Fix Proxmox workload refresh coherence

docs/release-control/v6/internal/subsystems/monitoring.md

@@ -27,6 +27,14 @@ Monitoring owns source freshness cadence for Proxmox, PBS, and PMG resources:
 the stale threshold is derived from the configured polling interval with a
 minimum floor, so API-facing resource status must not degrade merely because a
 healthy source is between normal poll cycles.
+Proxmox guest enumeration is a generation boundary. VM and LXC collection and
+enrichment must finish before one `State.UpdateGuestsForInstance` publication,
+so readers never observe a VM-only or LXC-only intermediate snapshot. A failed
+online cluster member retains only that member's last coherent guests and their
+source-native `{instance}:{node}:{vmid}` IDs; a successful empty member
+enumeration is authoritative and removes genuinely deleted guests. Collection
+failure remains visible through source freshness/error state and must not be
+converted into an authoritative empty inventory.
 Host-agent report liveness is server-observed, not agent-clock-observed:
 `ApplyHostReport` must stamp `Host.LastSeen`, agent-sourced Ceph cluster
 freshness, and host-agent cluster sensor freshness from Pulse receipt time, so
@@ -213,6 +221,7 @@ node-local Agent evidence.
 36. `internal/dockeragent/docker_client.go`
 37. `pkg/agents/docker/report.go`
 38. `internal/models/models.go`
+38a. `internal/models/proxmox_guest_state.go`
 39. `internal/models/models_frontend.go`
 40. `internal/models/converters.go`
 41. `internal/models/deepcopy.go`

docs/release-control/v6/internal/subsystems/performance-and-scalability.md

@@ -854,6 +854,19 @@ shell clickable behind another overlay.
 
 ## Current State
 
+### Workload refreshes retain one coherent paged generation
+
+The workload polling cache accepts a REST refresh only when every advertised
+page succeeds. A failed later page leaves the prior array and its stable row
+identities in place, ends loading, and exposes the refresh error; it must not
+publish a partial list, clear rows as a loading transition, or silently hide
+the error. Proxmox workload filtering consumes the source-authored
+`proxmox.runtimeStatus` before aggregate health so an availability-only
+freshness update cannot collapse a Running view. A later complete response may
+still remove an authoritatively deleted guest. These rules preserve sort,
+selection, drawer, and virtualized viewport state without adding another
+resource scan, websocket subscription, or browser-local source of truth.
+
 ### Canonical mutation-plane dependency
 
 Router wiring now exposes only typed action planning for model-originated

docs/release-control/v6/internal/subsystems/registry.json

@@ -5186,6 +5186,7 @@
         "internal/models/deepcopy.go",
         "internal/models/models.go",
         "internal/models/models_frontend.go",
+        "internal/models/proxmox_guest_state.go",
         "internal/proxmoxidentity/backup_identity.go",
         "pkg/agents/docker/report.go",
         "pkg/agents/host/report.go",
@@ -5398,6 +5399,7 @@
               "internal/models/deepcopy.go",
               "internal/models/models.go",
               "internal/models/models_frontend.go",
+              "internal/models/proxmox_guest_state.go",
               "pkg/agents/docker/report.go"
             ],
             "allow_same_subsystem_tests": false,

docs/release-control/v6/internal/subsystems/storage-recovery.md

@@ -1917,6 +1917,17 @@ that Safe auto-fix or Autopilot remediation is verified.
 
 ## Current State
 
+### Proxmox runtime continuity is not protection evidence
+
+The additive `ProxmoxData.RuntimeStatus` field preserves VM/LXC power-state
+presentation while a platform snapshot is stale or refreshing. Storage and
+recovery consumers must not interpret `running`, `stopped`, a retained guest
+row, or an availability facet as backup freshness, protection coverage,
+restore-chain verification, or recoverability. Authoritative guest deletion
+may remove the inventory row and its canonical history identity, while
+Recovery Assurance remains governed solely by its own backup and verification
+evidence.
+
 ### Storage history remount state is bounded
 
 Storage summary history no longer treats every node/range combination visited

docs/release-control/v6/internal/subsystems/unified-resources.md

@@ -43,6 +43,18 @@ comparable fallback because LXC observes the shared kernel and QEMU accounting
 is independently scoped. Agent CPU may fill the field only when the platform
 has no CPU observation. Agent-only fields that the platform does not provide
 remain eligible per metric.
+Proxmox guest power state is likewise source-authored semantic state, separate
+from aggregate resource health. `ProxmoxData.RuntimeStatus` retains the last
+coherent `running`/`stopped` observation while `Resource.Status` and
+`SourceStatus` continue to report stale, warning, offline, availability, and
+error evidence honestly. Availability checks may add facets and influence
+aggregate health after Proxmox becomes stale, but they must not overwrite or
+erase the Proxmox runtime state used by workload power filters. Full monitor
+registry generations and incremental agent/availability mutations serialize
+at the adapter boundary; a mutation that arrives during a rebuild must apply to
+the newly published registry rather than a superseded pointer. Authoritative
+snapshot omissions still remove resources and emit the normal canonical
+history change.
 Physical-disk resources own cross-source disk identity. When Proxmox inventory
 and host-agent SMART telemetry describe the same device, the merged resource
 must retain Proxmox node/instance source payloads while carrying SMART

frontend-modern/src/hooks/__tests__/useWorkloads.test.ts

@@ -228,6 +228,126 @@ describe('useWorkloads', () => {
     dispose();
   });
 
+  it('keeps Proxmox power state stable while aggregate freshness changes, then removes an authoritative deletion', async () => {
+    const guest = (vmid: number, status: string, runtimeStatus: string) => ({
+      ...sampleResource,
+      id: `cluster-a-pve1-${vmid}`,
+      type: 'system-container',
+      name: `lxc-${vmid}`,
+      status,
+      vmid,
+      proxmox: {
+        vmid,
+        nodeName: 'pve1',
+        instance: 'cluster-a',
+        runtimeStatus,
+      },
+    });
+
+    apiFetchJSONMock.mockResolvedValueOnce({
+      data: [
+        guest(101, 'online', 'running'),
+        guest(102, 'online', 'running'),
+        guest(103, 'online', 'running'),
+      ],
+      meta: { totalPages: 1 },
+    });
+
+    let dispose = () => {};
+    let result: ReturnType<UseWorkloadsModule['useWorkloads']> | undefined;
+    createRoot((d) => {
+      dispose = d;
+      const [enabled] = createSignal(true);
+      result = useWorkloads(enabled);
+    });
+
+    await waitForWorkloadCount(() => result!.workloads().length, 3);
+    const initialIds = result!.workloads().map((workload) => workload.id);
+
+    apiFetchJSONMock.mockResolvedValueOnce({
+      data: [
+        guest(101, 'warning', 'running'),
+        {
+          ...guest(102, 'online', 'running'),
+          availability: { targetId: 'probe-102', protocol: 'icmp', enabled: true, available: true },
+        },
+        guest(103, 'warning', 'running'),
+      ],
+      meta: { totalPages: 1 },
+    });
+    await result!.refetch();
+
+    expect(result!.workloads().map((workload) => workload.id)).toEqual(initialIds);
+    expect(result!.workloads().map((workload) => workload.status)).toEqual([
+      'running',
+      'running',
+      'running',
+    ]);
+
+    apiFetchJSONMock.mockResolvedValueOnce({
+      data: [guest(101, 'online', 'running'), guest(102, 'online', 'running')],
+      meta: { totalPages: 1 },
+    });
+    await result!.refetch();
+
+    expect(result!.workloads().map((workload) => workload.id)).toEqual(initialIds.slice(0, 2));
+    expect(result!.error()).toBeUndefined();
+
+    dispose();
+  });
+
+  it('rejects a partial paged refresh and retains the last coherent snapshot', async () => {
+    const secondResource = {
+      ...sampleResource,
+      id: 'cluster-a-pve1-102',
+      name: 'vm-102',
+      vmid: 102,
+    };
+    apiFetchJSONMock.mockResolvedValueOnce({
+      data: [sampleResource, secondResource],
+      meta: { totalPages: 1 },
+    });
+
+    let dispose = () => {};
+    let result: ReturnType<UseWorkloadsModule['useWorkloads']> | undefined;
+    createRoot((d) => {
+      dispose = d;
+      const [enabled] = createSignal(true);
+      result = useWorkloads(enabled);
+    });
+
+    await waitForWorkloadCount(() => result!.workloads().length, 2);
+    const coherentSnapshot = result!.workloads();
+
+    apiFetchJSONMock
+      .mockResolvedValueOnce({
+        data: [sampleResource],
+        meta: { totalPages: 2 },
+      })
+      .mockRejectedValueOnce(new Error('page 2 unavailable'));
+
+    await expect(result!.refetch()).rejects.toThrow('page 2 unavailable');
+    expect(result!.workloads()).toBe(coherentSnapshot);
+    expect(result!.workloads()).toHaveLength(2);
+    expect(result!.error()).toBeInstanceOf(Error);
+
+    apiFetchJSONMock
+      .mockResolvedValueOnce({
+        data: [sampleResource],
+        meta: { totalPages: 2 },
+      })
+      .mockResolvedValueOnce({
+        data: [],
+        meta: { totalPages: 2 },
+      });
+    await result!.refetch();
+
+    expect(result!.workloads()).toHaveLength(1);
+    expect(result!.error()).toBeUndefined();
+
+    dispose();
+  });
+
   it('does not apply in-flight workload results after the hook is disabled', async () => {
     const pendingFetch = deferred<unknown>();
     apiFetchJSONMock.mockImplementationOnce(() => pendingFetch.promise as Promise<any>);

frontend-modern/src/hooks/useWorkloads.ts

@@ -87,6 +87,7 @@ type APIResource = {
   node?: string;
   instance?: string;
   proxmox?: {
+    runtimeStatus?: string;
     nodeName?: string;
     clusterName?: string;
     instance?: string;
@@ -451,7 +452,9 @@ const mapResourceToWorkload = (resource: APIResource): WorkloadGuest | null => {
     node,
     instance,
     status: normalizeWorkloadStatus(
-      resource.status || (platformType === 'vmware-vsphere' ? resource.vmware?.powerState : null),
+      resource.proxmox?.runtimeStatus ||
+        resource.status ||
+        (platformType === 'vmware-vsphere' ? resource.vmware?.powerState : null),
     ),
     type:
       workloadType === 'vm'
@@ -587,10 +590,9 @@ async function fetchWorkloads(): Promise<WorkloadGuest[]> {
     for (let page = 2; page <= totalPages; page++) {
       pageRequests.push(apiFetchJSON<unknown>(buildWorkloadsUrl(page), { cache: 'no-store' }));
     }
-    const settled = await Promise.allSettled(pageRequests);
-    for (const result of settled) {
-      if (result.status !== 'fulfilled') continue;
-      const pageData = resolveWorkloadsPayload(result.value);
+    const responses = await Promise.all(pageRequests);
+    for (const response of responses) {
+      const pageData = resolveWorkloadsPayload(response);
       allResources.push(...pageData.data);
     }
   }
@@ -775,8 +777,12 @@ export function useWorkloads(enabled: Accessor<boolean> = () => true) {
         }
         applyWorkloads(data, scope);
         setError(undefined);
-      } catch {
-        // Silently ignore poll errors; keep showing last data
+      } catch (err) {
+        // Keep the last coherent snapshot while making the failed refresh
+        // observable to the surface.
+        if (scope === resolveActiveOrgScope()) {
+          setError(err);
+        }
       }
     }, DEFAULT_POLL_INTERVAL_MS);
     onCleanup(() => clearInterval(id));

frontend-modern/src/types/__tests__/resource.test.ts

@@ -240,6 +240,7 @@ describe('Resource Helper Functions', () => {
         vmid: 101,
         nodeName: 'pve-a',
         instance: 'cluster-a',
+        runtimeStatus: 'running',
         diskStatusReason: 'agent-not-running',
         guestAgentStatus: 'expected-unreachable',
         guestAgentExpected: true,
@@ -262,6 +263,7 @@ describe('Resource Helper Functions', () => {
 
       expect(resource.proxmox?.guestAgentStatus).toBe('expected-unreachable');
       expect(resource.proxmox?.guestAgentExpected).toBe(true);
+      expect(resource.proxmox?.runtimeStatus).toBe('running');
       expect(resource.proxmox?.diskStatusReason).toBe('agent-not-running');
       expect(resource.incidents?.[0]?.source).toBe('qemu-guest-agent');
     });

frontend-modern/src/types/resource.ts

@@ -644,6 +644,8 @@ export interface ResourceAgentMeta {
 }
 
 export interface ResourceProxmoxMeta {
+  /** Proxmox-authored VM/LXC power state, kept separate from collection freshness. */
+  runtimeStatus?: string;
   vmid?: number;
   node?: string;
   nodeName?: string;

internal/api/resources_test.go

@@ -779,11 +779,12 @@ func TestResourceListDerivesProxmoxWorkloadParentFromUnifiedSeed(t *testing.T) {
 				Sources:  []unified.DataSource{unified.SourceProxmox},
 				Identity: unified.ResourceIdentity{Hostnames: []string{"cloudflared"}},
 				Proxmox: &unified.ProxmoxData{
-					SourceID:    "delly:delly:104",
-					NodeName:    "delly",
-					ClusterName: "homelab",
-					Instance:    "delly",
-					VMID:        104,
+					SourceID:      "delly:delly:104",
+					RuntimeStatus: "running",
+					NodeName:      "delly",
+					ClusterName:   "homelab",
+					Instance:      "delly",
+					VMID:          104,
 				},
 			},
 		},
@@ -815,6 +816,9 @@ func TestResourceListDerivesProxmoxWorkloadParentFromUnifiedSeed(t *testing.T) {
 	if resource.Proxmox == nil || resource.Proxmox.NodeName != "delly" || resource.Proxmox.ClusterName != "homelab" {
 		t.Fatalf("expected cloudflared Proxmox node metadata, got %+v", resource.Proxmox)
 	}
+	if resource.Proxmox.RuntimeStatus != "running" {
+		t.Fatalf("expected source-authored runtime status, got %+v", resource.Proxmox)
+	}
 }
 
 func TestStateEndpointDerivesProxmoxWorkloadParentFromSupplementalRecords(t *testing.T) {

internal/models/proxmox_guest_state.go

@@ -0,0 +1,34 @@
+package models
+
+import "time"
+
+// UpdateGuestsForInstance replaces the VM and container projections for one
+// Proxmox instance under a single state lock. Pollers collect and enrich both
+// guest kinds before calling this method so readers cannot observe a mixed
+// generation while a refresh is in flight.
+func (s *State) UpdateGuestsForInstance(instanceName string, vms []VM, containers []Container) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+
+	s.VMs = updateSliceByInstanceWithBackup(
+		s.VMs, vms, instanceName,
+		func(vm VM) string { return vm.ID },
+		func(vm VM) string { return vm.Instance },
+		func(vm VM) int { return vm.VMID },
+		func(vm VM) time.Time { return vm.LastBackup },
+		func(vm VM, t time.Time) VM { vm.LastBackup = t; return vm },
+		cloneVM,
+		func(items []VM, i, j int) bool { return items[i].VMID < items[j].VMID },
+	)
+	s.Containers = updateSliceByInstanceWithBackup(
+		s.Containers, containers, instanceName,
+		func(ct Container) string { return ct.ID },
+		func(ct Container) string { return ct.Instance },
+		func(ct Container) int { return ct.VMID },
+		func(ct Container) time.Time { return ct.LastBackup },
+		func(ct Container, t time.Time) Container { ct.LastBackup = t; return ct },
+		cloneContainer,
+		func(items []Container, i, j int) bool { return items[i].VMID < items[j].VMID },
+	)
+	s.LastUpdate = time.Now()
+}

internal/models/state_host_test.go

@@ -1327,6 +1327,44 @@ func TestSyncGuestBackupTimesVMContainerCollision(t *testing.T) {
 	}
 }
 
+func TestUpdateGuestsForInstancePublishesCoherentGeneration(t *testing.T) {
+	state := NewState()
+	previousBackup := time.Now().Add(-2 * time.Hour).UTC()
+
+	state.UpdateVMs([]VM{
+		{ID: "lab-a:node-a:101", VMID: 101, Instance: "lab-a", Node: "node-a", LastBackup: previousBackup},
+		{ID: "lab-b:node-b:201", VMID: 201, Instance: "lab-b", Node: "node-b"},
+	})
+	state.UpdateContainers([]Container{
+		{ID: "lab-a:node-a:102", VMID: 102, Instance: "lab-a", Node: "node-a", LastBackup: previousBackup},
+		{ID: "lab-b:node-b:202", VMID: 202, Instance: "lab-b", Node: "node-b"},
+	})
+	before := state.GetSnapshot().LastUpdate
+
+	state.UpdateGuestsForInstance(
+		"lab-a",
+		[]VM{{ID: "lab-a:node-a:101", VMID: 101, Instance: "lab-a", Node: "node-a", Status: "running"}},
+		[]Container{{ID: "lab-a:node-a:103", VMID: 103, Instance: "lab-a", Node: "node-a", Status: "running"}},
+	)
+
+	snapshot := state.GetSnapshot()
+	if !snapshot.LastUpdate.After(before) {
+		t.Fatalf("LastUpdate did not advance: before=%v after=%v", before, snapshot.LastUpdate)
+	}
+	if len(snapshot.VMs) != 2 || len(snapshot.Containers) != 2 {
+		t.Fatalf("unexpected coherent generation sizes: vms=%d containers=%d", len(snapshot.VMs), len(snapshot.Containers))
+	}
+	if snapshot.VMs[0].ID != "lab-a:node-a:101" || !snapshot.VMs[0].LastBackup.Equal(previousBackup) {
+		t.Fatalf("updated VM did not retain backup state: %+v", snapshot.VMs[0])
+	}
+	if snapshot.Containers[0].ID != "lab-a:node-a:103" {
+		t.Fatalf("authoritatively deleted container remained in state: %+v", snapshot.Containers)
+	}
+	if snapshot.VMs[1].Instance != "lab-b" || snapshot.Containers[1].Instance != "lab-b" {
+		t.Fatalf("other instance was not isolated: vms=%+v containers=%+v", snapshot.VMs, snapshot.Containers)
+	}
+}
+
 func TestUpdateStorageBackupsForInstance(t *testing.T) {
 	state := NewState()
 

internal/monitoring/canonical_guardrails_test.go

@@ -1110,11 +1110,14 @@ func TestProxmoxGuestDockerInventoryUsesCanonicalReportIngestPath(t *testing.T)
 		},
 		"monitor_pve_guest_poll.go": {
 			"m.CollectProxmoxGuestDockerInventory(ctx, allContainers)",
-			"m.state.UpdateContainersForInstance(instanceName, allContainers)",
+			"m.state.UpdateGuestsForInstance(instanceName, allVMs, allContainers)",
 		},
 		"monitor_polling_containers.go": {
 			"m.CollectProxmoxGuestDockerInventory(ctx, allContainers)",
-			"m.state.UpdateContainersForInstance(instanceName, allContainers)",
+			"return allContainers",
+		},
+		"monitor_pve.go": {
+			"m.state.UpdateGuestsForInstance(instanceName, vms, containers)",
 		},
 	}
 

internal/monitoring/memory_trust_characterization_test.go

@@ -22,6 +22,38 @@ type vmMemoryTrustStubClient struct {
 	vmAgentMemCalls     int
 }
 
+func TestCollectVMsWithNodesRetainsFailedNodeRuntimeState(t *testing.T) {
+	monitor := newTestPVEMonitor("lab")
+	defer monitor.alertManager.Stop()
+	defer monitor.notificationMgr.Stop()
+	monitor.state.UpdateVMsForInstance("lab", []models.VM{{
+		ID:       "lab:node-b:201",
+		VMID:     201,
+		Name:     "database",
+		Node:     "node-b",
+		Instance: "lab",
+		Status:   "running",
+	}})
+
+	client := &partialNodeGuestClient{
+		stubPVEClient: &stubPVEClient{},
+		failedNodes:   map[string]bool{"node-b": true},
+	}
+	vms := monitor.collectVMsWithNodes(
+		context.Background(),
+		"lab",
+		"",
+		false,
+		client,
+		[]proxmox.Node{{Node: "node-b", Status: "online"}},
+		map[string]string{"node-b": "online"},
+	)
+
+	if len(vms) != 1 || vms[0].ID != "lab:node-b:201" || vms[0].Status != "running" {
+		t.Fatalf("failed-node VM continuity = %+v, want retained source ID and running state", vms)
+	}
+}
+
 func (s *vmMemoryTrustStubClient) GetVMs(ctx context.Context, node string) ([]proxmox.VM, error) {
 	return s.vms, nil
 }

internal/monitoring/monitor_extra_coverage_test.go

@@ -1172,12 +1172,12 @@ func TestMonitor_PreviousGuestContextForInstance_Extra(t *testing.T) {
 	if len(prev.vms) != 1 || prev.vms[0].VMID != 101 || prev.vms[0].Instance != "pve1" || prev.vms[0].Name != "vm1" {
 		t.Fatalf("expected only pve1 VMs, got %#v", prev.vms)
 	}
-	canonicalID := prev.vms[0].ID
-	if len(prev.vmsByID) != 2 || prev.vmsByID[canonicalID].VMID != 101 || prev.vmsByID[makeGuestID("pve1", "", 101)].VMID != 101 {
-		t.Fatalf("expected previous VM lookup to be indexed by canonical and runtime guest IDs, got %#v", prev.vmsByID)
+	guestID := makeGuestID("pve1", "", 101)
+	if prev.vms[0].ID != guestID || len(prev.vmsByID) != 1 || prev.vmsByID[guestID].VMID != 101 {
+		t.Fatalf("expected previous VM lookup to retain the source-authored guest ID, got %#v", prev.vmsByID)
 	}
-	if prev.vmsByID[canonicalID].Disk.Total != 100 || prev.vmsByID[canonicalID].Disk.Used != 40 {
-		t.Fatalf("expected previous VM projection to preserve aggregate disk summary, got %#v", prev.vmsByID[canonicalID].Disk)
+	if prev.vmsByID[guestID].Disk.Total != 100 || prev.vmsByID[guestID].Disk.Used != 40 {
+		t.Fatalf("expected previous VM projection to preserve aggregate disk summary, got %#v", prev.vmsByID[guestID].Disk)
 	}
 	if len(prev.containers) != 2 {
 		t.Fatalf("expected only pve1 containers, got %#v", prev.containers)

internal/monitoring/monitor_polling_containers.go

@@ -13,7 +13,7 @@ import (
 	"github.com/rs/zerolog/log"
 )
 
-func (m *Monitor) pollContainersWithNodes(ctx context.Context, instanceName string, clusterName string, isCluster bool, client PVEClientInterface, nodes []proxmox.Node, nodeEffectiveStatus map[string]string) {
+func (m *Monitor) collectContainersWithNodes(ctx context.Context, instanceName string, clusterName string, isCluster bool, client PVEClientInterface, nodes []proxmox.Node, nodeEffectiveStatus map[string]string) []models.Container {
 	startTime := time.Now()
 
 	// Channel to collect container results from each node
@@ -303,10 +303,12 @@ func (m *Monitor) pollContainersWithNodes(ctx context.Context, instanceName stri
 	lxcTemplateSubjects := make(map[string]struct{})
 	successfulNodes := 0
 	failedNodes := 0
+	failedNodeNames := make(map[string]struct{})
 
 	for result := range resultChan {
 		if result.err != nil {
 			failedNodes++
+			failedNodeNames[result.node] = struct{}{}
 		} else {
 			successfulNodes++
 			allContainers = append(allContainers, result.containers...)
@@ -319,28 +321,27 @@ func (m *Monitor) pollContainersWithNodes(ctx context.Context, instanceName stri
 		m.updatePVEBackupTemplateSubjectsForType(instanceName, "lxc", lxcTemplateSubjects)
 	}
 
-	// If we got ZERO containers but had containers before (likely cluster health issue),
-	// preserve previous containers instead of clearing them
-	if len(allContainers) == 0 && len(nodes) > 0 {
-		allContainers = append(allContainers, prevGuests.containers...)
-		prevContainerCount := len(prevGuests.containers)
-		if prevContainerCount > 0 {
-			log.Warn().
-				Str("instance", instanceName).
-				Int("prevContainers", prevContainerCount).
-				Int("successfulNodes", successfulNodes).
-				Int("totalNodes", len(nodes)).
-				Msg("Traditional polling returned zero containers but had containers before - preserving previous containers")
+	preservedContainers := 0
+	if len(failedNodeNames) > 0 {
+		for _, container := range prevGuests.containers {
+			if _, failed := failedNodeNames[container.Node]; failed {
+				allContainers = append(allContainers, container)
+				preservedContainers++
+			}
 		}
 	}
+	if preservedContainers > 0 {
+		log.Warn().
+			Str("instance", instanceName).
+			Int("preservedContainers", preservedContainers).
+			Int("failedNodes", failedNodes).
+			Msg("Preserved prior containers for nodes whose enumeration failed")
+	}
 
 	// Check Docker presence for containers that need it (new, restarted, started)
 	allContainers = m.CheckContainersForDocker(ctx, allContainers)
 	m.CollectProxmoxGuestDockerInventory(ctx, allContainers)
 
-	// Update state with all containers
-	m.state.UpdateContainersForInstance(instanceName, allContainers)
-
 	// Record guest metrics history for running containers (enables sparkline/trends view)
 	if !shouldSkipNativeMockStateMetricWrites() {
 		now := time.Now()
@@ -362,6 +363,16 @@ func (m *Monitor) pollContainersWithNodes(ctx context.Context, instanceName stri
 		Int("failedNodes", failedNodes).
 		Dur("duration", duration).
 		Msg("Parallel container polling completed")
+
+	return allContainers
+}
+
+// pollContainersWithNodes retains the focused single-kind polling entry point
+// used by tests and maintenance callers. The production guest cycle uses
+// collectContainersWithNodes and publishes both guest kinds atomically.
+func (m *Monitor) pollContainersWithNodes(ctx context.Context, instanceName string, clusterName string, isCluster bool, client PVEClientInterface, nodes []proxmox.Node, nodeEffectiveStatus map[string]string) {
+	containers := m.collectContainersWithNodes(ctx, instanceName, clusterName, isCluster, client, nodes, nodeEffectiveStatus)
+	m.state.UpdateContainersForInstance(instanceName, containers)
 }
 
 // pollStorageWithNodes polls storage from all nodes in parallel using goroutines

internal/monitoring/monitor_polling_vm.go

@@ -12,7 +12,7 @@ import (
 	"github.com/rs/zerolog/log"
 )
 
-func (m *Monitor) pollVMsWithNodes(ctx context.Context, instanceName string, clusterName string, isCluster bool, client PVEClientInterface, nodes []proxmox.Node, nodeEffectiveStatus map[string]string) {
+func (m *Monitor) collectVMsWithNodes(ctx context.Context, instanceName string, clusterName string, isCluster bool, client PVEClientInterface, nodes []proxmox.Node, nodeEffectiveStatus map[string]string) []models.VM {
 	startTime := time.Now()
 
 	type nodeResult struct {
@@ -85,10 +85,12 @@ func (m *Monitor) pollVMsWithNodes(ctx context.Context, instanceName string, clu
 	qemuTemplateSubjects := make(map[string]struct{})
 	successfulNodes := 0
 	failedNodes := 0
+	failedNodeNames := make(map[string]struct{})
 
 	for result := range resultChan {
 		if result.err != nil {
 			failedNodes++
+			failedNodeNames[result.node] = struct{}{}
 			continue
 		}
 		successfulNodes++
@@ -101,20 +103,22 @@ func (m *Monitor) pollVMsWithNodes(ctx context.Context, instanceName string, clu
 		m.updatePVEBackupTemplateSubjectsForType(instanceName, "qemu", qemuTemplateSubjects)
 	}
 
-	if len(allVMs) == 0 && len(nodes) > 0 {
-		allVMs = append(allVMs, prevGuests.vms...)
-		prevVMCount := len(prevGuests.vms)
-		if prevVMCount > 0 {
-			log.Warn().
-				Str("instance", instanceName).
-				Int("prevVMs", prevVMCount).
-				Int("successfulNodes", successfulNodes).
-				Int("totalNodes", len(nodes)).
-				Msg("Traditional polling returned zero VMs but had VMs before - preserving previous VMs")
+	preservedVMs := 0
+	if len(failedNodeNames) > 0 {
+		for _, vm := range prevGuests.vms {
+			if _, failed := failedNodeNames[vm.Node]; failed {
+				allVMs = append(allVMs, vm)
+				preservedVMs++
+			}
 		}
 	}
-
-	m.state.UpdateVMsForInstance(instanceName, allVMs)
+	if preservedVMs > 0 {
+		log.Warn().
+			Str("instance", instanceName).
+			Int("preservedVMs", preservedVMs).
+			Int("failedNodes", failedNodes).
+			Msg("Preserved prior VMs for nodes whose enumeration failed")
+	}
 
 	if !shouldSkipNativeMockStateMetricWrites() {
 		now := time.Now()
@@ -136,4 +140,14 @@ func (m *Monitor) pollVMsWithNodes(ctx context.Context, instanceName string, clu
 		Int("failedNodes", failedNodes).
 		Dur("duration", duration).
 		Msg("Parallel VM polling completed")
+
+	return allVMs
+}
+
+// pollVMsWithNodes retains the focused single-kind polling entry point used by
+// tests and maintenance callers. The production guest cycle uses
+// collectVMsWithNodes and publishes both guest kinds atomically.
+func (m *Monitor) pollVMsWithNodes(ctx context.Context, instanceName string, clusterName string, isCluster bool, client PVEClientInterface, nodes []proxmox.Node, nodeEffectiveStatus map[string]string) {
+	vms := m.collectVMsWithNodes(ctx, instanceName, clusterName, isCluster, client, nodes, nodeEffectiveStatus)
+	m.state.UpdateVMsForInstance(instanceName, vms)
 }

internal/monitoring/monitor_previous_state.go

@@ -93,14 +93,17 @@ func previousVMFromView(vm *unifiedresources.VMView) models.VM {
 	if vm == nil {
 		return models.VM{}
 	}
+	instance := vm.Instance()
+	node := vm.Node()
+	vmid := vm.VMID()
 	return models.VM{
-		ID:           vm.ID(),
-		Instance:     vm.Instance(),
-		Node:         vm.Node(),
-		VMID:         vm.VMID(),
+		ID:           makeGuestID(instance, node, vmid),
+		Instance:     instance,
+		Node:         node,
+		VMID:         vmid,
 		Name:         vm.Name(),
 		Type:         "qemu",
-		Status:       string(vm.Status()),
+		Status:       vm.RuntimeStatus(),
 		IPAddresses:  vm.IPAddresses(),
 		OSName:       vm.OSName(),
 		OSVersion:    vm.OSVersion(),
@@ -122,13 +125,16 @@ func previousContainerFromView(ct *unifiedresources.ContainerView) models.Contai
 	if ct == nil {
 		return models.Container{}
 	}
+	instance := ct.Instance()
+	node := ct.Node()
+	vmid := ct.VMID()
 	return models.Container{
-		ID:       ct.ID(),
-		Instance: ct.Instance(),
-		Node:     ct.Node(),
-		VMID:     ct.VMID(),
+		ID:       makeGuestID(instance, node, vmid),
+		Instance: instance,
+		Node:     node,
+		VMID:     vmid,
 		Name:     ct.Name(),
-		Status:   string(ct.Status()),
+		Status:   ct.RuntimeStatus(),
 		Type:     ct.ContainerType(),
 		IsOCI:    ct.IsOCI(),
 		LastSeen: ct.LastSeen(),

internal/monitoring/monitor_pve.go

@@ -803,12 +803,16 @@ func (m *Monitor) pollGuestsWithFallback(
 		}
 
 		// Use optimized parallel polling for better performance
+		previous := m.previousGuestContextForInstance(instanceName)
+		vms := previous.vms
+		containers := previous.containers
 		if instanceCfg.MonitorVMs {
-			m.pollVMsWithNodes(ctx, instanceName, instanceCfg.ClusterName, instanceCfg.IsCluster, client, nodes, nodeEffectiveStatus)
+			vms = m.collectVMsWithNodes(ctx, instanceName, instanceCfg.ClusterName, instanceCfg.IsCluster, client, nodes, nodeEffectiveStatus)
 		}
 		if instanceCfg.MonitorContainers {
-			m.pollContainersWithNodes(ctx, instanceName, instanceCfg.ClusterName, instanceCfg.IsCluster, client, nodes, nodeEffectiveStatus)
+			containers = m.collectContainersWithNodes(ctx, instanceName, instanceCfg.ClusterName, instanceCfg.IsCluster, client, nodes, nodeEffectiveStatus)
 		}
+		m.state.UpdateGuestsForInstance(instanceName, vms, containers)
 	}
 
 	return nil

internal/monitoring/monitor_pve_guest_poll.go

@@ -44,15 +44,14 @@ func (m *Monitor) pollVMsAndContainersEfficient(ctx context.Context, instanceNam
 
 	allVMs, allContainers = m.preserveGuestsForGracePeriod(instanceName, resources, prevGuests.vms, prevGuests.containers, nodeEffectiveStatus, allVMs, allContainers)
 
-	// Always update state when using efficient polling path
-	// Even if arrays are empty, we need to update to clear out VMs from genuinely offline nodes
-	m.state.UpdateVMsForInstance(instanceName, allVMs)
-
 	// Check Docker presence for containers that need it (new, restarted, started)
 	allContainers = m.CheckContainersForDocker(ctx, allContainers)
 	m.CollectProxmoxGuestDockerInventory(ctx, allContainers)
 
-	m.state.UpdateContainersForInstance(instanceName, allContainers)
+	// Publish the complete guest generation only after both VM and container
+	// collection/enrichment has finished. Empty authoritative results still
+	// remove genuinely deleted guests.
+	m.state.UpdateGuestsForInstance(instanceName, allVMs, allContainers)
 
 	m.recordGuestMetrics(allVMs, allContainers)
 

internal/monitoring/proxmox_guest_snapshot_coherence_test.go

@@ -0,0 +1,116 @@
+package monitoring
+
+import (
+	"context"
+	"errors"
+	"testing"
+
+	"github.com/rcourtman/pulse-go-rewrite/internal/config"
+	"github.com/rcourtman/pulse-go-rewrite/internal/models"
+	"github.com/rcourtman/pulse-go-rewrite/pkg/proxmox"
+)
+
+type partialNodeGuestClient struct {
+	*stubPVEClient
+	failedNodes      map[string]bool
+	vmsByNode        map[string][]proxmox.VM
+	containersByNode map[string][]proxmox.Container
+}
+
+func (c *partialNodeGuestClient) GetClusterResources(context.Context, string) ([]proxmox.ClusterResource, error) {
+	return nil, errors.New("cluster resources unavailable")
+}
+
+func (c *partialNodeGuestClient) GetVMs(_ context.Context, node string) ([]proxmox.VM, error) {
+	if c.failedNodes[node] {
+		return nil, errors.New("VM enumeration unavailable")
+	}
+	return c.vmsByNode[node], nil
+}
+
+func (c *partialNodeGuestClient) GetContainers(_ context.Context, node string) ([]proxmox.Container, error) {
+	if c.failedNodes[node] {
+		return nil, errors.New("container enumeration unavailable")
+	}
+	return c.containersByNode[node], nil
+}
+
+func TestPollGuestsWithFallbackRetainsOnlyFailedNodeGeneration(t *testing.T) {
+	monitor := newTestPVEMonitor("lab")
+	defer monitor.alertManager.Stop()
+	defer monitor.notificationMgr.Stop()
+
+	monitor.state.UpdateGuestsForInstance(
+		"lab",
+		[]models.VM{
+			{ID: "lab:node-a:101", VMID: 101, Name: "deleted-vm", Node: "node-a", Instance: "lab", Status: "running"},
+			{ID: "lab:node-b:201", VMID: 201, Name: "retained-vm", Node: "node-b", Instance: "lab", Status: "running"},
+		},
+		[]models.Container{
+			{ID: "lab:node-a:102", VMID: 102, Name: "deleted-ct", Node: "node-a", Instance: "lab", Status: "running"},
+			{ID: "lab:node-b:202", VMID: 202, Name: "retained-ct", Node: "node-b", Instance: "lab", Status: "running"},
+		},
+	)
+	monitor.state.UpdateGuestsForInstance(
+		"other",
+		[]models.VM{{ID: "other:node-c:301", VMID: 301, Node: "node-c", Instance: "other"}},
+		[]models.Container{{ID: "other:node-c:302", VMID: 302, Node: "node-c", Instance: "other"}},
+	)
+
+	client := &partialNodeGuestClient{
+		stubPVEClient:    &stubPVEClient{},
+		failedNodes:      map[string]bool{"node-b": true},
+		vmsByNode:        map[string][]proxmox.VM{"node-a": {}},
+		containersByNode: map[string][]proxmox.Container{"node-a": {}},
+	}
+	nodes := []proxmox.Node{
+		{Node: "node-a", Status: "online"},
+		{Node: "node-b", Status: "online"},
+	}
+	nodeStatus := map[string]string{"node-a": "online", "node-b": "online"}
+	cfg := &config.PVEInstance{MonitorVMs: true, MonitorContainers: true}
+
+	if err := monitor.pollGuestsWithFallback(context.Background(), "lab", cfg, client, nodes, nodeStatus); err != nil {
+		t.Fatalf("partial poll failed: %v", err)
+	}
+
+	snapshot := monitor.GetState()
+	assertGuestIDs(t, snapshot.VMs, []string{"lab:node-b:201", "other:node-c:301"})
+	assertGuestIDs(t, snapshot.Containers, []string{"lab:node-b:202", "other:node-c:302"})
+	if snapshot.VMs[0].Status != "running" || snapshot.Containers[0].Status != "running" {
+		t.Fatalf("failed-node power state was not retained: vm=%q container=%q", snapshot.VMs[0].Status, snapshot.Containers[0].Status)
+	}
+
+	client.failedNodes["node-b"] = false
+	client.vmsByNode["node-b"] = []proxmox.VM{}
+	client.containersByNode["node-b"] = []proxmox.Container{}
+
+	if err := monitor.pollGuestsWithFallback(context.Background(), "lab", cfg, client, nodes, nodeStatus); err != nil {
+		t.Fatalf("recovery poll failed: %v", err)
+	}
+
+	snapshot = monitor.GetState()
+	assertGuestIDs(t, snapshot.VMs, []string{"other:node-c:301"})
+	assertGuestIDs(t, snapshot.Containers, []string{"other:node-c:302"})
+}
+
+func assertGuestIDs[T models.VM | models.Container](t *testing.T, guests []T, want []string) {
+	t.Helper()
+	got := make([]string, 0, len(guests))
+	for _, guest := range guests {
+		switch typed := any(guest).(type) {
+		case models.VM:
+			got = append(got, typed.ID)
+		case models.Container:
+			got = append(got, typed.ID)
+		}
+	}
+	if len(got) != len(want) {
+		t.Fatalf("guest ids = %v, want %v", got, want)
+	}
+	for i := range want {
+		if got[i] != want[i] {
+			t.Fatalf("guest ids = %v, want %v", got, want)
+		}
+	}
+}

internal/unifiedresources/adapter_coverage_test.go

@@ -200,6 +200,19 @@ func TestMonitorAdapterUsesConfiguredProxmoxStaleThreshold(t *testing.T) {
 	if defaultVMs[0].Status() != StatusWarning {
 		t.Fatalf("default stale threshold status = %q, want warning", defaultVMs[0].Status())
 	}
+	if defaultVMs[0].RuntimeStatus() != "running" {
+		t.Fatalf("default stale runtime status = %q, want running", defaultVMs[0].RuntimeStatus())
+	}
+	defaultResources := defaultAdapter.GetByType(ResourceTypeVM)
+	if len(defaultResources) != 1 || defaultResources[0].Proxmox == nil {
+		t.Fatalf("expected one Proxmox VM resource, got %+v", defaultResources)
+	}
+	if got := defaultResources[0].Proxmox.RuntimeStatus; got != "running" {
+		t.Fatalf("Proxmox runtime status = %q, want running while collection is stale", got)
+	}
+	if got := defaultResources[0].SourceStatus[SourceProxmox].Status; got != "stale" {
+		t.Fatalf("Proxmox source status = %q, want stale", got)
+	}
 
 	adapter := NewMonitorAdapterWithStaleThresholds(NewRegistry(nil), map[DataSource]time.Duration{
 		SourceProxmox: 10 * time.Minute,

internal/unifiedresources/adapters.go

@@ -1676,6 +1676,7 @@ func resourceFromVM(vm models.VM) (Resource, ResourceIdentity) {
 	metrics := metricsFromVM(vm)
 	proxmox := &ProxmoxData{
 		SourceID:           sourceID,
+		RuntimeStatus:      vm.Status,
 		NodeName:           vm.Node,
 		Pool:               vm.Pool,
 		Instance:           vm.Instance,
@@ -1742,6 +1743,7 @@ func resourceFromContainer(ct models.Container) (Resource, ResourceIdentity) {
 	metrics := metricsFromContainer(ct)
 	proxmox := &ProxmoxData{
 		SourceID:          sourceID,
+		RuntimeStatus:     ct.Status,
 		NodeName:          ct.Node,
 		Pool:              ct.Pool,
 		Instance:          ct.Instance,

internal/unifiedresources/canonical_ids_types_test.go

@@ -22,6 +22,23 @@ func TestCanonicalResourceTypeDoesNotAliasHost(t *testing.T) {
 	}
 }
 
+func TestProxmoxRuntimeStatusJSONContract(t *testing.T) {
+	payload := ProxmoxData{
+		SourceID:      "lab:node-a:101",
+		RuntimeStatus: "running",
+		NodeName:      "node-a",
+		VMID:          101,
+	}
+
+	data, err := json.Marshal(payload)
+	if err != nil {
+		t.Fatalf("marshal ProxmoxData: %v", err)
+	}
+	if !strings.Contains(string(data), `"runtimeStatus":"running"`) {
+		t.Fatalf("ProxmoxData JSON did not carry runtimeStatus: %s", data)
+	}
+}
+
 func TestHostSMARTMetaCarriesSizeBytesJSONContract(t *testing.T) {
 	payload := HostSMARTMeta{
 		Device:    "/dev/sda",

internal/unifiedresources/monitor_adapter.go

@@ -14,6 +14,11 @@ import (
 type MonitorAdapter struct {
 	registry *ResourceRegistry
 
+	// mutationMu serializes complete registry generations with incremental
+	// supplemental updates. Registry construction intentionally happens while
+	// readers keep using the prior pointer, but two writers must never publish
+	// out of order or mutate a registry after it has been replaced.
+	mutationMu      sync.Mutex
 	mu              sync.RWMutex
 	activeAlerts    []models.Alert
 	lastRebuiltAt   time.Time
@@ -142,6 +147,12 @@ func (a *MonitorAdapter) ResolveCanonicalResourceID(ref string) (string, bool) {
 }
 
 func (a *MonitorAdapter) replaceRegistry(snapshot models.StateSnapshot, recordsBySource map[DataSource][]IngestRecord) {
+	if a == nil {
+		return
+	}
+	a.mutationMu.Lock()
+	defer a.mutationMu.Unlock()
+
 	registry := a.currentRegistry()
 	if registry == nil {
 		return
@@ -360,6 +371,12 @@ func (a *MonitorAdapter) PopulateSnapshotAndSupplemental(snapshot models.StateSn
 // PopulateSupplementalRecords ingests source-native records emitted outside the
 // legacy state snapshot pipeline.
 func (a *MonitorAdapter) PopulateSupplementalRecords(source DataSource, records []IngestRecord) {
+	if a == nil {
+		return
+	}
+	a.mutationMu.Lock()
+	defer a.mutationMu.Unlock()
+
 	registry := a.currentRegistry()
 	if registry == nil || len(records) == 0 || strings.TrimSpace(string(source)) == "" {
 		return

internal/unifiedresources/monitor_adapter_generation_test.go

@@ -0,0 +1,101 @@
+package unifiedresources
+
+import (
+	"sync"
+	"testing"
+	"time"
+
+	"github.com/rcourtman/pulse-go-rewrite/internal/models"
+)
+
+type blockingChangeStore struct {
+	*MemoryStore
+	entered chan struct{}
+	release chan struct{}
+	once    sync.Once
+}
+
+func (s *blockingChangeStore) RecordChange(change ResourceChange) error {
+	s.once.Do(func() {
+		close(s.entered)
+		<-s.release
+	})
+	return s.MemoryStore.RecordChange(change)
+}
+
+func TestMonitorAdapterSerializesSupplementalMutationAfterSnapshotPublication(t *testing.T) {
+	store := &blockingChangeStore{
+		MemoryStore: NewMemoryStore(),
+		entered:     make(chan struct{}),
+		release:     make(chan struct{}),
+	}
+	adapter := NewMonitorAdapter(NewRegistry(store))
+
+	rebuildDone := make(chan struct{})
+	go func() {
+		adapter.PopulateFromSnapshot(models.StateSnapshot{
+			LastUpdate: time.Now().UTC(),
+			VMs: []models.VM{{
+				ID:       "lab:node-a:101",
+				VMID:     101,
+				Name:     "database",
+				Node:     "node-a",
+				Instance: "lab",
+				Status:   "running",
+				LastSeen: time.Now().UTC(),
+			}},
+		})
+		close(rebuildDone)
+	}()
+
+	select {
+	case <-store.entered:
+	case <-time.After(time.Second):
+		t.Fatal("snapshot rebuild did not reach change publication")
+	}
+
+	supplementalDone := make(chan struct{})
+	go func() {
+		adapter.PopulateSupplementalRecords(SourceAgent, []IngestRecord{{
+			SourceID: "host-supplemental",
+			Resource: Resource{
+				Type:     ResourceTypeAgent,
+				Name:     "host-supplemental",
+				Status:   StatusOnline,
+				LastSeen: time.Now().UTC(),
+			},
+		}})
+		close(supplementalDone)
+	}()
+
+	select {
+	case <-supplementalDone:
+		t.Fatal("supplemental mutation bypassed the in-flight snapshot generation")
+	case <-time.After(20 * time.Millisecond):
+	}
+
+	close(store.release)
+	select {
+	case <-rebuildDone:
+	case <-time.After(time.Second):
+		t.Fatal("snapshot rebuild did not complete")
+	}
+	select {
+	case <-supplementalDone:
+	case <-time.After(time.Second):
+		t.Fatal("supplemental mutation did not resume")
+	}
+
+	resources := adapter.GetAll()
+	if len(resources) != 2 {
+		t.Fatalf("final generation contains %d resources, want snapshot plus supplemental record: %+v", len(resources), resources)
+	}
+	var foundVM, foundSupplemental bool
+	for _, resource := range resources {
+		foundVM = foundVM || resource.Name == "database"
+		foundSupplemental = foundSupplemental || resource.Name == "host-supplemental"
+	}
+	if !foundVM || !foundSupplemental {
+		t.Fatalf("final generation lost a writer: %+v", resources)
+	}
+}

internal/unifiedresources/monitor_adapter_read_state_test.go

@@ -1,6 +1,7 @@
 package unifiedresources
 
 import (
+	"fmt"
 	"testing"
 	"time"
 
@@ -525,6 +526,89 @@ func TestMonitorAdapterIngestsAvailabilityAfterCorrelatableSupplementalSources(t
 	}
 }
 
+func TestMonitorAdapterStalenessDoesNotEmitRemovalButAuthoritativeOmissionDoes(t *testing.T) {
+	store := NewMemoryStore()
+	adapter := NewMonitorAdapter(NewRegistry(store))
+	now := time.Now().UTC()
+	container := func(vmid int, name string, seen time.Time) models.Container {
+		return models.Container{
+			ID:       fmt.Sprintf("lab:node-a:%d", vmid),
+			VMID:     vmid,
+			Name:     name,
+			Node:     "node-a",
+			Instance: "lab",
+			Status:   "running",
+			Type:     "lxc",
+			LastSeen: seen,
+		}
+	}
+
+	adapter.PopulateFromSnapshot(models.StateSnapshot{
+		LastUpdate: now,
+		Containers: []models.Container{
+			container(101, "alpha", now),
+			container(102, "beta", now),
+		},
+	})
+	initial := adapter.GetByType(ResourceTypeSystemContainer)
+	if len(initial) != 2 {
+		t.Fatalf("initial container count = %d, want 2", len(initial))
+	}
+	var removedID string
+	for _, resource := range initial {
+		if resource.Name == "beta" {
+			removedID = resource.ID
+		}
+	}
+	if removedID == "" {
+		t.Fatal("beta canonical ID not found")
+	}
+
+	staleSeen := now.Add(-5 * time.Minute)
+	adapter.PopulateFromSnapshot(models.StateSnapshot{
+		LastUpdate: now.Add(time.Second),
+		Containers: []models.Container{
+			container(101, "alpha", staleSeen),
+			container(102, "beta", staleSeen),
+		},
+	})
+	if got := len(adapter.GetByType(ResourceTypeSystemContainer)); got != 2 {
+		t.Fatalf("stale refresh container count = %d, want 2", got)
+	}
+	if changes, err := store.GetRecentChanges(removedID, time.Time{}, 20); err != nil {
+		t.Fatalf("GetRecentChanges before deletion: %v", err)
+	} else {
+		for _, change := range changes {
+			if change.Metadata["changeType"] == "resource_removed" {
+				t.Fatalf("staleness emitted a removal: %+v", change)
+			}
+		}
+	}
+
+	adapter.PopulateFromSnapshot(models.StateSnapshot{
+		LastUpdate: now.Add(2 * time.Second),
+		Containers: []models.Container{
+			container(101, "alpha", now.Add(2*time.Second)),
+		},
+	})
+	if got := len(adapter.GetByType(ResourceTypeSystemContainer)); got != 1 {
+		t.Fatalf("authoritative deletion container count = %d, want 1", got)
+	}
+	changes, err := store.GetRecentChanges(removedID, time.Time{}, 20)
+	if err != nil {
+		t.Fatalf("GetRecentChanges after deletion: %v", err)
+	}
+	removals := 0
+	for _, change := range changes {
+		if change.Metadata["changeType"] == "resource_removed" {
+			removals++
+		}
+	}
+	if removals != 1 {
+		t.Fatalf("resource removal history count = %d, want 1: %+v", removals, changes)
+	}
+}
+
 func TestMonitorAdapterRecordChangeForwardsToStore(t *testing.T) {
 	store := NewMemoryStore()
 	adapter := NewMonitorAdapter(NewRegistry(store))

internal/unifiedresources/registry.go

@@ -3372,6 +3372,9 @@ func mergeProxmoxData(existing *ProxmoxData, incoming *ProxmoxData) *ProxmoxData
 	if incoming.HostURL != "" {
 		merged.HostURL = incoming.HostURL
 	}
+	if incoming.RuntimeStatus != "" {
+		merged.RuntimeStatus = incoming.RuntimeStatus
+	}
 	if incoming.VMID != 0 {
 		merged.VMID = incoming.VMID
 	}

internal/unifiedresources/registry_test.go

@@ -5306,6 +5306,71 @@ func TestMarkStaleRecomputesFromRemainingFreshSources(t *testing.T) {
 	}
 }
 
+func TestMarkStaleKeepsProxmoxRuntimeStateIndependentOfAvailabilityFacet(t *testing.T) {
+	rr := NewRegistry(nil)
+	staleSeen := time.Now().UTC().Add(-5 * time.Minute)
+	freshSeen := time.Now().UTC()
+
+	for _, sourceID := range []string{"lab:node-a:101", "lab:node-a:102"} {
+		rr.IngestRecords(SourceProxmox, []IngestRecord{{
+			SourceID: sourceID,
+			Resource: Resource{
+				Type:     ResourceTypeSystemContainer,
+				Name:     sourceID,
+				Status:   StatusOnline,
+				LastSeen: staleSeen,
+				Proxmox:  &ProxmoxData{RuntimeStatus: "running", NodeName: "node-a"},
+			},
+		}})
+	}
+
+	resources := rr.ListByType(ResourceTypeSystemContainer)
+	if len(resources) != 2 {
+		t.Fatalf("expected two Proxmox containers, got %d", len(resources))
+	}
+	var checkedID string
+	for _, resource := range resources {
+		if resource.Name == "lab:node-a:102" {
+			checkedID = resource.ID
+			break
+		}
+	}
+	if checkedID == "" {
+		t.Fatal("availability target container not found")
+	}
+
+	rr.IngestRecords(SourceAvailability, []IngestRecord{{
+		SourceID: "probe-102",
+		Resource: Resource{
+			Type:     ResourceTypeNetworkEndpoint,
+			Name:     "probe-102",
+			Status:   StatusOnline,
+			LastSeen: freshSeen,
+			Availability: &AvailabilityData{
+				TargetID: "probe-102", LinkedResourceID: checkedID,
+				Address: "192.0.2.102", Protocol: "icmp", Enabled: true, Available: true,
+			},
+		},
+	}})
+
+	rr.MarkStale(freshSeen, nil)
+
+	resources = rr.ListByType(ResourceTypeSystemContainer)
+	statuses := make(map[string]ResourceStatus, len(resources))
+	for _, resource := range resources {
+		statuses[resource.Name] = resource.Status
+		if resource.Proxmox == nil || resource.Proxmox.RuntimeStatus != "running" {
+			t.Fatalf("runtime status changed with source freshness for %s: %+v", resource.Name, resource.Proxmox)
+		}
+	}
+	if got := statuses["lab:node-a:101"]; got != StatusWarning {
+		t.Fatalf("unfaceted stale container status = %q, want warning", got)
+	}
+	if got := statuses["lab:node-a:102"]; got != StatusOnline {
+		t.Fatalf("availability-faceted stale container status = %q, want online", got)
+	}
+}
+
 func TestResourceRegistryUsesConfiguredProxmoxStaleThresholds(t *testing.T) {
 	seen := time.Now().UTC().Add(-90 * time.Second).Truncate(time.Millisecond)
 	snapshot := models.StateSnapshot{

internal/unifiedresources/types.go

@@ -331,7 +331,8 @@ type MetricValue struct {
 
 // ProxmoxData contains Proxmox-specific data for a resource.
 type ProxmoxData struct {
-	SourceID                     string              `json:"sourceId,omitempty"` // raw model ID from source snapshot
+	SourceID                     string              `json:"sourceId,omitempty"`      // raw model ID from source snapshot
+	RuntimeStatus                string              `json:"runtimeStatus,omitempty"` // source-authored VM/LXC power state, independent of collection freshness
 	NodeName                     string              `json:"nodeName,omitempty"`
 	Pool                         string              `json:"pool,omitempty"`
 	ClusterName                  string              `json:"clusterName,omitempty"`

internal/unifiedresources/views.go

@@ -97,6 +97,13 @@ func (v VMView) Status() ResourceStatus {
 	return v.r.Status
 }
 
+func (v VMView) RuntimeStatus() string {
+	if v.r == nil || v.r.Proxmox == nil {
+		return ""
+	}
+	return strings.TrimSpace(v.r.Proxmox.RuntimeStatus)
+}
+
 func (v VMView) VMID() int {
 	if v.r == nil || v.r.Proxmox == nil {
 		return 0
@@ -375,6 +382,13 @@ func (v ContainerView) Status() ResourceStatus {
 	return v.r.Status
 }
 
+func (v ContainerView) RuntimeStatus() string {
+	if v.r == nil || v.r.Proxmox == nil {
+		return ""
+	}
+	return strings.TrimSpace(v.r.Proxmox.RuntimeStatus)
+}
+
 func (v ContainerView) VMID() int {
 	if v.r == nil || v.r.Proxmox == nil {
 		return 0

tests/integration/tests/64-workloads-proxmox-refresh-stability.spec.ts

@@ -1,7 +1,12 @@
 import fs from "node:fs";
 import path from "node:path";
 import { fileURLToPath } from "node:url";
-import { expect, test as base, type Locator, type Page } from "@playwright/test";
+import {
+  expect,
+  test as base,
+  type Locator,
+  type Page,
+} from "@playwright/test";
 
 import {
   createAuthenticatedStorageState,
@@ -113,6 +118,49 @@ async function readGuestDrawerActiveTab(detailRow: Locator): Promise<string> {
   return active;
 }
 
+function issue1611Container(
+  vmid: number,
+  name: string,
+  status: "online" | "warning",
+  availability = false,
+) {
+  return {
+    id: `lab-node-a-${vmid}`,
+    type: "system-container",
+    name,
+    status,
+    lastSeen: "2026-07-24T08:00:00Z",
+    vmid,
+    node: "node-a",
+    instance: "lab",
+    sources: ["proxmox", ...(availability ? ["availability"] : [])],
+    platformScopes: ["proxmox-pve"],
+    metrics: {
+      cpu: { percent: 0.12 },
+      memory: { used: 1024, total: 4096, percent: 25 },
+      disk: { used: 2048, total: 8192, percent: 25 },
+    },
+    proxmox: {
+      runtimeStatus: "running",
+      nodeName: "node-a",
+      instance: "lab",
+      vmid,
+      cpus: 2,
+      uptime: 3600,
+    },
+    ...(availability
+      ? {
+          availability: {
+            targetId: `probe-${vmid}`,
+            protocol: "icmp",
+            enabled: true,
+            available: true,
+          },
+        }
+      : {}),
+  };
+}
+
 test.describe.serial("Workloads Proxmox refresh stability", () => {
   test.setTimeout(180_000);
 
@@ -172,7 +220,9 @@ test.describe.serial("Workloads Proxmox refresh stability", () => {
     await expect(discoveryButton).toBeVisible();
     await discoveryButton.click();
 
-    await expect.poll(() => readGuestDrawerActiveTab(detailRow)).toBe("discovery");
+    await expect
+      .poll(() => readGuestDrawerActiveTab(detailRow))
+      .toBe("discovery");
 
     const beforePollScrollTop = await readPrimaryViewportScrollTop(page);
 
@@ -188,4 +238,95 @@ test.describe.serial("Workloads Proxmox refresh stability", () => {
       Math.max(10, beforePollScrollTop - 80),
     );
   });
+
+  test("retains running LXC rows through stale availability projections and removes a confirmed deletion", async ({
+    page,
+  }, testInfo) => {
+    test.skip(
+      testInfo.project.name.startsWith("mobile-"),
+      "Desktop-only workload refresh proof",
+    );
+
+    await ensureMockModeEnabled(page);
+
+    let workloadRequests = 0;
+    let staleProjectionResponses = 0;
+    let deletionResponses = 0;
+    let publishDeletion = false;
+    await page.route("**/api/resources?**", async (route) => {
+      const url = new URL(route.request().url());
+      if (
+        url.pathname !== "/api/resources" ||
+        url.searchParams.get("type") !== "vm,system-container,app-container,pod"
+      ) {
+        await route.continue();
+        return;
+      }
+
+      workloadRequests += 1;
+      let data;
+      if (workloadRequests === 1) {
+        data = [
+          issue1611Container(101, "lxc-alpha", "online"),
+          issue1611Container(102, "lxc-beta", "online", true),
+          issue1611Container(103, "lxc-gamma", "online"),
+        ];
+      } else if (!publishDeletion) {
+        staleProjectionResponses += 1;
+        data = [
+          issue1611Container(101, "lxc-alpha", "warning"),
+          issue1611Container(102, "lxc-beta", "online", true),
+          issue1611Container(103, "lxc-gamma", "warning"),
+        ];
+      } else {
+        deletionResponses += 1;
+        data = [
+          issue1611Container(101, "lxc-alpha", "online"),
+          issue1611Container(102, "lxc-beta", "online", true),
+        ];
+      }
+
+      await route.fulfill({
+        status: 200,
+        contentType: "application/json",
+        body: JSON.stringify({
+          data,
+          meta: { page: 1, limit: 200, total: data.length, totalPages: 1 },
+        }),
+      });
+    });
+
+    await page.goto(
+      "/proxmox/workloads?type=system-container&platform=proxmox-pve&status=running",
+      { waitUntil: "domcontentloaded" },
+    );
+
+    const rows = page.locator("tr[data-guest-id]");
+    await expect(rows).toHaveCount(3, { timeout: 60_000 });
+    await expect(rows.first()).toBeVisible();
+
+    await page.locator("th").filter({ hasText: "Name" }).last().click();
+    const retainedRow = rows.filter({ hasText: "lxc-beta" });
+    await retainedRow.click();
+    const detailRow = page.locator(
+      'tr[data-inline-detail-for="lab:node-a:102"]',
+    );
+    await expect(detailRow).toBeVisible();
+
+    await expect
+      .poll(() => staleProjectionResponses, { timeout: 15_000 })
+      .toBeGreaterThan(0);
+    await expect(rows).toHaveCount(3);
+    await expect(rows.filter({ hasText: "lxc-alpha" })).toBeVisible();
+    await expect(rows.filter({ hasText: "lxc-gamma" })).toBeVisible();
+    await expect(detailRow).toBeVisible();
+
+    publishDeletion = true;
+    await expect
+      .poll(() => deletionResponses, { timeout: 15_000 })
+      .toBeGreaterThan(0);
+    await expect(rows).toHaveCount(2);
+    await expect(rows.filter({ hasText: "lxc-gamma" })).toHaveCount(0);
+    await expect(detailRow).toBeVisible();
+  });
 });