Make solismon resilient: never crash, retry until recovered

[dnviti]

Summary

Makes the monitor never crash. Any error is now logged and the process keeps running, actively retrying until it recovers and then continuing its work. Previously, routine transient conditions permanently killed the daemon.

Why

• scrape_solis() called exit(1) on a connect failure or after 3 failed reads. Because from sys import exit raises SystemExit (a BaseException), it bypassed every except Exception and killed the process — so the inverter being briefly unreachable (overnight, Wi-Fi blip, reboot) permanently stopped monitoring.
• The top-level handler also exit(1)'d, which under Home Assistant boot: auto becomes a restart/crash-loop.
• A fresh Modbus socket (+ background reader thread) was opened every cycle and never closed → fd/thread leak that eventually crashes a long-running host.
• SIGTERM (what Docker/HA send to stop) killed the process at the C level with no clean-shutdown path.

Changes

main.py

• Remove all exit()/SystemExit paths; scrape_solis() logs and returns True/False. Bounded read retries return False (no exit, no hot-loop, no stale/unbound regs).
• Run scraping in a background daemon thread (scrape_loop) that retries with exponential backoff capped at CHECK_INTERVAL on failure and resets to normal cadence on success — recovers fast, never hangs or hammers, and can never die.
• Guarantee Modbus teardown in a finally (fixes the leak) and always disconnect the MQTT client.
• Build metrics into a local dict and publish atomically on success, so a failed cycle keeps the last-good snapshot (and avoids a dict-resize race).
• Prometheus collect() now only serves the latest snapshot (instant, never blocks/raises); exporter startup retries a port-bind failure instead of exiting.
• Guard timestamp parsing and register indexing against malformed frames.
• Install SIGTERM/SIGINT handlers for clean, logged shutdown.

config/config.py

• Config parsing is total: a bad env var / option logs a warning and falls back to its default instead of raising at import time.
• Clamp CHECK_INTERVAL to >= 1 (prevents sleep(-30) crashing and 0 busy-looping).

Behavior change

Prometheus mode now polls the inverter every CHECK_INTERVAL in a background thread (and /metrics serves the latest snapshot instantly), instead of polling only when scraped. This is what enables retry-until-recovered without hanging HTTP requests.

Verification

• Unit tests: connect failure, persistent read failure, short/extra register frames, MQTT broker down, missing modified-metric inputs, and the full backoff loop (grow → cap → reset → never-raise → clean stop) — all pass.
• Live runs of main.py with a mock inverter that fails twice then recovers: it retried, recovered, kept publishing, and on SIGTERM exited 0 with no traceback — in both MQTT and Prometheus modes.

🤖 Generated with Claude Code

[Copilot]

Copilot encountered an error and was unable to review this pull request. You can try again by re-requesting a review.