Skip to content

armgit / Pulse

Public Imported from GitHub
0

Real-time monitoring dashboard for Proxmox VE, PBS, Docker, Kubernetes, TrueNAS and vSphere. Self-hosted, with alerts for failed backups, full disks, unhealthy containers and offline nodes.

Compose scoped backup activity with protected main

pulse-triage[bot]6cc313502d71
diff --cc docs/release-control/v6/internal/subsystems/storage-recovery.md
index e643c3ca8a,a01da392fa..7c27558b45
--- a/docs/release-control/v6/internal/subsystems/storage-recovery.md
+++ b/docs/release-control/v6/internal/subsystems/storage-recovery.md
@@@ -21,63 -21,24 +21,81 @@@
## Purpose
+### Disabled local backup targets keep their own unread scope
+
+A skipped local backup-content read cannot be covered by a successful
+same-named disk on another PVE node. Known live files retain their exact read
+origin, identity and dates; durable points retain their prior evidence and an
+incomplete provider observation, including when the live cache is empty.
+Accepted sibling rows still update or disappear on an accepted empty read.
+Disabled content endpoints are never queried to recover membership.
+
+PVE also lists a node-restricted local target as disabled on excluded nodes.
+Only for skipped local targets with an accepted same-name peer, one ordinary
+bounded datacenter storage-catalogue read resolves that ambiguity. An explicit
+`nodes` restriction can establish exclusion; matching names, missing/unrestricted
+configuration and failed config reads cannot. Those unknown scopes stay unread,
+not permission-denied or confirmed absent. Shared/inherently remote storage
+retains its existing accepted-peer semantics. A cancelled catalogue read cannot
+publish new inventory or provider evidence. Normal re-enable and complete reads
+still establish absence and recovery. No public fields, identifiers, freshness
+policy, access grants or guest/backup/recovery operations change.
+
+`pve_disabled_local_storage_test.go` covers VM/LXC subjects, both node orders,
+warm/cold live inventory, repeated skips, empty readable peers, unavailable or
+denied configuration, explicit node restrictions, shared controls, cancellation,
+connection isolation, all-disabled reads and ordinary recovery. Its normal-suite
+lifecycle optionally exports actual poller/ingest/posture frames for production
+Backups/API-hook desktop/narrow browser proof. This is monitoring correctness,
+not archive availability, installed guest/filesystem recovery or a diagnosis of
+#3248's original stale dates.
+
+### PBS unread namespaces retain only their own last accepted live files
+
+When a PBS group-list request fails transiently, its previous live snapshot
+rows remain inspectable even if a sibling namespace succeeds. A transient
+namespace-discovery failure similarly retains known but unqueried namespaces,
+without guessing or issuing hidden-namespace requests. Accepted rows take
+priority over carried rows at the existing live-state bound. Carried dates,
+identities and snapshot-cache times are unchanged, and carried rows never enter
+the observed-group batch or renew recovery-point evidence. Provider history
+stays partial/unavailable rather than claiming protection.
+
+An actual access rejection still withdraws that namespace's raw rows, including
+when every namespace failed and another failure was transient. Wholly unread
+mixed failures keep access Unknown rather than inventing universal denial;
+uniformly rejected collection still reports Denied. Failed discovery
+cannot override a namespace request actually made: neither a rejection nor an
+accepted empty listing can be undone by carry-forward. Durable historical facts
+remain qualified by collection/access evidence until a complete accepted scan
+can reconcile absence. Ordinary successful reads restore the inventory. Public
+JSON, identifiers, bounds, policy and native backup/recovery operations are
+unchanged. `pbs_namespace_continuity_test.go` exercises VM/container scopes,
+repeated transient/malformed reads, discovery loss, mixed denial, accepted empty
+siblings, connection isolation and complete recovery/removal. Its optional
+lifecycle frames drive the production Backups consumer and API hooks in
+desktop/narrow browser proof. This is source continuity, not archive availability,
+installed guest recovery or #3248's diagnosed stale-date cause.
+
+
+ ### Backup activity shares the list's inventory scope
+
+ By date activity counts and volume use the same search, node, workload type,
+ exact backup location and source/verification filters as the restore-point
+ list. Unresolved locations and empty matches cannot draw another scope's
+ activity. Day selection is applied afterwards only to the list, so the chart
+ continues to offer all dates in that inventory scope. Unknown-date notices
+ also describe only the matching inventory. Completed-only admission, local
+ calendar buckets, aggregate server-owned posture, query cadence and all
+ backup/recovery operations are unchanged.
+
+ `ProxmoxBackupsTable.test.tsx` exercises every facet, combined filters,
+ case-sensitive/ambiguous locations, empty matches, count/volume, and switching
+ and clearing days/filters without extra reads. Source-bound desktop/narrow
+ browser proof exercises the actual inventory parser/model/chart/list and
+ shared filter controls. This corrects inspection, not #3248's original stale
+ timestamps or native guest recovery.
+
### Backup activity ranges follow local calendar dates
The existing 7/30/90/365-day Proxmox backup activity windows contain exactly