Changeset 0.33.0 (#207)

Co-authored-by: Jeffrey Smith <jasafpro@gmail.com>
Co-committed-by: Jeffrey Smith <jasafpro@gmail.com>

server/metrics/db_collector.go

@@ -0,0 +1,26 @@
+package metrics
+
+import (
+	"database/sql"
+	"time"
+)
+
+// StartDBCollector begins a background goroutine that reads sql.DBStats
+// every interval and updates the Prometheus DB pool gauges.
+func StartDBCollector(db *sql.DB, interval time.Duration) {
+	if db == nil {
+		return
+	}
+	go func() {
+		ticker := time.NewTicker(interval)
+		defer ticker.Stop()
+		for range ticker.C {
+			stats := db.Stats()
+			DBOpenConnections.Set(float64(stats.OpenConnections))
+			DBInUseConnections.Set(float64(stats.InUse))
+			DBIdleConnections.Set(float64(stats.Idle))
+			DBWaitCount.Set(float64(stats.WaitCount))
+			DBWaitDuration.Set(stats.WaitDuration.Seconds())
+		}
+	}()
+}

server/metrics/metrics.go

@@ -0,0 +1,97 @@
+// Package metrics defines all Prometheus metrics for Chat Switchboard.
+// All metrics use the "switchboard_" prefix. Registered via promauto
+// so they are available on the default registry's /metrics endpoint.
+package metrics
+
+import (
+	"github.com/prometheus/client_golang/prometheus"
+	"github.com/prometheus/client_golang/prometheus/promauto"
+)
+
+// ── HTTP Request Metrics ────────────────────
+
+var (
+	HTTPRequestsTotal = promauto.NewCounterVec(prometheus.CounterOpts{
+		Name: "switchboard_http_requests_total",
+		Help: "Total HTTP requests by method, path pattern, and status code.",
+	}, []string{"method", "path_pattern", "status"})
+
+	HTTPRequestDuration = promauto.NewHistogramVec(prometheus.HistogramOpts{
+		Name:    "switchboard_http_request_duration_seconds",
+		Help:    "HTTP request latency in seconds.",
+		Buckets: []float64{0.01, 0.05, 0.1, 0.25, 0.5, 1, 2.5, 5, 10, 30},
+	}, []string{"method", "path_pattern"})
+)
+
+// ── WebSocket Metrics ───────────────────────
+
+var WebSocketConnections = promauto.NewGauge(prometheus.GaugeOpts{
+	Name: "switchboard_websocket_connections",
+	Help: "Current number of active WebSocket connections.",
+})
+
+// ── Completion / Token Metrics ──────────────
+
+var (
+	CompletionTokensTotal = promauto.NewCounterVec(prometheus.CounterOpts{
+		Name: "switchboard_completion_tokens_total",
+		Help: "Total tokens processed by direction (prompt|completion) and model.",
+	}, []string{"direction", "model_id"})
+
+	CompletionsTotal = promauto.NewCounterVec(prometheus.CounterOpts{
+		Name: "switchboard_completions_total",
+		Help: "Total completion requests by provider, model, and status.",
+	}, []string{"provider_config_id", "model_id", "status"})
+
+	CompletionDuration = promauto.NewHistogramVec(prometheus.HistogramOpts{
+		Name:    "switchboard_completion_duration_seconds",
+		Help:    "Completion request latency (wall time including tool loops).",
+		Buckets: []float64{0.1, 0.5, 1, 2, 5, 10, 30, 60, 120},
+	}, []string{"provider_config_id", "model_id"})
+)
+
+// ── Provider Health ─────────────────────────
+
+var ProviderStatus = promauto.NewGaugeVec(prometheus.GaugeOpts{
+	Name: "switchboard_provider_status",
+	Help: "Provider health status: 0=unknown, 1=healthy, 2=degraded, 3=down.",
+}, []string{"provider_config_id"})
+
+// ── Database Pool Metrics ───────────────────
+
+var (
+	DBOpenConnections = promauto.NewGauge(prometheus.GaugeOpts{
+		Name: "switchboard_db_open_connections",
+		Help: "Number of open database connections.",
+	})
+	DBInUseConnections = promauto.NewGauge(prometheus.GaugeOpts{
+		Name: "switchboard_db_in_use_connections",
+		Help: "Number of database connections currently in use.",
+	})
+	DBIdleConnections = promauto.NewGauge(prometheus.GaugeOpts{
+		Name: "switchboard_db_idle_connections",
+		Help: "Number of idle database connections.",
+	})
+	DBWaitCount = promauto.NewGauge(prometheus.GaugeOpts{
+		Name: "switchboard_db_wait_count_total",
+		Help: "Total number of connections waited for.",
+	})
+	DBWaitDuration = promauto.NewGauge(prometheus.GaugeOpts{
+		Name: "switchboard_db_wait_duration_seconds_total",
+		Help: "Total time blocked waiting for a new connection (seconds).",
+	})
+)
+
+// ── Task Scheduler Metrics ──────────────────
+
+var TaskExecutionsTotal = promauto.NewCounterVec(prometheus.CounterOpts{
+	Name: "switchboard_task_executions_total",
+	Help: "Total task executions by status (success|error).",
+}, []string{"status"})
+
+// ── Sandbox Metrics ─────────────────────────
+
+var SandboxExecutionsTotal = promauto.NewCounterVec(prometheus.CounterOpts{
+	Name: "switchboard_sandbox_executions_total",
+	Help: "Total Starlark sandbox executions by entry point and status.",
+}, []string{"entry_point", "status"})