemp)
-			names = append(names, fmt.Sprintf("GPU %d", idx))
-			if len(labels) == 0 {
-				labels = l
-			}
-		}
-		h.ringsMu.Unlock()
-		yMin = floatPtr(0)
-		yMax = autoMax120(datasets...)
-
-	// ── Per-GPU sub-charts ────────────────────────────────────────────────
-	case strings.HasPrefix(path, "gpu/"):
-		rest := strings.TrimPrefix(path, "gpu/")
-		// rest is either "{idx}-load", "{idx}-temp", "{idx}-power", or legacy "{idx}"
-		sub := ""
-		if i := strings.LastIndex(rest, "-"); i > 0 {
-			sub = rest[i+1:]
-			rest = rest[:i]
-		}
-		idx := 0
-		fmt.Sscanf(rest, "%d", &idx)
-		h.ringsMu.Lock()
-		var gr *gpuRings
-		if idx < len(h.gpuRings) {
-			gr = h.gpuRings[idx]
-		}
-		h.ringsMu.Unlock()
-		if gr == nil {
-			http.NotFound(w, r)
-			return
-		}
-		switch sub {
-		case "load":
-			vUtil, l := gr.Util.snapshot()
-			vMemUtil, _ := gr.MemUtil.snapshot()
-			labels = l
-			title = fmt.Sprintf("GPU %d Load", idx)
-			datasets = [][]float64{vUtil, vMemUtil}
-			names = []string{"Load %", "Mem %"}
-			yMin = floatPtr(0)
-			yMax = floatPtr(100)
-		case "temp":
-			vTemp, l := gr.Temp.snapshot()
-			labels = l
-			title = fmt.Sprintf("GPU %d Temperature", idx)
-			datasets = [][]float64{vTemp}
-			names = []string{"Temp °C"}
-			yMin = floatPtr(0)
-			yMax = autoMax120(vTemp)
-		default: // "power" or legacy (no sub)
-			vPower, l := gr.Power.snapshot()
-			labels = l
-			title = fmt.Sprintf("GPU %d Power", idx)
-			datasets = [][]float64{vPower}
-			names = []string{"Power W"}
-			yMin = floatPtr(0)
-			yMax = autoMax120(vPower)
-		}
-
-	default:
-		http.NotFound(w, r)
 		return
 	}
 
@@ -840,6 +644,7 @@ func namedTempDatasets(samples []platform.LiveMetricSample, group string) ([][]f
 			}
 		}
 	}
+	sort.Strings(names)
 	datasets := make([][]float64, 0, len(names))
 	for _, name := range names {
 		ds := make([]float64, len(samples))
@@ -867,6 +672,7 @@ func namedFanDatasets(samples []platform.LiveMetricSample) ([][]float64, []strin
 			}
 		}
 	}
+	sort.Strings(names)
 	datasets := make([][]float64, 0, len(names))
 	for _, name := range names {
 		ds := make([]float64, len(samples))
@@ -878,7 +684,7 @@ func namedFanDatasets(samples []platform.LiveMetricSample) ([][]float64, []strin
 				}
 			}
 		}
-		datasets = append(datasets, ds)
+		datasets = append(datasets, normalizeFanSeries(ds))
 	}
 	return datasets, names
 }
@@ -894,6 +700,7 @@ func gpuDatasets(samples []platform.LiveMetricSample, pick func(platform.GPUMetr
 			}
 		}
 	}
+	sort.Ints(indices)
 	datasets := make([][]float64, 0, len(indices))
 	names := make([]string, 0, len(indices))
 	for _, idx := range indices {
@@ -953,6 +760,27 @@ func normalizePowerSeries(ds []float64) []float64 {
 	return out
 }
 
+func normalizeFanSeries(ds []float64) []float64 {
+	if len(ds) == 0 {
+		return nil
+	}
+	out := make([]float64, len(ds))
+	var lastPositive float64
+	for i, v := range ds {
+		if v > 0 {
+			lastPositive = v
+			out[i] = v
+			continue
+		}
+		if lastPositive > 0 {
+			out[i] = lastPositive
+			continue
+		}
+		out[i] = 0
+	}
+	return out
+}
+
 // floatPtr returns a pointer to a float64 value.
 func floatPtr(v float64) *float64 { return &v }
 
@@ -1044,15 +872,17 @@ func renderChartSVG(title string, datasets [][]float64, names []string, labels [
 	opt.Title = gocharts.TitleOption{Text: title}
 	opt.XAxis.Labels = sparse
 	opt.Legend = gocharts.LegendOption{SeriesNames: names}
+	if chartLegendVisible(len(names)) {
+		opt.Legend.Offset = gocharts.OffsetStr{Top: gocharts.PositionBottom}
+		opt.Legend.OverlayChart = gocharts.Ptr(false)
+	} else {
+		opt.Legend.Show = gocharts.Ptr(false)
+	}
 	opt.Symbol = gocharts.SymbolNone
 	// Right padding: reserve space for the MarkLine label (library recommendation).
 	opt.Padding = gocharts.NewBox(20, 20, 80, 20)
 	if yMin != nil || yMax != nil {
-		opt.YAxis = []gocharts.YAxisOption{{
+		opt.YAxis = []gocharts.YAxisOption{chartYAxisOption(yMin, yMax)}
-			Min:            yMin,
-			Max:            yMax,
-			ValueFormatter: chartLegendNumber,
-		}}
 	}
 
 	// Add a single peak mark line on the series that holds the global maximum.
@@ -1064,7 +894,7 @@ func renderChartSVG(title string, datasets [][]float64, names []string, labels [
 	p := gocharts.NewPainter(gocharts.PainterOptions{
 		OutputFormat: gocharts.ChartOutputSVG,
 		Width:        1400,
-		Height:       240,
+		Height:       chartCanvasHeight(len(names)),
 	}, gocharts.PainterThemeOption(gocharts.GetTheme("grafana")))
 	if err := p.LineChart(opt); err != nil {
 		return nil, err
@@ -1072,6 +902,26 @@ func renderChartSVG(title string, datasets [][]float64, names []string, labels [
 	return p.Bytes()
 }
 
+func chartLegendVisible(seriesCount int) bool {
+	return seriesCount <= 8
+}
+
+func chartCanvasHeight(seriesCount int) int {
+	if chartLegendVisible(seriesCount) {
+		return 360
+	}
+	return 288
+}
+
+func chartYAxisOption(yMin, yMax *float64) gocharts.YAxisOption {
+	return gocharts.YAxisOption{
+		Min:            yMin,
+		Max:            yMax,
+		LabelCount:     11,
+		ValueFormatter: chartYAxisNumber,
+	}
+}
+
 // globalPeakSeries returns the index of the series containing the global maximum
 // value across all datasets, and that maximum value.
 func globalPeakSeries(datasets [][]float64) (idx int, peak float64) {
@@ -1159,6 +1009,28 @@ func snapshotNamedRings(rings []*namedMetricsRing) ([][]float64, []string, []str
 	return datasets, names, labels
 }
 
+func snapshotFanRings(rings []*metricsRing, fanNames []string) ([][]float64, []string, []string) {
+	var datasets [][]float64
+	var names []string
+	var labels []string
+	for i, ring := range rings {
+		if ring == nil {
+			continue
+		}
+		vals, l := ring.snapshot()
+		datasets = append(datasets, normalizeFanSeries(vals))
+		name := "Fan"
+		if i < len(fanNames) {
+			name = fanNames[i]
+		}
+		names = append(names, name+" RPM")
+		if len(labels) == 0 {
+			labels = l
+		}
+	}
+	return datasets, names, labels
+}
+
 func chartLegendNumber(v float64) string {
 	neg := v < 0
 	if v < 0 {
@@ -1181,6 +1053,23 @@ func chartLegendNumber(v float64) string {
 	return out
 }
 
+func chartYAxisNumber(v float64) string {
+	neg := v < 0
+	if neg {
+		v = -v
+	}
+	var out string
+	if v >= 1000 {
+		out = fmt.Sprintf("%dк", int((v+500)/1000))
+	} else {
+		out = fmt.Sprintf("%.0f", v)
+	}
+	if neg {
+		return "-" + out
+	}
+	return out
+}
+
 func sparseLabels(labels []string, n int) []string {
 	out := make([]string, len(labels))
 	step := len(labels) / n

audit/internal/webui/server_test.go

@@ -89,6 +89,53 @@ func TestChartDataFromSamplesUsesFullHistory(t *testing.T) {
 	}
 }
 
+func TestChartDataFromSamplesKeepsStableGPUSeriesOrder(t *testing.T) {
+	samples := []platform.LiveMetricSample{
+		{
+			Timestamp: time.Now().Add(-2 * time.Minute),
+			GPUs: []platform.GPUMetricRow{
+				{GPUIndex: 7, PowerW: 170},
+				{GPUIndex: 2, PowerW: 120},
+				{GPUIndex: 0, PowerW: 100},
+			},
+		},
+		{
+			Timestamp: time.Now().Add(-1 * time.Minute),
+			GPUs: []platform.GPUMetricRow{
+				{GPUIndex: 0, PowerW: 101},
+				{GPUIndex: 7, PowerW: 171},
+				{GPUIndex: 2, PowerW: 121},
+			},
+		},
+	}
+
+	datasets, names, _, title, _, _, ok := chartDataFromSamples("gpu-all-power", samples)
+	if !ok {
+		t.Fatal("chartDataFromSamples returned ok=false")
+	}
+	if title != "GPU Power" {
+		t.Fatalf("title=%q", title)
+	}
+	wantNames := []string{"GPU 0", "GPU 2", "GPU 7"}
+	if len(names) != len(wantNames) {
+		t.Fatalf("names len=%d want %d: %v", len(names), len(wantNames), names)
+	}
+	for i := range wantNames {
+		if names[i] != wantNames[i] {
+			t.Fatalf("names[%d]=%q want %q; full=%v", i, names[i], wantNames[i], names)
+		}
+	}
+	if got := datasets[0]; len(got) != 2 || got[0] != 100 || got[1] != 101 {
+		t.Fatalf("GPU 0 dataset=%v want [100 101]", got)
+	}
+	if got := datasets[1]; len(got) != 2 || got[0] != 120 || got[1] != 121 {
+		t.Fatalf("GPU 2 dataset=%v want [120 121]", got)
+	}
+	if got := datasets[2]; len(got) != 2 || got[0] != 170 || got[1] != 171 {
+		t.Fatalf("GPU 7 dataset=%v want [170 171]", got)
+	}
+}
+
 func TestNormalizePowerSeriesHoldsLastPositive(t *testing.T) {
 	got := normalizePowerSeries([]float64{0, 480, 0, 0, 510, 0})
 	want := []float64{0, 480, 480, 480, 510, 510}
@@ -102,6 +149,117 @@ func TestNormalizePowerSeriesHoldsLastPositive(t *testing.T) {
 	}
 }
 
+func TestRenderMetricsUsesBufferedChartRefresh(t *testing.T) {
+	body := renderMetrics()
+	if !strings.Contains(body, "const probe = new Image();") {
+		t.Fatalf("metrics page should preload chart images before swap: %s", body)
+	}
+	if !strings.Contains(body, "el.dataset.loading === '1'") {
+		t.Fatalf("metrics page should avoid overlapping chart reloads: %s", body)
+	}
+}
+
+func TestChartLegendVisible(t *testing.T) {
+	if !chartLegendVisible(8) {
+		t.Fatal("legend should stay visible for charts with up to 8 series")
+	}
+	if chartLegendVisible(9) {
+		t.Fatal("legend should be hidden for charts with more than 8 series")
+	}
+}
+
+func TestChartYAxisNumber(t *testing.T) {
+	tests := []struct {
+		in   float64
+		want string
+	}{
+		{in: 999, want: "999"},
+		{in: 1000, want: "1к"},
+		{in: 1370, want: "1к"},
+		{in: 1500, want: "2к"},
+		{in: 10200, want: "10к"},
+		{in: -1499, want: "-1к"},
+	}
+	for _, tc := range tests {
+		if got := chartYAxisNumber(tc.in); got != tc.want {
+			t.Fatalf("chartYAxisNumber(%v)=%q want %q", tc.in, got, tc.want)
+		}
+	}
+}
+
+func TestChartCanvasHeight(t *testing.T) {
+	if got := chartCanvasHeight(4); got != 360 {
+		t.Fatalf("chartCanvasHeight(4)=%d want 360", got)
+	}
+	if got := chartCanvasHeight(12); got != 288 {
+		t.Fatalf("chartCanvasHeight(12)=%d want 288", got)
+	}
+}
+
+func TestNormalizeFanSeriesHoldsLastPositive(t *testing.T) {
+	got := normalizeFanSeries([]float64{4200, 0, 0, 4300, 0})
+	want := []float64{4200, 4200, 4200, 4300, 4300}
+	if len(got) != len(want) {
+		t.Fatalf("len=%d want %d", len(got), len(want))
+	}
+	for i := range want {
+		if got[i] != want[i] {
+			t.Fatalf("got[%d]=%v want %v", i, got[i], want[i])
+		}
+	}
+}
+
+func TestChartYAxisOption(t *testing.T) {
+	min := floatPtr(0)
+	max := floatPtr(100)
+	opt := chartYAxisOption(min, max)
+	if opt.Min != min || opt.Max != max {
+		t.Fatalf("chartYAxisOption min/max mismatch: %#v", opt)
+	}
+	if opt.LabelCount != 11 {
+		t.Fatalf("chartYAxisOption labelCount=%d want 11", opt.LabelCount)
+	}
+	if got := opt.ValueFormatter(1000); got != "1к" {
+		t.Fatalf("chartYAxisOption formatter(1000)=%q want 1к", got)
+	}
+}
+
+func TestSnapshotFanRingsUsesTimelineLabels(t *testing.T) {
+	r1 := newMetricsRing(4)
+	r2 := newMetricsRing(4)
+	r1.push(1000)
+	r1.push(1100)
+	r2.push(1200)
+	r2.push(1300)
+
+	datasets, names, labels := snapshotFanRings([]*metricsRing{r1, r2}, []string{"FAN_A", "FAN_B"})
+	if len(datasets) != 2 {
+		t.Fatalf("datasets=%d want 2", len(datasets))
+	}
+	if len(names) != 2 || names[0] != "FAN_A RPM" || names[1] != "FAN_B RPM" {
+		t.Fatalf("names=%v", names)
+	}
+	if len(labels) != 2 {
+		t.Fatalf("labels=%v want 2 entries", labels)
+	}
+	if labels[0] == "" || labels[1] == "" {
+		t.Fatalf("labels should contain timeline values, got %v", labels)
+	}
+}
+
+func TestRenderNetworkInlineSyncsPendingState(t *testing.T) {
+	body := renderNetworkInline()
+	if !strings.Contains(body, "d.pending_change") {
+		t.Fatalf("network UI should read pending network state from API: %s", body)
+	}
+	if !strings.Contains(body, "setInterval(loadNetwork, 5000)") {
+		t.Fatalf("network UI should periodically refresh network state: %s", body)
+	}
+	if !strings.Contains(body, "showNetPending(NET_ROLLBACK_SECS)") {
+		t.Fatalf("network UI should show pending confirmation immediately on apply: %s", body)
+	}
+}
+
 func TestRootRendersDashboard(t *testing.T) {
 	dir := t.TempDir()
 	path := filepath.Join(dir, "audit.json")

audit/internal/webui/tasks.go

@@ -83,16 +83,17 @@ func taskDisplayName(target, profile, loader string) string {
 
 // Task represents one unit of work in the queue.
 type Task struct {
-	ID        string     `json:"id"`
+	ID         string     `json:"id"`
-	Name      string     `json:"name"`
+	Name       string     `json:"name"`
-	Target    string     `json:"target"`
+	Target     string     `json:"target"`
-	Priority  int        `json:"priority"`
+	Priority   int        `json:"priority"`
-	Status    string     `json:"status"`
+	Status     string     `json:"status"`
-	CreatedAt time.Time  `json:"created_at"`
+	CreatedAt  time.Time  `json:"created_at"`
-	StartedAt *time.Time `json:"started_at,omitempty"`
+	StartedAt  *time.Time `json:"started_at,omitempty"`
-	DoneAt    *time.Time `json:"done_at,omitempty"`
+	DoneAt     *time.Time `json:"done_at,omitempty"`
-	ErrMsg    string     `json:"error,omitempty"`
+	ElapsedSec int        `json:"elapsed_sec,omitempty"`
-	LogPath   string     `json:"log_path,omitempty"`
+	ErrMsg     string     `json:"error,omitempty"`
+	LogPath    string     `json:"log_path,omitempty"`
 
 	// runtime fields (not serialised)
 	job    *jobState
@@ -101,11 +102,11 @@ type Task struct {
 
 // taskParams holds optional parameters parsed from the run request.
 type taskParams struct {
-	Duration          int    `json:"duration,omitempty"`
+	Duration           int      `json:"duration,omitempty"`
-	DiagLevel         int    `json:"diag_level,omitempty"`
+	DiagLevel          int      `json:"diag_level,omitempty"`
-	GPUIndices        []int  `json:"gpu_indices,omitempty"`
+	GPUIndices         []int    `json:"gpu_indices,omitempty"`
-	ExcludeGPUIndices []int  `json:"exclude_gpu_indices,omitempty"`
+	ExcludeGPUIndices  []int    `json:"exclude_gpu_indices,omitempty"`
-	Loader            string `json:"loader,omitempty"`
+	Loader             string   `json:"loader,omitempty"`
 	BurnProfile        string   `json:"burn_profile,omitempty"`
 	DisplayName        string   `json:"display_name,omitempty"`
 	Device             string   `json:"device,omitempty"` // for install
@@ -311,6 +312,7 @@ func (q *taskQueue) snapshot() []Task {
 	out := make([]Task, len(q.tasks))
 	for i, t := range q.tasks {
 		out[i] = *t
+		out[i].ElapsedSec = taskElapsedSec(&out[i], time.Now())
 	}
 	sort.SliceStable(out, func(i, j int) bool {
 		si := statusOrder(out[i].Status)
@@ -769,6 +771,7 @@ func (q *taskQueue) loadLocked() {
 		q.assignTaskLogPathLocked(t)
 		if t.Status == TaskPending || t.Status == TaskRunning {
 			t.Status = TaskPending
+			t.StartedAt = nil
 			t.DoneAt = nil
 			t.ErrMsg = ""
 		}
@@ -808,3 +811,21 @@ func (q *taskQueue) persistLocked() {
 	}
 	_ = os.Rename(tmp, q.statePath)
 }
+
+func taskElapsedSec(t *Task, now time.Time) int {
+	if t == nil || t.StartedAt == nil || t.StartedAt.IsZero() {
+		return 0
+	}
+	start := *t.StartedAt
+	if !t.CreatedAt.IsZero() && start.Before(t.CreatedAt) {
+		start = t.CreatedAt
+	}
+	end := now
+	if t.DoneAt != nil && !t.DoneAt.IsZero() {
+		end = *t.DoneAt
+	}
+	if end.Before(start) {
+		return 0
+	}
+	return int(end.Sub(start).Round(time.Second) / time.Second)
+}

audit/internal/webui/tasks_test.go

@@ -55,6 +55,9 @@ func TestTaskQueuePersistsAndRecoversPendingTasks(t *testing.T) {
 	if got.Status != TaskPending {
 		t.Fatalf("status=%q want %q", got.Status, TaskPending)
 	}
+	if got.StartedAt != nil {
+		t.Fatalf("started_at=%v want nil for recovered pending task", got.StartedAt)
+	}
 	if got.params.Duration != 300 || got.params.BurnProfile != "smoke" {
 		t.Fatalf("params=%+v", got.params)
 	}
@@ -236,6 +239,26 @@ func TestRunTaskBuildsSupportBundleWithoutApp(t *testing.T) {
 	}
 }
 
+func TestTaskElapsedSecClampsInvalidStartedAt(t *testing.T) {
+	now := time.Date(2026, 4, 1, 19, 10, 0, 0, time.UTC)
+	created := time.Date(2026, 4, 1, 19, 4, 5, 0, time.UTC)
+	started := time.Time{}
+	task := &Task{
+		Status:    TaskRunning,
+		CreatedAt: created,
+		StartedAt: &started,
+	}
+	if got := taskElapsedSec(task, now); got != 0 {
+		t.Fatalf("taskElapsedSec(zero start)=%d want 0", got)
+	}
+
+	stale := created.Add(-24 * time.Hour)
+	task.StartedAt = &stale
+	if got := taskElapsedSec(task, now); got != int(now.Sub(created).Seconds()) {
+		t.Fatalf("taskElapsedSec(stale start)=%d want %d", got, int(now.Sub(created).Seconds()))
+	}
+}
+
 func TestRunTaskInstallUsesSharedCommandStreaming(t *testing.T) {
 	q := &taskQueue{
 		opts: &HandlerOptions{},

bible-local/architecture/charting.md

@@ -9,6 +9,34 @@ All live metrics charts in the web UI are server-side SVG images served by Go
 and polled by the browser every 2 seconds via `<img src="...?t=now">`.
 There is no client-side canvas or JS chart library.
 
+## Rule: live charts must be visually uniform
+
+Live charts are a single UI family, not a set of one-off widgets. New charts and
+changes to existing charts must keep the same rendering model and presentation
+rules unless there is an explicit architectural decision to diverge.
+
+Default expectations:
+
+- same server-side SVG pipeline for all live metrics charts
+- same refresh behaviour and failure handling in the browser
+- same canvas size class and card layout
+- same legend placement policy across charts
+- same axis, title, and summary conventions
+- no chart-specific visual exceptions added as a quick fix
+
+Current default for live charts:
+
+- legend below the plot area when a chart has 8 series or fewer
+- legend hidden when a chart has more than 8 series
+- 10 equal Y-axis steps across the chart height
+- 1400 x 360 SVG canvas with legend
+- 1400 x 288 SVG canvas without legend
+- full-width card rendering in a single-column stack
+
+If one chart needs a different layout or legend behaviour, treat that as a
+design-level decision affecting the whole chart family, not as a local tweak to
+just one endpoint.
+
 ### Why go-analyze/charts
 
 - Pure Go, no CGO — builds cleanly inside the live-build container
@@ -29,7 +57,8 @@ self-contained SVG renderer used **only** for completed SAT run reports
 | `GET /api/metrics/chart/server.svg` | CPU temp, CPU load %, mem load %, power W, fan RPMs |
 | `GET /api/metrics/chart/gpu/{idx}.svg` | GPU temp °C, load %, mem %, power W |
 
-Charts are 1400 × 280 px SVG. The page renders them at `width: 100%` in a
+Charts are 1400 × 360 px SVG when the legend is shown, and 1400 × 288 px when
+the legend is hidden. The page renders them at `width: 100%` in a
 single-column layout so they always fill the viewport width.
 
 ### Ring buffers

bible-local/decisions/2026-04-01-memtest-build-strategy.md

@@ -1,7 +1,7 @@
 # Decision: Treat memtest as explicit ISO content, not as trusted live-build magic
 
 **Date:** 2026-04-01
-**Status:** active
+**Status:** resolved
 
 ## Context
 
@@ -58,6 +58,18 @@ Root cause of the false alarm:
 - as a result, we re-entered the same memtest investigation loop even though
   the real ISO was already correct
 
+Additional correction from the subsequent `v3.21` build logs dated 2026-04-01:
+
+- once ISO reading was fixed, the post-build debug correctly showed the raw ISO
+  still carried live-build's default memtest layout (`live/memtest.bin`,
+  `live/memtest.efi`, `boot/grub/memtest.cfg`, `isolinux/memtest.cfg`)
+- that mismatch is expected to trigger project recovery, because `bee` requires
+  `boot/memtest86+x64.bin` / `boot/memtest86+x64.efi` plus matching menu paths
+- however, `build.sh` exited before recovery because `set -e` treated a direct
+  `iso_memtest_present` return code of `1` as fatal
+- so the next repeated loop was caused by shell control flow, not by proof that
+  the recovery design itself was wrong
+
 ## Known Failed Attempts
 
 These approaches were already tried and should not be repeated blindly:
@@ -102,6 +114,8 @@ Any future memtest fix must explicitly identify:
 - and a post-build proof from a real ISO, not only from intermediate workdir files
 - whether the ISO inspection step itself succeeded, rather than merely whether
   the validator printed a memtest warning
+- whether a non-zero probe is intentionally handled inside an `if` / `case`
+  context rather than accidentally tripping `set -e`
 
 ## Decision
 
@@ -134,6 +148,8 @@ Current implementation direction:
   - install a stable ISO reader in the builder image
   - fail with an explicit reader error if ISO listing/extraction fails
   - do not treat reader failure as evidence that memtest is missing
+  - do not call a probe that may return "needs recovery" as a bare command under
+    `set -e`; wrap it in explicit control flow
 
 ## Consequences
 
@@ -144,3 +160,65 @@ Current implementation direction:
 - But validation output is only trustworthy if ISO reading itself succeeded. A
   "missing memtest" warning without a successful ISO read is not evidence.
 - If we change memtest strategy again, we must update this ADR with the exact build evidence that justified the change.
+
+## Working Solution (confirmed 2026-04-01, commits 76a9100 → 2baf3be)
+
+This approach was confirmed working in ISO `easy-bee-nvidia-v3.20-5-g76a9100-amd64.iso`
+and validated again in subsequent builds. The final ISO contains all required memtest artifacts.
+
+### Components
+
+**1. Binary hook `config/hooks/normal/9100-memtest.hook.binary`**
+
+Runs inside the live-build binary phase. Does not patch bootloader files at hook time —
+those files may not exist yet. Instead:
+
+- Tries to copy `memtest86+x64.bin` / `memtest86+x64.efi` from `chroot/boot/` first.
+- Falls back to extracting from the cached `.deb` (via `dpkg-deb -x`) if `chroot/boot/` is empty.
+- Appends GRUB and isolinux menu entries only if the respective cfg files already exist at hook time.
+  If they do not exist, the hook warns and continues (does not fail).
+
+Controlled by `BEE_REQUIRE_MEMTEST=1` env var to turn warnings into hard errors when needed.
+
+**2. Post-`lb build` recovery step in `build.sh`**
+
+After `lb build` completes, `build.sh` checks whether the fully materialized `binary/` tree
+contains all required memtest artifacts. If not:
+
+- Copies/extracts memtest binaries into `binary/boot/`.
+- Patches `binary/boot/grub/grub.cfg` and `binary/isolinux/live.cfg` directly.
+- Reruns the late binary stages (`binary_checksums`, `binary_iso`, `binary_zsync`) to rebuild
+  the ISO with the patched tree.
+
+This is the deterministic safety net: even if the hook runs at the wrong time, the recovery
+step handles the final `binary/` tree after live-build has written all bootloader configs.
+
+**3. ISO validation hardening**
+
+The memtest probe in `build.sh` is wrapped in explicit `if` / `case` control flow, not called
+as a bare command under `set -e`. A non-zero probe return (needs recovery) is intentional and
+handled — it does not abort the build prematurely.
+
+ISO reading (`xorriso -indev -ls` / extraction) is treated as a separate prerequisite.
+If the reader fails, the validator reports a reader error explicitly, not a memtest warning.
+This prevents the false-negative loop that burned 2026-04-01 v3.14–v3.19.
+
+### Why this works when earlier attempts did not
+
+The earlier patterns all shared a single flaw: they assumed a single build-time point
+(hook or source template) would be the last writer of bootloader configs and memtest payloads.
+In live-build on Debian Bookworm that assumption is false — live-build continues writing
+bootloader files after custom hooks run, and `chroot/boot/` does not reliably hold memtest payloads.
+
+The recovery step sidesteps the ordering problem entirely: it acts on the fully materialized
+`binary/` tree after `lb build` finishes, then rebuilds the ISO from that patched tree.
+There is no ordering dependency to get wrong.
+
+### Do not revert
+
+Do not remove the recovery step or the hook without a fresh real ISO build proving
+live-build alone produces all four required artifacts:
+- `boot/memtest86+x64.bin`
+- `boot/memtest86+x64.efi`
+- memtest entry in `boot/grub/grub.cfg`
+- memtest entry in `isolinux/live.cfg`

iso/builder/VERSIONS

@@ -8,7 +8,7 @@ NCCL_TESTS_VERSION=2.13.10
 NVCC_VERSION=12.8
 CUBLAS_VERSION=13.0.2.14-1
 CUDA_USERSPACE_VERSION=13.0.96-1
-DCGM_VERSION=4.5.2-1
+DCGM_VERSION=4.5.3-1
 JOHN_JUMBO_COMMIT=67fcf9fe5a
 ROCM_VERSION=6.3.4
 ROCM_SMI_VERSION=7.4.0.60304-76~22.04

iso/builder/auto/config

@@ -32,7 +32,7 @@ lb config noauto \
     --memtest memtest86+ \
     --iso-volume "EASY_BEE_${BEE_GPU_VENDOR_UPPER:-NVIDIA}" \
     --iso-application "EASY-BEE-${BEE_GPU_VENDOR_UPPER:-NVIDIA}" \
-    --bootappend-live "boot=live components video=1920x1080 console=tty0 console=ttyS0,115200n8 loglevel=7 username=bee user-fullname=Bee modprobe.blacklist=nouveau" \
+    --bootappend-live "boot=live components video=1920x1080 console=tty0 console=ttyS0,115200n8 loglevel=3 username=bee user-fullname=Bee modprobe.blacklist=nouveau,snd_hda_intel,snd_hda_codec_realtek,snd_hda_codec_generic,soundcore" \
     --apt-recommends false \
     --chroot-squashfs-compression-type zstd \
     "${@}"

iso/builder/build.sh

@@ -862,7 +862,6 @@ rm -f \
     "${OVERLAY_STAGE_DIR}/etc/bee-release" \
     "${OVERLAY_STAGE_DIR}/root/.ssh/authorized_keys" \
     "${OVERLAY_STAGE_DIR}/usr/local/bin/bee" \
-    "${OVERLAY_STAGE_DIR}/usr/local/bin/bee-nccl-gpu-stress" \
     "${OVERLAY_STAGE_DIR}/usr/local/bin/john" \
     "${OVERLAY_STAGE_DIR}/usr/local/lib/bee/bee-gpu-burn-worker" \
     "${OVERLAY_STAGE_DIR}/usr/local/bin/bee-smoketest" \
@@ -1136,13 +1135,16 @@ fi
 ISO_RAW="${LB_DIR}/live-image-amd64.hybrid.iso"
 if [ -f "$ISO_RAW" ]; then
     dump_memtest_debug "post-build" "${LB_DIR}" "$ISO_RAW"
-    iso_memtest_present "$ISO_RAW"
+    if iso_memtest_present "$ISO_RAW"; then
-    memtest_status=$?
+        :
-    if [ "$memtest_status" -eq 1 ]; then
+    else
-        recover_iso_memtest "${LB_DIR}" "$ISO_RAW"
+        memtest_status=$?
-        dump_memtest_debug "post-recovery" "${LB_DIR}" "$ISO_RAW"
+        if [ "$memtest_status" -eq 1 ]; then
-    elif [ "$memtest_status" -eq 2 ]; then
+            recover_iso_memtest "${LB_DIR}" "$ISO_RAW"
-        memtest_fail "failed to inspect ISO for memtest before recovery" "$ISO_RAW"
+            dump_memtest_debug "post-recovery" "${LB_DIR}" "$ISO_RAW"
+        elif [ "$memtest_status" -eq 2 ]; then
+            memtest_fail "failed to inspect ISO for memtest before recovery" "$ISO_RAW"
+        fi
     fi
     validate_iso_memtest "$ISO_RAW"
     cp "$ISO_RAW" "$ISO_OUT"

iso/builder/config/package-lists/bee-amd.list.chroot

@@ -1,3 +1,6 @@
+# AMD GPU firmware
+firmware-amd-graphics
+
 # AMD ROCm — GPU monitoring, bandwidth test, and compute stress (RVS GST)
 rocm-smi-lib=%%ROCM_SMI_VERSION%%
 rocm-bandwidth-test=%%ROCM_BANDWIDTH_TEST_VERSION%%

iso/builder/config/package-lists/bee.list.chroot

@@ -71,9 +71,7 @@ lightdm
 firmware-linux-free
 firmware-linux-nonfree
 firmware-misc-nonfree
-firmware-amd-graphics
 firmware-realtek
-firmware-intel-sound
 firmware-bnx2
 firmware-bnx2x
 firmware-cavium

iso/builder/smoketest.sh

@@ -52,6 +52,14 @@ else
     fail "nvidia-smi: NOT FOUND"
 fi
 
+for tool in bee-gpu-burn bee-john-gpu-stress bee-nccl-gpu-stress all_reduce_perf; do
+    if p=$(PATH="/usr/local/bin:$PATH" command -v "$tool" 2>/dev/null); then
+        ok "$tool found: $p"
+    else
+        fail "$tool: NOT FOUND"
+    fi
+done
+
 echo ""
 echo "-- NVIDIA modules --"
 KO_DIR="/usr/local/lib/nvidia"

iso/overlay/usr/local/bin/bee-john-gpu-stress

@@ -190,4 +190,16 @@ CHOSEN_FORMAT=$(choose_format) || {
 }
 
 echo "format=${CHOSEN_FORMAT}"
-exec ./john --test="${SECONDS}" --format="${CHOSEN_FORMAT}" --devices="${JOHN_DEVICES}"
+PIDS=""
+_first=1
+for opencl_id in $(echo "${JOHN_DEVICES}" | tr ',' ' '); do
+    [ "${_first}" = "1" ] || sleep 3
+    _first=0
+    ./john --test="${SECONDS}" --format="${CHOSEN_FORMAT}" --devices="${opencl_id}" &
+    PIDS="${PIDS} $!"
+done
+FAIL=0
+for pid in ${PIDS}; do
+    wait "${pid}" || FAIL=$((FAIL+1))
+done
+[ "${FAIL}" -eq 0 ] || { echo "john: ${FAIL} device(s) failed" >&2; exit 1; }

iso/overlay/usr/local/bin/bee-openbox-session

@@ -24,7 +24,7 @@ chromium \
     --no-first-run \
     --disable-session-crashed-bubble \
     --disable-features=TranslateUI \
-    --start-fullscreen \
+    --start-maximized \
     http://localhost/ &
 
 exec openbox

iso/overlay/usr/local/bin/netconf

@@ -3,6 +3,11 @@
 # Type 'a' at any prompt to abort, 'b' to go back.
 set -e
 
+# Requires root for ip/dhclient/resolv.conf — re-exec via sudo if needed.
+if [ "$(id -u)" -ne 0 ]; then
+    exec sudo "$0" "$@"
+fi
+
 abort() { echo "Aborted."; exit 0; }
 
 ask() {