Skip to content

CHANGELOG.md

Latest commit

 

History

History
312 lines (284 loc) · 17.7 KB

CHANGELOG.md

File metadata and controls

312 lines (284 loc) · 17.7 KB
Raw

Changelog

0.7.0

Versionsanzeige im Header und automatische Update-Hinweise.

  • Versions-Badge: Neben „Sun2000 · HmIP" steht jetzt die installierte Version. Ein Klick darauf öffnet die GitHub-Releases-Seite.
  • Update-Erkennung: Ein Hintergrund-Check (alle 6 h, Node-https, keine neue Abhängigkeit) vergleicht die installierte Version mit dem neuesten GitHub-Release. Netz-/Parse-Fehler sind nicht-fatal.
  • Benachrichtigung mit Link: Wird erstmals eine neuere Version erkannt, erscheint eine einmalige Meldung (mit Release-Link) im Notification Center und – sofern aktiviert – per Telegram. Neue, abschaltbare Kategorie plugin-update.
  • Menüpunkt „Update installieren": taucht bei verfügbarem Update als hervorgehobener Link in der Tab-Leiste auf und führt direkt zum GitHub-Release; das Badge hebt sich dann ebenfalls hervor.
  • /api/version liefert jetzt zusätzlich latest, updateAvailable und releaseUrl. Fünf neue Tests (Semver-Vergleich + Checker-Verhalten).

0.6.0

Persistent history, inverter alarms, and a round of polish.

Persistent history & export

  • History survives restarts: the hourly (≤96 h) and daily (≤30 d) tiers are now persisted to /data/history.json and restored on startup, so the Verlauf view is no longer empty after every container restart. Writes are atomic (temp file + fsync + rename) so a crash never leaves a torn file, and every write failure is logged but non-fatal.
  • Export: new CSV and JSON download buttons in the Verlauf tab (/api/history/export.csv|json, LAN-gated, served as attachments). The CSV has a fixed, stable column header.
  • Restore is resilient: a missing, corrupt, or unknown-version file simply starts empty; out-of-window and malformed entries are pruned/skipped; restored hourly buckets are de-duplicated so the current hour never doubles up.

Inverter alarms

  • Sun2000 alarm registers 32008/32009/32010 are now read and decoded into a named, severity-classified alarm list (unknown bits degrade to a generic alarm-<addr>-bit<n> identifier rather than being dropped).
  • New inverter-alarm notification category: edge-triggered (one message per newly-active alarm, nothing while it stays active, re-armed after it clears), with critical/warning severity mapping. Enabled by default.
  • Diagnose tab gains an "Aktive Alarme" card.

Polish

  • Login rate limiting: per-IP failed-login throttling on /api/admin/login (default 5 attempts / 15 min) — a blocked IP gets 429 + Retry-After and the password is never even evaluated. A successful login resets the counter.
  • Config backup/restore: admin+LAN endpoints to download the full config and restore it from a file (validated, merged over defaults; invalid documents are rejected with the running config left unchanged). The backup contains plaintext secrets by design — keep it safe.
  • Theme: initial theme now honours the OS prefers-color-scheme when no preference is stored.
  • i18n: the translation table is checked for bidirectional key parity; the table now lives in its own i18n.js so it is unit-testable.
  • PWA: a static manifest.webmanifest makes the dashboard installable. No service worker is registered, so the no-cache contract is preserved.

Tooling

  • Added 18 fast-check property tests (history round-trip/retention/bounds/ resilience/raw-window/no-duplicate/CSV header, alarm decode, alarm edge-trigger/ gate/severity, rate-limit threshold/per-IP, config restore round-trip/defaults/ invalid, i18n parity, initial theme) plus example tests for the store I/O.

0.5.0

Configurable notifications: Telegram delivery + an in-dashboard Notification Center.

Notifications

  • Configurable catalog: each category can be enabled/disabled with a minimum severity — connection/standby transitions, Modbus errors and reconnect lockdown, HCU WebSocket connect/disconnect, battery SOC low/full, daily energy milestones, power peaks, and device-status changes.
  • Configurable thresholds: low/full battery SOC, daily energy milestone increment, power-peak level (validated on save).
  • Grouping/batching: related events within a configurable window are coalesced into a single digest message instead of many; critical events flush immediately, carrying along everything already collected.
  • Quiet hours & rate limiting: during quiet hours only critical digests are delivered immediately, the rest are deferred and sent afterwards; once the rate limit is hit, further digests are coalesced and delivered (never dropped).
  • Telegram channel: delivery via the Telegram Bot HTTP API using Node's built-in https (no Python, no new runtime), with bounded exponential-backoff retry that honours a 429 retry_after. A "Telegram testen" button verifies setup. The bot token is redacted like cloudPassword/adminPassword and never written to logs.
  • Notification Center: a new "Meldungen" tab lists unread events grouped by category with mark-as-read / mark-all-read (admin-gated), and an unread badge on the tab that updates live via the existing SSE snapshot.

Architecture & tooling

  • New additive subsystem under src/notifications/ (detector, grouping, dispatcher, telegram, store, format, facade) wired via passive subscriptions to the poller and HCU — existing flows are untouched.
  • New /api/notifications* endpoints behind the LAN gate + requireAdmin.
  • Config persists the full notifications block in /data/config.json with a deep-merge so absent keys fall back to defaults; hot-reloaded without restart.
  • Added a fast-check property-test suite covering 20 correctness properties (digest completeness, critical immediate-flush, unread-count invariant, store bound, severity filtering, Telegram eligibility, quiet-hours routing, rate-limit coalescing without loss, SOC/milestone edges, bounded retries, token-never-logged, redaction and config round-trips).

0.4.1

  • Admin login was unreachable: when an admin password was set, there was no visible field to enter it — unlocking relied solely on the header lock, which only prompted when the client's cached adminProtected flag happened to be set. Added a dedicated Admin-Modus card at the top of the Config tab (password field + Anmelden/Abmelden, with a live status line). Also made ensureAdmin() robust: a 403 from the server now always prompts for the password and retries, even if the cached state was stale.

0.4.0

Security and UI overhaul.

Security

  • LAN-only access: the dashboard and its API now reject requests from non-private source IPs (lanOnly, default on). This reliably blocks access from the internet even if the HCU's port gets forwarded. An optional allowedSubnets CIDR allowlist (e.g. 192.168.10.0/24) narrows it further. /healthz stays open for the HCU's own probing. (NAT caveat: behind the HCU's port mapping the source IP may be the bridge gateway, so the practical guarantee is "block non-private", which is the property that matters.)
  • Admin mode: every write — battery/charge control, register writes, config changes, resets, slave-id probe — now requires an authenticated admin session. Set adminPassword for real protection; when empty, admin mode is a soft guard against accidental writes. Login issues a 2 h token; secrets are redacted in /api/config.

Reliability

  • Standby instead of red: when the TCP link is up but the inverter is asleep (night mode), the status pill shows amber "Wechselrichter im Standby" and /healthz returns 200 — so the HCU no longer restart-loops overnight.
  • Block-read fallback: if one unsupported/de-energised register fails an atomic block read (typical for PV strings at dusk) but the inverter is otherwise responsive, the block is retried register-by-register so the readable fields still come through.
  • Accurate daily energy sourced from the inverter's own counters (daily yield, battery daily, meter deltas) rather than integrated power.

UI / UX

  • Chart hover tooltips with crosshair and per-series values.
  • Smart W/kW unit formatting.
  • New Verlauf tab visualising the long-term hourly/daily aggregates.
  • Autarky donut, KPI sparklines, "Energie heute" card.
  • Tab bar wraps on narrow screens; favicon; live "vor X s" ticker; loading state; keyboard-focus and ARIA tweaks.

Tooling

  • ESLint (flat config) and a node:test suite (decode/encode, history energy, Modbus reconnect state machine, server access/admin wiring).
  • Backend performance: SSE payload is skipped when no client is connected; self-sufficiency is computed once per snapshot instead of per request.

0.3.10

  • Critical: debug dashboard was completely broken. A stray ASCII double quote inside a German confirm() string (the "reset device identity" button introduced in 0.3.7) terminated the string early, causing a syntax error in app.js. The whole script aborted before buildTabs()/render() ran, so the dashboard showed only the base layout — no tabs, no data, and a permanently red status dot (the CSS default, not a real status). Fixed by escaping the quote. app.js and chart.js are now part of npm run lint (node --check), which would have caught this.
  • Model string NUL: decodeString now cuts at the first NUL byte, so the inverter model reads as SUN2000-8KTL-M1 instead of SUN2000-8KTL-M1\u000001074314-006. The clean value is what gets sent to the HCU as modelType and shown on the dashboard.
  • Tab bar: wraps onto multiple rows on narrow screens instead of scrolling the last tabs (Logs, Diagnose) off the edge with no visible scrollbar.

0.3.9

  • Modbus reconnect protection fixed (real bug): the escalating cooldown after a socket closed by peer and the 10-minute lockdown were dead code — they referenced an undefined constant (PEER_CLOSE_COOLDOWN_MS) and the lockdown gate was never read. Under "use strict" this threw on the first reconnect after a peer-close, so the SDongle's rate-limiter was never actually respected. Now the cooldown escalates (30s → 60s → 120s …, capped at 5 min), the lockdown is enforced (no connect attempts for 10 min after repeated peer-closes), and both reset on the first successful read.
  • STATUS_EVENT only on real change: device state is no longer re-sent to the HCU every poll when nothing changed. Cuts HCU traffic during flat periods (night/idle) and stops re-asserting controllable devices (the force-charge switch), which is the documented anti-pattern. Inclusion and post-control updates still force an emit.
  • Tiered, bounded data retention: history keeps full 10s resolution for 6h, hourly aggregates (avg/min/max + integrated energy) up to 96h, then one condensed summary per day for ~30 days. Logs and the HCU message log keep everything for 96h and are reduced to the essentials (warnings/errors and non-routine messages) beyond that. Memory stays bounded regardless of uptime. New GET /api/history/aggregate exposes the long-term tiers.
  • Cleanup: removed dead readMany code; npm run lint now covers all Node-side modules (including modbus.js, which was previously unchecked).

0.3.8

  • Dashboard no-cache: the dashboard HTML/JS/CSS were served with a 1-hour cache header. After a plugin update the browser kept serving the old bundle, which called the newer backend API and showed only empty values (and a stale "v0.2" footer). Assets are now served with Cache-Control: no-cache, must-revalidate so a normal reload always picks up the running version.
  • Live version in footer: new GET /api/version endpoint; the dashboard footer now shows the actually running plugin version instead of a hard-coded string. Makes "am I on the new build?" obvious.
  • Quiet nights: Modbus exception 4 (Slave device failure) and exception 2 (Illegal data address) on PV/AC registers are the inverter's normal response while shutting down for the night. They are now treated like an "inverter asleep" timeout (throttled info) instead of a hard error, so the log stays clean overnight.

0.3.7

  • Dashboard → Konfig: zwei neue Wartungs-Knöpfe.
    • „Geräte-Identität zurücksetzen" löscht nur die persistierte Inverter-SN. Der Rest der Konfig bleibt. Beim nächsten erfolgreichen Modbus-Read wird eine frische SN gespeichert. Sinnvoll wenn man die HmIP-Geräte komplett neu anlegen lassen will.
    • „Komplett-Reset" schreibt eine /data/.reset_on_next_boot-Marker- Datei und beendet das Plugin nach 2 s. Beim nächsten Start (HCU restartet den Container automatisch) wird /data/config.json gelöscht und das Plugin startet mit Default-Werten. Doppelt bestätigt (Confirm + „RESET" eintippen).
  • API-Endpoints POST /api/config/clear-sn und POST /api/config/reset (letzterer braucht {"confirm":"RESET"} im Body).

0.3.6

  • Soft config updates: changes to non-Modbus settings (dashboard, hardware flags, cloud) no longer tear down the Modbus TCP connection. The previous behavior triggered the SDongle's reconnect rate-limiter, which then RST'd every connection for 10+ minutes after every config save. Only changes to host / port / unit-id reset the poller now.
  • 30 s cooldown after socket closed by peer: the SDongle keeps a rate-limiter on rapid reconnects. When it RSTs us, we now wait 30 s before the next connect attempt instead of immediately reconnecting. This breaks the "connect → RST → connect → RST" loop that filled the logs every 11 s.
  • Static info merge instead of overwrite: a flaky read no longer blanks out the previously known SN/model/FW. The "Inverter: ? SN ? FW ?" log line is gone for good — startup logs only the first successful identification.
  • Logs: identical "closed by peer" warnings throttled to 1×/min.

0.3.5

  • Stable HmIP device IDs across plugin restarts: the inverter serial number is now persisted to /data/config.json (persistedSn) on the first successful Modbus read. Later restarts always use the same value, so HmIP devices no longer get re-registered as duplicates when the very first static read fails (slow Modbus, night mode, firmware update).
  • Config form: new read-only "Used serial number" entry under a new "Device identity" group, so users can verify which SN drives the IDs.
  • Troubleshooting docs (DE + EN) extended:
    • "Comparing values: FusionSolar vs. plugin" — explains why DC vs AC, cloud lag and battery idle noise look like discrepancies.
    • "Duplicate devices in the HmIP app" — explains the pre-0.3.5 bug and how to clean up.
    • "Getting an installer account" — three concrete paths (your installer, self-register, Huawei support).
    • "Firmware update via the FusionSolar web portal" — full step-by-step.

0.3.4

  • Modbus block reads: instead of issuing one request per register, contiguous registers are now read in a single Modbus call. A poll cycle that previously fired ~30 separate requests now fires 4–6, which is much friendlier to the SDongleA-05 when other Modbus masters (FusionSolar cloud sync, HA, etc.) share the inverter.
  • Adaptive poll interval: failures back off exponentially (10s → 20s → 40s → 60s) and reset to base on the next successful read.
  • Diagnose: new "Verbindung stabil" check that detects the typical "socket closed by peer" pattern and points the user at the FusionSolar Modbus-TCP setting ("uneingeschränkt").

0.3.3

  • Modbus: keep TCP socket open when reads time out (typical at night when the inverter is sleeping). Avoids the spammy "Modbus connected" log loop.
  • Modbus: read timeout raised from 4 s to 8 s for slow wake-up responses.
  • Modbus: identical warnings throttled to once per minute (less log noise).
  • Modbus: socket-level errors now drop the connection (and reconnect), but timeouts no longer do.
  • Dashboard → Diagnose: new "Modbus-Statistik" panel (reads / OK / timeouts / errors / writes / last error).
  • Dashboard → Diagnose: TCP probe and Slave-ID probe buttons.
  • Diagnose check now distinguishes between "TCP connected" and "inverter actually answers", with a clear hint pointing at night mode / wrong slave id / blocking master when only TCP is up.

0.3.0

  • Plugin metadata: issuer set to Fabio Renner, GitHub URL and PayPal donation link appended to description rendered in the HCU plugin tile.
  • README.md / README.de.md: plugin icon at the top, GitHub link, PayPal donate form, updated download link.
  • New plugin icon (icon.svg).

0.2.0

  • Massively expanded debug dashboard:
    • Server-Sent Events (SSE) for instant live updates.
    • 11 tabs: Overview, Live, Inverter, Battery, Grid, Control, Modbus, HCU, Config, Logs, Diagnostics.
    • Animated SVG energy-flow diagram.
    • Vanilla-canvas multi-series line charts (no external deps).
    • Light/dark theme toggle, DE/EN UI.
    • Per-string PV view, per-phase grid view, battery SOC gauge.
    • Control panel for storage mode, max charge/discharge, target SOC, active power limit.
    • Register browser with search/filter/hex toggle, CSV export, raw-range scanner.
    • HCU message log viewer.
    • Today's peaks (PV, house, import/export, charge/discharge, SOC min/max), self-sufficiency %.
  • 6 h history ring buffer (10 s resolution).
  • Additional Modbus registers: PV strings 1–4 (V/A), inverter per-phase voltage/current.

0.1.0

  • Initial release.
  • Modbus TCP polling of Sun2000 inverter, LUNA2000 battery, DTSU666-H smart meter.
  • HCU Connect API integration: virtual INVERTER, BATTERY, GRID_CONNECTION_POINT, ENERGY_METER devices.
  • Optional virtual SWITCH devices for battery forced-charge / forced-discharge.
  • Optional FusionSolar cloud fallback (read-only).
  • HCUweb config page with grouped properties.
  • Local debug dashboard on port 8088 with live values, register browser and write-register tool.