fix(store): 数据库连接池限 25 条并设语句与空闲事务超时，迁移锁固定在取锁的连接上释放，API 导出连接池指标

internal/api/metrics.go

@@ -1,6 +1,7 @@
 package api
 
 import (
+	"database/sql"
 	"net/http"
 
 	"felis.lolicon.best/internal/metrics"
@@ -15,10 +16,12 @@ import (
 // build reconcile loop lives in cmd/felis.reconcileBuilds), so this endpoint is
 // that counter's sole scrape path; the operator's :8080 carries the fleet
 // gauges instead.
-var apiMetricsHandler = newAPIMetricsHandler()
+var (
+	apiRegistry       = prometheus.NewRegistry()
+	apiMetricsHandler = newAPIMetricsHandler(apiRegistry)
+)
 
-func newAPIMetricsHandler() http.Handler {
-	reg := prometheus.NewRegistry()
+func newAPIMetricsHandler(reg *prometheus.Registry) http.Handler {
 	// A fresh registry cannot already hold a collector, so Register's
 	// AlreadyRegistered tolerance arm never triggers here.
 	_ = metrics.Register(reg)
@@ -31,6 +34,13 @@ func newAPIMetricsHandler() http.Handler {
 	return promhttp.HandlerFor(reg, promhttp.HandlerOpts{})
 }
 
+// RegisterStorePool adds the store pool's go_sql_* series (db_name="felis"):
+// connections open, in use and idle against the cap, and how often and how
+// long requests waited for one, which is where a pool pinned at its cap shows.
+func RegisterStorePool(db *sql.DB) error {
+	return apiRegistry.Register(collectors.NewDBStatsCollector(db, "felis"))
+}
+
 // handleMetrics is mounted as a Public route on the internal face (the same
 // stance as the health probes): the internal listener is ClusterIP-only and a
 // Prometheus scrape carries no token. The external face never serves metrics.

internal/api/metrics_test.go

@@ -1,11 +1,14 @@
 package api
 
 import (
+	"database/sql"
 	"net/http"
 	"strings"
 	"testing"
 
 	"felis.lolicon.best/internal/metrics"
+
+	_ "github.com/jackc/pgx/v5/stdlib"
 )
 
 // TestMetricsEndpoint locks the scrape surface: the internal face serves the
@@ -42,3 +45,26 @@ func TestMetricsEndpoint(t *testing.T) {
 		t.Fatalf("external GET /metrics = %d, want 404 (metrics stay internal)", w.Code)
 	}
 }
+
+// The store pool's counts reach the internal scrape once registered.
+func TestMetricsEndpointCarriesTheStorePool(t *testing.T) {
+	pool, err := sql.Open("pgx", "postgres://[email protected]:1/felis")
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer pool.Close()
+	pool.SetMaxOpenConns(7)
+	if err := RegisterStorePool(pool); err != nil {
+		t.Fatal(err)
+	}
+
+	w := do((&API{}).InternalHandler(), "GET", "/metrics", "", nil)
+	for _, want := range []string{
+		`go_sql_max_open_connections{db_name="felis"} 7`,
+		`go_sql_wait_count_total{db_name="felis"} 0`,
+	} {
+		if !strings.Contains(w.Body.String(), want) {
+			t.Errorf("metrics exposition missing %q", want)
+		}
+	}
+}