Prometheus metrics and structured logging

[FumingPower3925]

Summary

Prometheus-compatible metrics exposition and structured JSON logging for key system events.

Prometheus Metrics (SDD 12.3)

Exposed at /metrics when EnableMetrics: true.

Metric	Type	Labels
celeris_requests_total	Counter	method, path, status
celeris_request_duration_seconds	Histogram	method, path
celeris_connections_active	Gauge	engine
celeris_connections_total	Counter	engine
celeris_engine_switches_total	Counter	from, to
celeris_overload_stage	Gauge	—
celeris_overload_escalations_total	Counter	from_stage, to_stage
celeris_syscalls_total	Counter	type
celeris_buffer_utilization	Gauge	worker
celeris_iouring_sq_utilization	Gauge	worker
celeris_iouring_tier	Gauge	—

Structured Logging (SDD 12.2)

JSON format for key events:

• Engine switch: {"event":"engine_switch","from":"iouring","to":"epoll","reason":"h2c_bias","timestamp":"..."}
• Overload escalation: {"event":"overload_escalation","from_stage":1,"to_stage":2,"cpu_pct":82.5,"timestamp":"..."}
• Overload de-escalation: similar format
• Tier detection: {"event":"tier_detected","tier":"High","kernel":"5.19.0","features":["multishot_accept","pbuf"]}

Tests

• All metrics registered with correct types and labels
• /metrics endpoint returns valid Prometheus format
• /metrics returns 404 when EnableMetrics: false
• Log output is valid JSON
• Engine switch event logged with correct fields
• Overload escalation/de-escalation logged
• Integration: generate traffic → scrape /metrics → verify counter increments

SDD Reference

Sections 12.2, 12.3

Dependencies

Depends on #14, #15

[FumingPower3925]

Core has observe.Collector with Snapshot; Prometheus/debug endpoints moved to goceleris/middlewares by design.