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.

Merge published frontier for backup evidence review

pulse-triage[bot]b7b98d1cd0d7
diff --cc docs/release-control/v6/internal/subsystems/storage-recovery.md
index cfbe042c58,8e15447c5f..6b4098c097
--- a/docs/release-control/v6/internal/subsystems/storage-recovery.md
+++ b/docs/release-control/v6/internal/subsystems/storage-recovery.md
@@@ -21,30 -21,30 +21,54 @@@
## Purpose
+### Coverage provider outcomes are not backup job results
+
+Provider evidence identifies the source of each scoped assessment: PBS/PVE
+backup inventories have explicit names; legacy restore-point evidence is named
+as such, and unrecognised sources remain literal text (or Unknown when absent).
+The rollup's `jobState` is displayed as Reported outcome, never Job or Collection:
+an inventory observation can supply it, but a newer retained point can also
+supply the outcome beside an older observation's source. Visible wrapping copy
+keeps this evidence distinct from actual task and restore-point states. History,
+access, reported values, protection, dates and policy do not change. The shipped
+PBS/PVE comparison guides name the same Source/Reported outcome fields and
+keep successful inventory reads separate from backup-task completion. Rendered
+help tests preserve this distinction and every original-recovery safety limit.
+
+Normal-suite `ProxmoxCoverageTable.test.tsx` covers known/unknown/literal sources,
+all reported outcomes and narrow/wide rendering. `ProxmoxBackupsTable.test.tsx`
+passes parsed inventory and posture reads through the consuming surface,
+retaining a Running/unprotected point and a real OK task independently of
+successful or failed inventory evidence, including polling and owned focus.
+Desktop/narrow browser proof must inspect the production Backups consumer with
+source-bound collector frames, its separate task/evidence meanings and readable
+source/outcome/qualifier. This is safer inspection, not backup-job execution,
+original guest recovery, a tested restore or published availability.
+
+ ### Datastore discovery cannot turn unread history into absence
+
+ PBS datastore enumeration uses the same bounded, fully read and closed HTTP 200
+ transport and unambiguous envelope admission as namespace, group, snapshot and
+ writer inventories. Duplicate/case-aliased fields, non-null errors, invalid
+ UTF-8 and partial/other success statuses are unread, not an accepted empty outer
+ scope. Validate the entire enumeration before exclusions or any per-datastore
+ requests. Retain payload-free diagnostics and existing HTML/HTTPS guidance.
+ Explicit empty arrays and unique future metadata remain valid; filtering,
+ status collection, permissions, auth, poll budgets and guest operations do not
+ change.
+
+ `TestDatastoreInventoryRequiresWholeResponse` and the complete-body/valid-scope
+ controls exercise the client, including read/close/bound failures and excluded
+ stores. Normal-suite `TestPBSDatastoreWholeResponseRetainsEvidence` drives the
+ public poll and background worker for VM/CT warm/cold-live states: unread
+ discovery retains raw and durable evidence without renewal or guessed child
+ reads, withdraws protection without claiming an access denial, and ordinary
+ complete/empty reads recover and reconcile absence. The consumer lifecycle
+ exports these exact producer results for desktop/narrow Backups interactions
+ and pixel inspection. Existing discovery denial, skip, other-connection and
+ connectivity-fallback controls remain. No original guest recovery, production
+ occurrence, issue cause or containing release is established by this proof.
+
### PBS backup listings require the whole response
Namespace, group and snapshot reads accept one unambiguous JSON envelope only