Expand logging

[446564]

Beacon currently uses the standard log.Printf throughout for all output with no way to filter verbosity or change output format. This makes it difficult to run quietly in production or debug specific subsystems without log noise.

Goals

• Configurable log level (debug, info, warn, error) via config or env var
• Structured logging with consistent fields (broker name, component, etc.) for easier log aggregation
• Minimal disruption to existing log call sites — ideally a drop-in migration path

Considerations

• log/slog (stdlib, Go 1.21+) is the natural fit given the Go version — no additional dependency
• Text handler for human-readable output in development, JSON handler for production
• Log level should be settable via env var (e.g. LOG_LEVEL=debug) and optionally config.yaml
• Output format (text/json) could also be configurable
• The ingest path is the noisiest — broker name is already included in most log lines but would benefit from being a structured field
• The cache layer currently has no logging at all — debug-level cache hit/miss logging would be useful here

Scope

• Add LogConfig to internal/config/config.go (level, format)
• Initialize a global slog.Logger in main.go and pass or set as default
• Migrate existing log.Printf call sites to slog — info for normal operation, warn for degraded states, error for failures
• Add debug-level cache hit/miss logging to internal/cache

Out of scope

• Log rotation or file output (use a reverse proxy or systemd for that)
• Per-subsystem log levels (can be revisited if needed)

[n30nex]

I’m implementing the logging controls with Go's standard log/slog: log.level/log.format, LOG_LEVEL/LOG_FORMAT overrides, explicit runtime severities, and component/broker fields. Startup failures must remain visible at every supported level. Runtime log sites need to move together so warn/error settings do not accidentally suppress failures left behind as legacy info messages.

I’ll keep cache hit/miss diagnostics in a separate small companion PR. Edge Caddy/Apache access logs and fail2ban policy stay unchanged; the app request logger will retain trusted client identity and panic handling without logging query strings. I’ll verify levels/formats, startup errors, cache behavior and native Pi ingestion, and update the local MQTT log-reading helper for structured records. No rotation, file-output system, per-component level controls or new logging dependency is planned.

[n30nex]

Submitted as two independent PRs against dev: #144 for logging controls/runtime migration and #145 for cache diagnostics. Both have green CI/CodeQL, native Pi 5 full tests with PostgreSQL enabled, and focused race coverage.

The Canadaverse preview now combines them with the existing candidates and runs JSON/info logging. Actual records parse as JSON, carry broker/component fields and preserve HTTP severity/client identity; both feeds advance. Public API/WebSocket and matching corresponding-source checks pass. Native tests cover debug/error thresholds, startup failure visibility and payload-free cache diagnostics. No schema change or new dependency was needed.

The combined build also preserves #141's pre-handshake attempt limiter and close-1013 concurrent-limit signal, and updates #140's log assertion for the new format. That merge interaction is called out in #144. The local MQTT log classifier now accepts text/JSON; #116 remains a separate stability investigation. Leaving #51 open for review and merge tracking.

[n30nex]

Review and closure tracking: #144 provides the log controls/call-site migration; #145 provides the cache hit/miss diagnostics required by this issue. Both parts are needed for completion.

The review requests recommend #145 first, followed by #144. Once the cache change is merged and the combined scope is accepted, the closing reference for this issue belongs on the remaining logging PR. Until then, neither individual PR has an automatic closing reference, so a partial merge will not prematurely mark this complete. If both have already landed, the complete issue can be closed after confirming both commits are present in dev.