fix(telegram): add request timeout for notifications

.env.example

@@ -26,6 +26,8 @@ TELEGRAM_NOTIFY_INFO=true
 TELEGRAM_NOTIFY_WARN=true
 TELEGRAM_NOTIFY_ALERT=true
 TELEGRAM_NOTIFY_REPORT=true
+TELEGRAM_REQUEST_TIMEOUT_SEC=10
+TELEGRAM_QUEUE_SIZE=256
 
 STRATEGY_ROLLING_SHORT=60
 STRATEGY_ROLLING_LONG=252

README.md

@@ -86,6 +86,8 @@ Common formats:
 | `TELEGRAM_NOTIFY_WARN` | `true` | Send warnings. |
 | `TELEGRAM_NOTIFY_ALERT` | `true` | Send critical alerts. |
 | `TELEGRAM_NOTIFY_REPORT` | `true` | Send daily reports. |
+| `TELEGRAM_REQUEST_TIMEOUT_SEC` | `10` | HTTP timeout for Telegram API calls, including the startup `getMe` check and message sends. |
+| `TELEGRAM_QUEUE_SIZE` | `256` | In-memory Telegram notification queue size. The queue smooths transient Telegram slowness but is not persisted across process restarts. |
 
 ### STRATEGY
 

README.ru.md

@@ -77,6 +77,8 @@ APP_MODE=backtest go run ./cmd/bot
 | `TELEGRAM_NOTIFY_WARN` | `true` или `false` | `true` | boolean | Включает предупреждения. |
 | `TELEGRAM_NOTIFY_ALERT` | `true` или `false` | `true` | boolean | Включает alert-сообщения. Они считаются критичными для доставки и ждут место в очереди. |
 | `TELEGRAM_NOTIFY_REPORT` | `true` или `false` | `true` | boolean | Включает дневные отчёты. |
+| `TELEGRAM_REQUEST_TIMEOUT_SEC` | целое число секунд | `10` | должно быть `> 0` | Таймаут HTTP-запросов к Telegram API, включая стартовую проверку `getMe` и отправку сообщений. |
+| `TELEGRAM_QUEUE_SIZE` | целое число | `256` | должно быть `> 0` | Размер in-memory очереди Telegram-уведомлений. Очередь сглаживает временную медленную доставку, но не сохраняется между рестартами процесса. |
 
 ### STRATEGY
 

internal/app/app.go

@@ -178,6 +178,8 @@ func Run(ctx context.Context, opts Options) error {
 		NotifyWarn:     cfg.Telegram.NotifyWarn,
 		NotifyAlert:    cfg.Telegram.NotifyAlert,
 		NotifyReport:   cfg.Telegram.NotifyReport,
+		RequestTimeout: time.Duration(cfg.Telegram.RequestTimeoutSec) * time.Second,
+		QueueSize:      cfg.Telegram.QueueSize,
 		AuditSink:      repo,
 	}, log)
 	if err != nil {

internal/config/config.go

@@ -72,6 +72,8 @@ type TelegramConfig struct {
 	NotifyWarn        bool   `env:"NOTIFY_WARN" envDefault:"true"`
 	NotifyAlert       bool   `env:"NOTIFY_ALERT" envDefault:"true"`
 	NotifyReport      bool   `env:"NOTIFY_REPORT" envDefault:"true"`
+	RequestTimeoutSec int    `env:"REQUEST_TIMEOUT_SEC" envDefault:"10"`
+	QueueSize         int    `env:"QUEUE_SIZE" envDefault:"256"`
 }
 
 type StrategyConfig struct {
@@ -201,6 +203,12 @@ func (c *Config) Validate() error {
 	if c.TInvest.RequestTimeoutSec <= 0 {
 		return errors.New("TINVEST_REQUEST_TIMEOUT_SEC must be positive")
 	}
+	if c.Telegram.RequestTimeoutSec <= 0 {
+		return errors.New("TELEGRAM_REQUEST_TIMEOUT_SEC must be positive")
+	}
+	if c.Telegram.QueueSize <= 0 {
+		return errors.New("TELEGRAM_QUEUE_SIZE must be positive")
+	}
 	if c.TInvest.TradingCalendarExchange == "" {
 		c.TInvest.TradingCalendarExchange = "MOEX"
 	}

internal/config/config_test.go

@@ -86,6 +86,39 @@ func TestLoadSchedulerKnobsFromEnv(t *testing.T) {
 	}
 }
 
+func TestLoadTelegramQueueDefaults(t *testing.T) {
+	t.Setenv("APP_MODE", "backtest")
+
+	cfg, err := Load()
+	if err != nil {
+		t.Fatal(err)
+	}
+	if cfg.Telegram.RequestTimeoutSec != 10 {
+		t.Fatalf("telegram request timeout=%d, want 10", cfg.Telegram.RequestTimeoutSec)
+	}
+	if cfg.Telegram.QueueSize != 256 {
+		t.Fatalf("telegram queue size=%d, want 256", cfg.Telegram.QueueSize)
+	}
+}
+
+func TestValidateRejectsInvalidTelegramRequestTimeout(t *testing.T) {
+	cfg := minimalBrokerConfig(domain.ModeSandbox)
+	cfg.Telegram.RequestTimeoutSec = 0
+	err := cfg.Validate()
+	if err == nil || !strings.Contains(err.Error(), "TELEGRAM_REQUEST_TIMEOUT_SEC") {
+		t.Fatalf("Validate err=%v, want TELEGRAM_REQUEST_TIMEOUT_SEC requirement", err)
+	}
+}
+
+func TestValidateRejectsInvalidTelegramQueueSize(t *testing.T) {
+	cfg := minimalBrokerConfig(domain.ModeSandbox)
+	cfg.Telegram.QueueSize = 0
+	err := cfg.Validate()
+	if err == nil || !strings.Contains(err.Error(), "TELEGRAM_QUEUE_SIZE") {
+		t.Fatalf("Validate err=%v, want TELEGRAM_QUEUE_SIZE requirement", err)
+	}
+}
+
 func minimalBrokerConfig(mode domain.Mode) Config {
 	return Config{
 		App: AppConfig{
@@ -99,6 +132,10 @@ func minimalBrokerConfig(mode domain.Mode) Config {
 			RequestTimeoutSec: 10,
 		},
 		DB: DBConfig{DSN: "user:pass@tcp(localhost:3306)/bot"},
+		Telegram: TelegramConfig{
+			RequestTimeoutSec: 10,
+			QueueSize:         256,
+		},
 		Execution: ExecutionConfig{
 			EntrySignalTime:     mustTOD("18:10:00"),
 			EntryWindowStart:    mustTOD("18:20:00"),

internal/notify/notify.go

@@ -5,6 +5,7 @@ import (
 	"errors"
 	"fmt"
 	"log/slog"
+	"net/http"
 	"strings"
 	"time"
 
@@ -13,7 +14,11 @@ import (
 	"overnight-trading-bot/internal/domain"
 )
 
-const mustDeliverEnqueueTimeout = 2 * time.Second
+const (
+	defaultQueueSize          = 256
+	defaultRequestTimeout     = 10 * time.Second
+	mustDeliverEnqueueTimeout = 2 * time.Second
+)
 
 type Notifier interface {
 	Info(ctx context.Context, msg string) error
@@ -38,6 +43,8 @@ type TelegramConfig struct {
 	NotifyWarn     bool
 	NotifyAlert    bool
 	NotifyReport   bool
+	RequestTimeout time.Duration
+	QueueSize      int
 	AuditSink      AuditSink
 }
 
@@ -52,6 +59,7 @@ type Telegram struct {
 	queue      chan outbound
 	done       chan struct{}
 	closed     chan struct{}
+	retryDelay func(error, int) time.Duration
 }
 
 type outbound struct {
@@ -61,13 +69,23 @@ type outbound struct {
 }
 
 func NewTelegram(cfg TelegramConfig, log *slog.Logger) (Notifier, error) {
+	return newTelegramWithEndpoint(cfg, log, tgbotapi.APIEndpoint)
+}
+
+func newTelegramWithEndpoint(cfg TelegramConfig, log *slog.Logger, apiEndpoint string) (Notifier, error) {
 	if cfg.BotToken == "" || cfg.ChatID == 0 {
 		if log != nil {
 			log.Warn("telegram notifier disabled; TELEGRAM_BOT_TOKEN or TELEGRAM_CHAT_ID is empty")
 		}
 		return Noop{}, nil
 	}
-	bot, err := tgbotapi.NewBotAPI(cfg.BotToken)
+	if cfg.RequestTimeout <= 0 {
+		cfg.RequestTimeout = defaultRequestTimeout
+	}
+	if cfg.QueueSize <= 0 {
+		cfg.QueueSize = defaultQueueSize
+	}
+	bot, err := tgbotapi.NewBotAPIWithClient(cfg.BotToken, apiEndpoint, &http.Client{Timeout: cfg.RequestTimeout})
 	if err != nil {
 		return nil, err
 	}
@@ -75,9 +93,10 @@ func NewTelegram(cfg TelegramConfig, log *slog.Logger) (Notifier, error) {
 		cfg:        cfg,
 		bot:        bot,
 		log:        log,
-		queue:  make(chan outbound, 256),
+		queue:      make(chan outbound, cfg.QueueSize),
 		done:       make(chan struct{}),
 		closed:     make(chan struct{}),
+		retryDelay: telegramRetryDelay,
 	}
 	go t.dispatch()
 	return t, nil
@@ -181,7 +200,7 @@ func (t *Telegram) send(item outbound) {
 	for attempt := 0; attempt < 3; attempt++ {
 		if _, err := t.bot.Send(msg); err != nil {
 			lastErr = err
-			delay := telegramRetryDelay(err, attempt)
+			delay := t.retryDelay(err, attempt)
 			if t.log != nil {
 				t.log.Warn("telegram send failed", "attempt", attempt+1, "err", err, "retry_in", delay)
 			}

internal/notify/telegram_test.go

@@ -1,6 +1,10 @@
 package notify
 
 import (
+	"context"
+	"net/http"
+	"net/http/httptest"
+	"strings"
 	"testing"
 	"time"
 
@@ -44,6 +48,131 @@ func TestTelegramRetryDelayUsesRetryAfter(t *testing.T) {
 	}
 }
 
+func TestNewTelegramUsesRequestTimeout(t *testing.T) {
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		time.Sleep(200 * time.Millisecond)
+	}))
+	defer server.Close()
+
+	start := time.Now()
+	_, err := newTelegramWithEndpoint(TelegramConfig{
+		BotToken:       "token",
+		ChatID:         123,
+		RequestTimeout: 20 * time.Millisecond,
+		QueueSize:      1,
+	}, nil, telegramTestEndpoint(server))
+	elapsed := time.Since(start)
+
+	if err == nil {
+		t.Fatal("NewTelegram err=nil, want timeout")
+	}
+	if elapsed > 150*time.Millisecond {
+		t.Fatalf("NewTelegram took %s, want request timeout to fire quickly", elapsed)
+	}
+}
+
+func TestNewTelegramUsesConfiguredQueueSize(t *testing.T) {
+	server := telegramTestServer(t, func(w http.ResponseWriter, r *http.Request) {
+		writeTelegramOK(w, r)
+	})
+	defer server.Close()
+
+	notifier, err := newTelegramWithEndpoint(TelegramConfig{
+		BotToken:       "token",
+		ChatID:         123,
+		RequestTimeout: time.Second,
+		QueueSize:      3,
+	}, nil, telegramTestEndpoint(server))
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer notifier.Close()
+
+	telegram, ok := notifier.(*Telegram)
+	if !ok {
+		t.Fatalf("notifier type=%T, want *Telegram", notifier)
+	}
+	if got := cap(telegram.queue); got != 3 {
+		t.Fatalf("queue cap=%d, want 3", got)
+	}
+}
+
+func TestTelegramSendTimeoutAuditsMustDeliverFailure(t *testing.T) {
+	audit := &recordingAuditSink{events: make(chan domain.RiskEvent, 1)}
+	server := telegramTestServer(t, func(w http.ResponseWriter, r *http.Request) {
+		if strings.HasSuffix(r.URL.Path, "/sendMessage") {
+			time.Sleep(50 * time.Millisecond)
+			writeTelegramOK(w, r)
+			return
+		}
+		writeTelegramOK(w, r)
+	})
+	defer server.Close()
+
+	notifier, err := newTelegramWithEndpoint(TelegramConfig{
+		BotToken:       "token",
+		ChatID:         123,
+		NotifyAlert:    true,
+		RequestTimeout: 10 * time.Millisecond,
+		QueueSize:      1,
+		AuditSink:      audit,
+	}, nil, telegramTestEndpoint(server))
+	if err != nil {
+		t.Fatal(err)
+	}
+	telegram := notifier.(*Telegram)
+	telegram.retryDelay = func(error, int) time.Duration { return time.Millisecond }
+	defer notifier.Close()
+
+	if err := notifier.Alert(context.Background(), "critical"); err != nil {
+		t.Fatal(err)
+	}
+
+	select {
+	case event := <-audit.events:
+		if event.EventType != "notification_undeliverable" {
+			t.Fatalf("event type=%s, want notification_undeliverable", event.EventType)
+		}
+		if event.Severity != domain.SeverityCritical {
+			t.Fatalf("event severity=%s, want critical", event.Severity)
+		}
+	case <-time.After(500 * time.Millisecond):
+		t.Fatal("timed out waiting for must-deliver audit event")
+	}
+}
+
 type assertErr struct{}
 
 func (assertErr) Error() string { return "boom" }
+
+type recordingAuditSink struct {
+	events chan domain.RiskEvent
+}
+
+func (s *recordingAuditSink) InsertRiskEvent(ctx context.Context, event domain.RiskEvent) error {
+	select {
+	case s.events <- event:
+	case <-ctx.Done():
+		return ctx.Err()
+	}
+	return nil
+}
+
+func telegramTestEndpoint(server *httptest.Server) string {
+	return server.URL + "/bot%s/%s"
+}
+
+func telegramTestServer(t *testing.T, handler http.HandlerFunc) *httptest.Server {
+	t.Helper()
+	return httptest.NewServer(handler)
+}
+
+func writeTelegramOK(w http.ResponseWriter, r *http.Request) {
+	w.Header().Set("Content-Type", "application/json")
+	switch {
+	case strings.HasSuffix(r.URL.Path, "/getMe"):
+		_, _ = w.Write([]byte(`{"ok":true,"result":{"id":1,"is_bot":true,"first_name":"test","username":"test_bot"}}`))
+	default:
+		_, _ = w.Write([]byte(`{"ok":true,"result":{"message_id":1,"date":0,"chat":{"id":123,"type":"private"},"text":"ok"}}`))
+	}
+}