legion_kk/docs/open.md

Open Issues

Last checked: 2026-06-17

This file tracks current reminders and unresolved product/engineering work. Completed review notes and stale refactor plans have been removed or squashed into the relevant open item.

Renderer / GUI

• Build the actual observation UI:
  • cloud/catalog status lines with current operation, last-finished timestamp, and last error
  • warning/error log notifications from observation:logs, wired to acknowledgement
  • metric summary components on top of the renderer observation metrics view model
  • node observation display for ready-divergent, provider-missing, degraded services, stale metrics, and log severity counts
  • render ActivityState.deploymentEvents as structured deployment steps with severity/status styling
• Continue consolidating reusable Svelte component primitives whose class/API conventions stay CSS-compatible with the Tribes tribes_ui plugin surface. Avoid Legion-only styling patterns that cannot be mirrored in plugin/HEEx UI.
• Improve the common “one DNS zone plus optional domain delegation” setup flow.
• Consider a reusable dialog wrapper for repeated <dialog> lifecycle patterns.
• Add runtime validation for critical IPC payloads, especially large payloads such as AppSnapshot, when the API surface settles.
• Add list virtualization only if managed resource lists become large enough to need it.

Map view

• Add a map component for planned/active infrastructure geography.
• Evaluate MapLibre GL for rendering.
• Evaluate Planetiler or another offline tile generation path.
• Bundle a low-zoom world map with Legion so the GUI does not depend on a live tile service for basic global/regional placement. High zoom levels are not required initially.

Observation, logs, and telemetry

• Decide whether cloud/catalog observation need backend-owned stale thresholds. Logs and metrics intentionally do not: subscribers can decide whether live events have gone quiet.
• Make cluster log backfill robust for late/out-of-order sync and switching queried cluster nodes:
  • track cursor/watermark per queried cluster node/source
  • prefer inserted_at or updated_at API filters/cursors over occurred_at
  • keep an overlap window and rely on local dedupe
  • add a force-full backfill trigger/CLI when needed
• Keep heavier streams separate: metric rollups for time-series summaries, deployment events for selected durable install/recovery steps, logs for warning/error notifications, and plugin-specific methods for plugin details.

Auto-reconciliation and repair

• Tighten repair detection and auto-heal policies beyond the current conservative safe repairs.
• Cover at least:
  • DNS-derived host inputs and node deployment metadata
  • sync certificate/config material
  • system target/channel generation
  • more precise cluster membership/control-plane state
  • more precise NBDE mode, Tang peer set, and local boot-key state
  • more precise provider firewall/security-group rules
• Decide which additional repair needs are auto-reconciled, which require operator confirmation, and which should block further materialization.

Provider cleanup

• Keep provider cleanup grounded in canonical planned/tracked/actual resources, actor state, operation logs, and billing evidence; do not reintroduce a separate provider recovery sidecar without a concrete ownership gap.
• Extend cleanup and adoption workflows beyond CLI actions:
  • richer resource detail pages with mismatch evidence, last refresh, cleanup/adoption attempts, and billing impact
  • typed confirmation for destructive GUI cleanup when the resource is especially ambiguous
  • resource-specific adoption that promotes safe observed resources into normal tracked state/bindings
• Model related billable resources explicitly as provider observations when needed: server/instance, SSH key, firewall/security group, volume/disk, floating IP, DNS zone, domain/order/cart where applicable.
• On failure, prioritize billable provider destroy first, then best-effort ancillary cleanup. Surface structured cleanup summaries to CLI/GUI.
• Extend cleanup/adoption coverage beyond Hetzner/Scaleway compute to OVH and DNS/domain provisioning, where billing semantics are more sensitive.

Pricing and cost tracking

• Extend the initial provider-neutral billing ledger beyond servers, domains, and observed orphans:
  • provider-specific volume/disk, floating IP, firewall/security-group, DNS zone, domain order/cart, and renewal semantics
  • richer hourly-capped-monthly, annual/fixed, unattached-IP, and stopped/suspended billing rules per provider
  • explicit provider invoice/API comparison when billing APIs are available
• Improve persisted monthly summaries so end-of-month snapshots are finalized even if Legion is not opened exactly at month boundary.
• Promote resource-specific adoption/cleanup actions into the billing ledger so cost and cleanup events share one evidence trail.

Engine / API quality

• Continue migrating deeper provider/planning failures to typed LegionEngineError paths when touched.
• Runtime payload validation remains intentionally limited; revisit for critical IPC/daemon boundaries once contracts stabilize.