Retry monitoring TG database connection on startup

cmd/classifier/main.go

@@ -15,6 +15,7 @@ import (
 	"time"
 
 	"monitoring-tg/internal/aiservice"
+	"monitoring-tg/internal/dbretry"
 
 	"github.com/jackc/pgx/v5/pgxpool"
 )
@@ -69,7 +70,7 @@ func main() {
 	ctx, stop := signal.NotifyContext(context.Background(), syscall.SIGINT, syscall.SIGTERM)
 	defer stop()
 
-	pool, err := pgxpool.New(ctx, cfg.databaseURL())
+	pool, err := dbretry.Connect(ctx, cfg.databaseURL(), 2*time.Minute)
 	if err != nil {
 		slog.Error("db_connect_failed", "error", err)
 		os.Exit(1)

cmd/server/main.go

@@ -21,6 +21,7 @@ import (
 	"time"
 
 	"monitoring-tg/internal/aiservice"
+	"monitoring-tg/internal/dbretry"
 
 	"github.com/jackc/pgx/v5"
 	"github.com/jackc/pgx/v5/pgxpool"
@@ -134,7 +135,7 @@ func main() {
 	ctx, stop := signal.NotifyContext(context.Background(), syscall.SIGINT, syscall.SIGTERM)
 	defer stop()
 
-	pool, err := pgxpool.New(ctx, cfg.databaseURL())
+	pool, err := dbretry.Connect(ctx, cfg.databaseURL(), 2*time.Minute)
 	if err != nil {
 		slog.Error("db_connect_failed", "error", err)
 		os.Exit(1)

internal/dbretry/dbretry.go

@@ -0,0 +1,44 @@
+package dbretry
+
+import (
+	"context"
+	"fmt"
+	"time"
+
+	"github.com/jackc/pgx/v5/pgxpool"
+)
+
+func Connect(ctx context.Context, databaseURL string, maxWait time.Duration) (*pgxpool.Pool, error) {
+	deadline := time.Now().Add(maxWait)
+	var lastErr error
+
+	for attempt := 1; ; attempt++ {
+		pool, err := pgxpool.New(ctx, databaseURL)
+		if err == nil {
+			if pingErr := pool.Ping(ctx); pingErr == nil {
+				return pool, nil
+			} else {
+				err = fmt.Errorf("ping postgres: %w", pingErr)
+				pool.Close()
+			}
+		} else {
+			err = fmt.Errorf("connect postgres: %w", err)
+		}
+		lastErr = err
+
+		if time.Now().After(deadline) {
+			return nil, fmt.Errorf("connect postgres after retry: %w", lastErr)
+		}
+		sleep := time.Duration(attempt) * time.Second
+		if sleep > 5*time.Second {
+			sleep = 5 * time.Second
+		}
+		timer := time.NewTimer(sleep)
+		select {
+		case <-ctx.Done():
+			timer.Stop()
+			return nil, fmt.Errorf("connect postgres cancelled: %w", ctx.Err())
+		case <-timer.C:
+		}
+	}
+}