CosmicClash/server/observability/slo.go

package observability

import (
	"sort"
	"time"
)

type SLOWindow struct {
	RegionalRTT []time.Duration
	Assignment  []time.Duration
	Connection  []time.Duration
	APILatency  []time.Duration
	Matches     []bool
	TickBacklog bool
	Headroom    float64
}

type SLOViolation struct {
	Metric string
	Reason string
}

func EvaluateSLO(window SLOWindow) []SLOViolation {
	violations := make([]SLOViolation, 0)
	if percentile(window.RegionalRTT, .95) > 80*time.Millisecond {
		violations = append(violations, SLOViolation{"regional_rtt_p95", "exceeds 80ms"})
	}
	if percentile(window.Assignment, .95) > 5*time.Second || percentile(window.Assignment, .99) > 10*time.Second {
		violations = append(violations, SLOViolation{"assignment_latency", "p95/p99 threshold exceeded"})
	}
	if percentile(window.Connection, .95) > 5*time.Second {
		violations = append(violations, SLOViolation{"connection_latency_p95", "exceeds 5s"})
	}
	if ratio(window.Matches) < .999 {
		violations = append(violations, SLOViolation{"allocation_result_success", "below 99.9%"})
	}
	if percentile(window.APILatency, .95) > 250*time.Millisecond {
		violations = append(violations, SLOViolation{"api_latency_p95", "exceeds 250ms"})
	}
	if window.TickBacklog {
		violations = append(violations, SLOViolation{"tick_health", "physics backlog detected"})
	}
	if window.Headroom > 0 && window.Headroom < .30 {
		violations = append(violations, SLOViolation{"resource_headroom", "below 30%"})
	}
	return violations
}

func percentile(values []time.Duration, p float64) time.Duration {
	if len(values) == 0 {
		return 0
	}
	ordered := append([]time.Duration(nil), values...)
	sort.Slice(ordered, func(i, j int) bool { return ordered[i] < ordered[j] })
	index := int(float64(len(ordered)-1) * p)
	return ordered[index]
}

func ratio(values []bool) float64 {
	if len(values) == 0 {
		return 1
	}
	success := 0
	for _, value := range values {
		if value {
			success++
		}
	}
	return float64(success) / float64(len(values))
}