SIGNALS Docs
Exit Reader View
100%
Theme

Reader View

Rental operations

Read-only rental lifecycle checks and observed conflict trends.

Overview

Rental operations runs read-only lifecycle checks across rentals and shows observed conflict trends. It never reprices, repairs, promotes or retries a rental.

Where it lives

Route: /admin/settings/rental-operations

Open Admin → System → Rental operations. It is also linked from System Health on self-hosted installs. Requires settings.manage and access to every warehouse; on Signals Cloud a fully selected tenant is also required.

Checks

The page shows when it last checked and how many rentals were included (Checked N of M rentals). If only a sample was checked, a note says other rentals may be omitted. Refresh runs the checks again.

Card What it counts
Duplicate quote versions Rentals with duplicate version numbers or more than one active version
Status and evidence Open orders whose status does not match their recorded fulfilment phase
Stored totals Rentals whose stored headline totals do not match a recalculation
Recorded handoff backlog Attachment handoffs not yet completed, plus invoice deliveries that are pending, retrying or failed
Conflict-rate regression Whether the observed conflict rate rose, comparing adjacent 15-minute windows

Each card carries a status: Attention required, No observed findings, Incomplete, or Unavailable (for example, when there is no baseline to compare against). Availability queue state is not observed, and alerts go to the local application log.

Dashboard details

Operators with settings management and unrestricted warehouse access can open Rental operations in the system settings or command palette, and from System Health on self-hosted installs. Cloud callers also require a real bound tenant whose database matches the selected connection. The panel reads duplicate-version preflight, the installed fulfilment phase, stored headline integrity, recorded attachment and invoice handoffs, and observed invocation conflicts. It never reprices, repairs, promotes or retries a rental.

Status, totals and invoice checks inspect the first 25 current rentals, with a 1,000-child-row cap for phase/headline checks. Sampled or indeterminate results are incomplete; unavailable readers show no count. Duplicate-version preflight and recorded attachment backlog use the selected database. Availability queue state and other unrecorded handoffs are not observed, so the panel makes no universal handoff-health claim.

Conflict regression compares adjacent 15-minute invocation windows, requiring at least 20 observations in each and an increase of 10 percentage points. Missing or unreadable windows are unavailable. Counters and one warning per family per 15-minute window use tenant-qualified cache keys. Refreshing the panel emits warnings for observed findings; it does not schedule background inspections. Warning records contain only the family and count and are written to the application log.

Rental lifecycle diagnostics

Operators with settings management, action-log view and rental view permissions can open Rental lifecycle diagnostics from the rental management menu. Warehouse and tenant access apply on every inspection. The view links the latest 50 recorded event identifiers, stable rental event types and times to the current aggregate version, tree revision and active quote version. Only recognized actor kinds and UUID correlation identifiers are shown. Unknown metadata is unavailable; event payloads, actor identities, commercial fields and snapshot bodies are excluded.

Snapshot event pointers are recorded separately. A matching pointer does not prove all downstream effects completed, and revision tokens are not event counts. A concurrent revision change asks the operator to refresh. Attachment and invoice statuses come only from installed receipts; invoice permissions still apply. Availability and webhook completion remain unavailable because this reader has no per-rental completion receipt. Missing evidence never means completed. History and handoff lists are bounded, with explicit additional-history notices. Refresh only reads and rechecks authorization; it does not replay, repair, retry or enqueue work.

Stored totals verification

Operators can inspect stored headline and historical pricing-manifest consistency with signals:rentals:verify-totals. It is read-only by default; explicit apply restores valid captured projections and records an atomic reason audit. Run the command with --help for scope, chunking and resume options.

Recovery after a committed movement

A stale availability view or missing notification does not mean that dispatch or return failed. First check the rental's assignment or bulk quantities, action log and original fulfilment command receipt, where a command key was supplied. Record the rental, affected catalogue item and warehouse, command key, effect job ID and any delivery log. Preserve the committed movement: do not repeat it with a new key or reverse it to repair a downstream effect.

For a failed queue job, use Settings → System health to inspect and retry the individual failed job, or inspect php artisan queue:failed in the correct tenant context and use php artisan queue:retry <failed-job-id>. Identify the effect and target, then correct the worker, queue or receiver fault before retrying. Avoid retrying every failed job. A repeated fulfilment command returns its original receipt; it is not a downstream-effect recovery command.

Incomplete effect Recovery and verification
Availability recalculation failed Retry the existing RecalculateAvailabilityJob for the affected catalogue-item/warehouse pair. It reads current stock and demand, so later committed returns remain authoritative. Check the resulting availability slots and the worker result.
Availability job is lost or snapshots are absent A push that never reached the queue has no failed-job record to retry. This release has no targeted operator screen or Artisan command for RebuildSnapshotsJob; its production callers are qualifying rental actions, including later returns. Escalate a lost job to maintenance rather than making another stock movement. The underlying job can upsert a bounded window without changing movement history, but a later-return rebuild does not establish operator recovery without another write. It does not broadcast availability or resend notifications. An integrity sweep that samples existing snapshots cannot guarantee recovery of a wholly absent pair.
Webhook enqueue failed Keep the existing pending outbox row and payload. Restore queue service; the scheduled outbox drain runs every five minutes, with a grace period and receiver backoff. Check that the row was dispatched and inspect WebhookLog.delivered_at and the response; dispatched alone does not prove receiver success.
Webhook or recipient delivery failed Retry only its existing eligible delivery job, preserving the original outbox payload or recipient context. Webhook delivery has six attempts; exhausted jobs dead-letter the outbox row. SendNotification retries recipients independently; check the recipient's communication log and delivery record. Previously successful recipients do not need a new domain notification.
Webhook is dead-lettered, privacy-fenced or disabled Stop automatic redelivery and investigate the recorded reason. Dead-lettered or privacy-redacted outbox rows are refused even by an original serialized delivery job. Re-enable or arrange an explicitly reviewed redelivery only when the current policy permits it; do not clear the fence or republish the stock movement.

Delivery is not exactly once. A receiver may have accepted a request before its acknowledgement was lost. Reconcile its receipt before a manual resend, and use its supported deduplication mechanism where available. Keep prior failed attempts and the recovery result in the incident record; do not edit movement, receipt or audit rows to make the effect appear successful.

Repairing lifecycle projections

Use php artisan signals:rentals:repair-status --rental=<id> --actor=<user-id> to preview a status correction from recorded lifecycle and fulfilment events. Use signals:rentals:repair-demands with the same selectors to preview demand activation against the recorded active version and availability history. Both commands default to dry run and return the proposed field changes as JSON. The actor must be active, have rental edit permission and access the rental's warehouse; Cloud callers must first select their tenant.

Omit --rental to preview a page, optionally setting --limit=1..100 (default 50) and --after=<last-rental-id>. Resume using the returned next_cursor while has_more is true. Pages follow rental ID order and include only the operator's warehouses. Archived rentals, issued invoices and accepted pricing manifests are report-only; they remain unchanged and the cursor advances past them. A refused rental returns an item with outcome: refused, a safe code and no changes; the page continues and its cursor advances past that rental. Batch apply is unavailable.

To apply a preview, add --repair --expected-version=<aggregate-version> --reason="<operator reason>" and confirm the prompt. --force skips only confirmation. Non-local repairs require maintenance mode and verified queue quiescence. A changed rental version, incomplete or conflicting history, unsupported status, or an existing projection repair lock refuses the operation without a partial repair. Check the rental's recorded history before retrying.

These commands restore projection fields, without replaying stock movements, lifecycle handlers, invoices or financial effects. Status repair does not infer a missing lifecycle event from today's projections. Demand repair changes only existing rows' active flags, preserves their captured quantities and windows, and schedules the normal availability recalculation after commit. It does not recreate missing demand rows or recalculate historical pricing or tenant policy. Each changed rental receives one audit recording the operator, reason and changed fields in the repair transaction; an audit failure rolls back the repair. A second successful repair reports no changes and creates no audit.

Checking rental data before an upgrade

Before upgrading an installation, run signals:rentals:version-constraints --json and signals:rentals:lifecycle-preflight --json against the selected installation or tenant. Duplicate version numbers, multiple active versions, unknown lifecycle definitions or project-tree conflicts are reported with their IDs; resolve them before migrating rather than renumbering versions or inventing events. The rental-version constraint migration runs the same check inline and refuses to add its constraints while conflicting data remains. After migrating, signals:rentals:version-constraints --require-installed --json succeeds only when the constraints are in place. On Cloud, central and config-only contexts are not a selected tenant, and each tenant is checked separately.

If a repair is refused, record the refusal code and IDs, re-read the rental and preview again before retrying. Held records and source history are preserved; a refusal is never permission to overwrite projections or replay customer effects.