Chart Fenris’s persistent TUI monitoring redesign #1
Assignees
xavierk (Xavier Karma)
Clear assignees
xavierk (Xavier Karma)
No Assignees
Notifications
Due Date
No due date set.
Dependencies
No dependencies set.
Reference: xavierk/Fenris#1
Reference in New Issue
Block a user
Destination
Produce an implementation-ready specification for replacing Fenris’s HTML dashboard with a keyboard-first TUI backed by an independent systemd collector, persistent compact observation storage, and a usage-adjusted theoretical lifespan whose projection confidence reflects the length, completeness, and stability of real usage history.
Notes
Domain: NVMe usage observation and endurance projection. Read
CONTEXT.md,docs/agents/domain.md, anddocs/agents/issue-tracker.md. Consultgrillinganddomain-modelingfor decision tickets; also consultresearchorprototypewhen required by the ticket type.Standing preferences: Linux/systemd only; one configured NVMe drive; fully local with no telemetry or vendor-data fetching; retain
statusandsample; use a maintained TUI framework; preserve existing observations through migration. This map plans the redesign and does not implement it.Decisions so far
Evaluate Python TUI frameworks for Fenris: Choose Textual for the prototype, contingent on raising Fenris to Python 3.9+, with Urwid as the low-dependency and terminal-control fallback.
Verify NVMe endurance signals and projection constraints: Project baseline consumption from DUW and wall-clock habits, prefer verified model-specific TBW, and disclose coarse vendor wear plus categorical evidence (research context).
Verify systemd lifecycle and privilege constraints — Use a system timer and short-lived privileged collector, keep the TUI unprivileged, and mediate explicit startup toggles through polkit; research context.
Define the persistent observation store and legacy migration: One SQLite database in WAL mode at
/var/lib/fenris/observations.db, root-written and group-read; entities for samples, hour observations, day aggregates, monitoring periods, controller segments, and the endurance baseline; 14-day raw retention; single-transaction idempotent migration fromhistory.jsonlas sole authority (ADR 0001).Define the lifespan projection and confidence model: One projection from the precedence-chosen baseline with Percentage Used as context only; headline rate from the most recent sustained regime (full history ≤90 days by default) with 7/28/90-day scenario ranges; UTC-day evidence gates and four-state categorical confidence with contributing facts (ADR 0002).
Define the collector, service, and CLI lifecycle: Timer-driven oneshot collector on a 5-minute
OnUnitInactiveSeccadence, one-key/etc/fenrisconfig, a polkit-guardedfenris-monitorhelper owning pause/resume and monitoring-period rows (amending the systemd decision's direct-systemctl toggle), read-onlystatus, helper-mediatedsample, retiredstart/stop/fenris.sh, shared freshness constants (ADR 0003).Prototype the TUI information architecture: Variant A “Panes” — one dense keyboard-first screen with confidence-as-evidence and four separate service facts; pause/resume asymmetry and Textual tty passthrough validated live (prototype branch).
Define installation, upgrade, and removal behavior: make install/upgrade/uninstall/purge over an /opt/fenris venv with a file manifest — dormant install with sanctioned-toggle opt-in, installer-run history.jsonl import, forward-only schema migrations behind a one-generation backup, exact dependency lockfile, and sanctioned-disable-before-uninstall (ADR 0004).
Define failure and recovery behavior: Refuse invariant-violating writes at the collector, never backfill gaps, treat store faults as visible degradation with human-sanctioned recovery, flat retry with no alerting, and collector-anchored periods for orphaned samples (ADR 0005).
Verify the controller identity that segments observation history: Controller segments key on the normalized, kernel-exposed subsystem NQN — kernel composite then
model|serialas fallbacks, FR as metadata only — with identity change and DUW decrease as independent axes; legacy history imports under a labeled model-scoped legacy identity (research context).Define endurance-baseline provenance and validation: Mandatory provenance is URL, revision, entry date, model, and nominal capacity; one active row with derived verification (machine match or recorded user attestation), an explicitly acknowledged unverified tier, unprivileged entry-time sysfs validation (normalized model containment, ±1% capacity), read-time applicability as a model match against the current controller segment (retained, never auto-deleted), and a polkit-guarded helper
baselineverb (amends ADR 0001 and ADR 0003).Decide controller-segment metadata columns: Segments carry a frozen-at-open, fully nullable metadata snapshot — normalized
subnqn/sn/mn/frplusvid/ssvid/transportand theidentity_degradedflag,cntlidexcluded; legacy segments importmn-only (amends ADR 0001).Define cross-cutting acceptance criteria: Accepted criteria committed at docs/spec/acceptance-criteria.md — subsystem-organized testable statements with A/P/M evidence classes, ADR-traceability only, exhaustive state-matrix and TUI/CLI parity gates, prohibition set, migration interruption gates; SLOT-A (degraded identity) and SLOT-B (acquisition path) await their tickets.
Decide how degraded identity affects projection confidence: Degraded means a blank identity key only; it caps confidence at Limited with the "replacement detection relies on write-counter continuity only" fact, and key changes to or from blank quarantine like any identity change (amends ADR 0002; fills SLOT-A as PR-15/PR-16 + ID-4).
Choose the collector's NVMe acquisition path: Pin to
smartctl -jfor counters plus sysfs for identity — hard pin, no fallback, no partial samples, normalization once at write time; vid/ssvid from the PCI node when present (ADR 0006; fills SLOT-B as AC-1–AC-5).Assemble the implementation-ready specification: One
docs/spec/fenris-redesign.md, data-flow ordered, normatively restating every operative contract while rationale stays in ADRs 0001–0006 cited per section; a two-way ADR↔criterion traceability matrix as appendix gates assembly; one AFK task ticket (Write the Fenris redesign specification and close the map) executes it and closes the map.Not yet specified
Out of scope