落地结构化日志与基础监控告警

2026-02-28 16:37:37 +08:00
parent 78f488fbbb
commit 5666dc61a0
7 changed files with 345 additions and 0 deletions
@@ -0,0 +1,83 @@
+package observability
+
+import (
+	"sync/atomic"
+	"time"
+)
+
+type Snapshot struct {
+	GeneratedAtUTC     string  `json:"generated_at_utc"`
+	UptimeSeconds      int64   `json:"uptime_seconds"`
+	TotalRequests      uint64  `json:"total_requests"`
+	ClientErrors       uint64  `json:"client_errors"`
+	ServerErrors       uint64  `json:"server_errors"`
+	ClientErrorRatePct float64 `json:"client_error_rate_pct"`
+	ServerErrorRatePct float64 `json:"server_error_rate_pct"`
+	AvgLatencyMs       float64 `json:"avg_latency_ms"`
+	MaxLatencyMs       float64 `json:"max_latency_ms"`
+}
+
+type Collector struct {
+	startedAt      time.Time
+	totalRequests  uint64
+	clientErrors   uint64
+	serverErrors   uint64
+	totalLatencyNs uint64
+	maxLatencyNs   uint64
+}
+
+func NewCollector() *Collector {
+	return &Collector{startedAt: time.Now()}
+}
+
+func (c *Collector) Observe(status int, latency time.Duration) {
+	atomic.AddUint64(&c.totalRequests, 1)
+	atomic.AddUint64(&c.totalLatencyNs, uint64(latency.Nanoseconds()))
+
+	if status >= 500 {
+		atomic.AddUint64(&c.serverErrors, 1)
+	} else if status >= 400 {
+		atomic.AddUint64(&c.clientErrors, 1)
+	}
+
+	latNs := uint64(latency.Nanoseconds())
+	for {
+		old := atomic.LoadUint64(&c.maxLatencyNs)
+		if latNs <= old {
+			break
+		}
+		if atomic.CompareAndSwapUint64(&c.maxLatencyNs, old, latNs) {
+			break
+		}
+	}
+}
+
+func (c *Collector) Snapshot() Snapshot {
+	now := time.Now()
+	total := atomic.LoadUint64(&c.totalRequests)
+	clientErr := atomic.LoadUint64(&c.clientErrors)
+	serverErr := atomic.LoadUint64(&c.serverErrors)
+	totalLatencyNs := atomic.LoadUint64(&c.totalLatencyNs)
+	maxLatencyNs := atomic.LoadUint64(&c.maxLatencyNs)
+
+	var clientRate float64
+	var serverRate float64
+	var avgLatencyMs float64
+	if total > 0 {
+		clientRate = float64(clientErr) * 100 / float64(total)
+		serverRate = float64(serverErr) * 100 / float64(total)
+		avgLatencyMs = float64(totalLatencyNs) / float64(total) / 1e6
+	}
+
+	return Snapshot{
+		GeneratedAtUTC:     now.UTC().Format(time.RFC3339),
+		UptimeSeconds:      int64(now.Sub(c.startedAt).Seconds()),
+		TotalRequests:      total,
+		ClientErrors:       clientErr,
+		ServerErrors:       serverErr,
+		ClientErrorRatePct: clientRate,
+		ServerErrorRatePct: serverRate,
+		AvgLatencyMs:       avgLatencyMs,
+		MaxLatencyMs:       float64(maxLatencyNs) / 1e6,
+	}
+}