fix(scheduled-runner): ретрай транзиентных ошибок отправки (522/5xx/429/сеть) вместо мгновенного failed + send_attempts

src/config/db.js

@@ -224,6 +224,7 @@ const migrate = async () => {
 
   // safe column alters (existing tables on prod may lack newer columns)
   await query(`ALTER TABLE categories ADD COLUMN IF NOT EXISTS is_active BOOLEAN DEFAULT true`);
+  await query(`ALTER TABLE scheduled_posts ADD COLUMN IF NOT EXISTS send_attempts INT NOT NULL DEFAULT 0`);
 
   console.log('[DB] Migrations applied');
 };

src/services/scheduledPostsRunner.js

@@ -401,6 +401,21 @@ async function publishOne(scheduledPost) {
 // массового ретрая всё накопившееся улетело бы в канал пачкой.
 const SKIP_OLDER_THAN_H = 3;
 
+// Ретрай транзиентных ошибок отправки (522/5xx/429/сетевые таймауты).
+// Разовый сбой Telegram-прокси (CF Worker) не должен насмерть ронять слот.
+const MAX_SEND_ATTEMPTS = 5;
+const RETRY_BACKOFF_MIN = [2, 5, 10, 20, 30]; // минуты между попытками
+function isTransientSendError(err) {
+  const status = err.response?.status;
+  if (status && status >= 500) return true;   // 500..599 включая 522
+  if (status === 429) return true;             // rate limit
+  const code = String(err.code || '');
+  if (['ECONNRESET','ETIMEDOUT','ECONNREFUSED','ECONNABORTED','EAI_AGAIN','EPIPE'].includes(code)) return true;
+  const m = String(err.message || '').toLowerCase();
+  if (m.includes('timeout') || m.includes('socket hang up') || m.includes('network')) return true;
+  return false;
+}
+
 async function runScheduled() {
   // 1) Помечаем слишком старые pending как skipped (не спамим канал задним числом)
   const { rows: skipped } = await query(
@@ -434,12 +449,24 @@ async function runScheduled() {
       console.log(`[scheduled-runner] sent id=${sp.id} channel=${sp.channel_id} article=${sp.article_id}`);
     } catch (err) {
       const msg = err.response?.data?.description || err.response?.data?.error?.error_msg || err.message;
-      await query(
-        `UPDATE scheduled_posts SET status='failed', error=$1 WHERE id=$2`,
-        [String(msg).slice(0, 1000), sp.id]
-      );
-      results.push({ id: sp.id, ok: false, error: msg });
-      console.error(`[scheduled-runner] failed id=${sp.id}: ${msg}`);
+      const attempts = (sp.send_attempts || 0) + 1;
+      if (isTransientSendError(err) && attempts < MAX_SEND_ATTEMPTS) {
+        const delayMin = RETRY_BACKOFF_MIN[Math.min(attempts - 1, RETRY_BACKOFF_MIN.length - 1)];
+        const retryAt = new Date(Date.now() + delayMin * 60_000);
+        await query(
+          `UPDATE scheduled_posts SET status='pending', scheduled_at=$1, send_attempts=$2, error=$3 WHERE id=$4`,
+          [retryAt, attempts, `retry ${attempts}/${MAX_SEND_ATTEMPTS} через ${delayMin}м: ${String(msg).slice(0, 300)}`, sp.id]
+        );
+        results.push({ id: sp.id, ok: false, retry: attempts, error: msg });
+        console.warn(`[scheduled-runner] transient id=${sp.id}, retry ${attempts}/${MAX_SEND_ATTEMPTS} in ${delayMin}m: ${msg}`);
+      } else {
+        await query(
+          `UPDATE scheduled_posts SET status='failed', send_attempts=$1, error=$2 WHERE id=$3`,
+          [attempts, String(msg).slice(0, 1000), sp.id]
+        );
+        results.push({ id: sp.id, ok: false, error: msg });
+        console.error(`[scheduled-runner] failed id=${sp.id} (attempts=${attempts}): ${msg}`);
+      }
     }
   }
   return { processed: rows.length, skipped: skipped.length, results };