[Gastown] PR 22: Observability

[jrf0110]

Parent: #204 | Phase 4: Hardening

Revised: Added container process logging and container-specific metrics.

Goal

Structured logging, real-time event streaming, and alerting for the Gastown system.

Components

Structured Logging

• Structured logging in Gastown worker (Sentry integration)
• Log all DO RPC calls with timing
• Container process logs forwarded to Workers Logs
• Agent lifecycle events (start, stop, crash, restart)

Event Streaming

• Bead event stream for real-time dashboard (DO → WebSocket or SSE)
• Agent status change events
• Convoy progress events

Alerting

• GUPP violations (agent stale for configurable threshold)
• Escalation rate spikes
• Review queue depth exceeding threshold
• Agent restart loops (same agent restarting N times in M minutes)
• Container OOM events
• Container cold start frequency

Usage Metrics

• Beads per day (by type, rig, town)
• Agents per day (by role, rig)
• LLM cost per bead
• Average bead completion time
• Container uptime and resource utilization

Dependencies

• All Phase 1–3 PRs

Tech

• Cloudflare analytics engine (as much as possible, put tons of data in analytics engine. over log report in analytics engine)
• Cloudflare Workers logs

Acceptance Criteria

• Structured logging with Sentry integration
• Container process logs in Workers Logs
• Real-time bead event stream
• Configurable alerting rules
• Usage metrics collection and dashboard
• Container health metrics (uptime, OOM events, cold starts)