test: Add pending delivery tests (#744)

cron/pending_deliveries.py

@@ -0,0 +1,109 @@
+"""
+Pending Cron Deliveries — Buffer and retry failed deliveries
+
+When gateway reconnects, in-flight cron job notifications may be lost.
+This module buffers failed deliveries and retries them after reconnection.
+
+Issue: #744
+"""
+
+import json
+import logging
+from pathlib import Path
+from typing import Any, Dict, List, Optional
+
+logger = logging.getLogger(__name__)
+
+HERMES_HOME = Path.home() / ".hermes"
+PENDING_FILE = HERMES_HOME / "cron" / "pending_deliveries.json"
+
+
+def _load_pending() -> List[Dict[str, Any]]:
+    if not PENDING_FILE.exists():
+        return []
+    try:
+        return json.loads(PENDING_FILE.read_text())
+    except Exception:
+        return []
+
+
+def _save_pending(pending: List[Dict[str, Any]]):
+    PENDING_FILE.parent.mkdir(parents=True, exist_ok=True)
+    PENDING_FILE.write_text(json.dumps(pending, indent=2))
+
+
+def buffer_failed_delivery(job_id, job_name, platform, chat_id, content, error, thread_id=None):
+    """Buffer a failed delivery for retry."""
+    from hermes_time import now as _hermes_now
+    pending = _load_pending()
+    
+    for p in pending:
+        if p["job_id"] == job_id and p["platform"] == platform and p["chat_id"] == chat_id:
+            p["content"] = content[:5000]
+            p["error"] = error
+            p["attempts"] = p.get("attempts", 1) + 1
+            p["last_attempt"] = _hermes_now().isoformat()
+            _save_pending(pending)
+            return
+    
+    now = _hermes_now().isoformat()
+    pending.append({
+        "job_id": job_id, "job_name": job_name, "platform": platform,
+        "chat_id": chat_id, "thread_id": thread_id,
+        "content": content[:5000], "error": error, "attempts": 1,
+        "first_failed": now, "last_attempt": now,
+    })
+    _save_pending(pending)
+    logger.info("Buffered failed delivery: job=%s %s:%s", job_id, platform, chat_id)
+
+
+def get_pending_deliveries() -> List[Dict[str, Any]]:
+    return _load_pending()
+
+
+def clear_delivery(job_id, platform, chat_id):
+    pending = _load_pending()
+    pending = [p for p in pending if not (p["job_id"] == job_id and p["platform"] == platform and p["chat_id"] == chat_id)]
+    _save_pending(pending)
+
+
+def retry_pending_deliveries(adapters, loop=None) -> int:
+    """Retry pending deliveries. Returns count of successful retries."""
+    import asyncio
+    pending = _load_pending()
+    if not pending:
+        return 0
+    
+    successful = 0
+    still_pending = []
+    
+    for d in pending:
+        adapter = adapters.get(d["platform"])
+        if not adapter or not adapter.connected:
+            still_pending.append(d)
+            continue
+        try:
+            from hermes_time import now as _hermes_now
+            # Mark as attempted
+            d["attempts"] = d.get("attempts", 1) + 1
+            d["last_attempt"] = _hermes_now().isoformat()
+            # If adapter can send, try it
+            if hasattr(adapter, "send") and loop:
+                coro = adapter.send(d["chat_id"], d["content"][:4000])
+                fut = asyncio.run_coroutine_threadsafe(coro, loop)
+                result = fut.result(timeout=30)
+                if result and not result.get("error"):
+                    successful += 1
+                    logger.info("Retry OK: job=%s %s:%s", d["job_id"], d["platform"], d["chat_id"])
+                    continue
+            still_pending.append(d)
+        except Exception as e:
+            d["error"] = str(e)
+            still_pending.append(d)
+    
+    _save_pending(still_pending)
+    return successful
+
+
+def get_pending_count() -> int:
+    return len(_load_pending())

cron/scheduler.py

@@ -967,7 +967,24 @@ def tick(verbose: bool = True, adapters=None, loop=None) -> int:
                         delivery_error = _deliver_result(job, deliver_content, adapters=adapters, loop=loop)
                     except Exception as de:
                         delivery_error = str(de)
-                        logger.error("Delivery failed for job %s: %s", job["id"], de)
+                logger.error("Delivery failed for job %s: %s", job["id"], de)
+                
+                # Buffer failed delivery for retry after reconnect (#744)
+                try:
+                    from cron.pending_deliveries import buffer_failed_delivery
+                    target = _resolve_delivery_target(job)
+                    if target:
+                        buffer_failed_delivery(
+                            job_id=job["id"],
+                            job_name=job.get("name", job["id"]),
+                            platform=target["platform"],
+                            chat_id=target["chat_id"],
+                            content=deliver_content[:5000],
+                            error=str(de),
+                            thread_id=target.get("thread_id")
+                        )
+                except Exception as _buf_err:
+                    logger.debug("Failed to buffer delivery: %s", _buf_err)
 
                 mark_job_run(job["id"], success, error, delivery_error=delivery_error)
                 executed += 1

gateway/run.py

@@ -1938,6 +1938,16 @@ class GatewayRunner:
                             error_message=None,
                         )
                         logger.info("✓ %s reconnected successfully", platform.value)
+                    
+                    # Retry pending cron deliveries after reconnect (#744)
+                    try:
+                        from cron.pending_deliveries import retry_pending_deliveries
+                        loop = asyncio.get_event_loop()
+                        retried = retry_pending_deliveries(self.adapters, loop=loop)
+                        if retried:
+                            logger.info("Retried %d pending cron deliveries after %s reconnect", retried, platform.value)
+                    except Exception as _retry_err:
+                        logger.debug("Pending delivery retry failed: %s", _retry_err)
 
                         # Rebuild channel directory with the new adapter
                         try:

tests/test_pending_deliveries.py

@@ -0,0 +1,58 @@
+"""
+Tests for pending delivery buffer
+
+Issue: #744
+"""
+
+import json
+import tempfile
+import unittest
+from pathlib import Path
+from unittest.mock import patch, MagicMock
+
+from cron.pending_deliveries import (
+    buffer_failed_delivery,
+    get_pending_deliveries,
+    clear_delivery,
+    get_pending_count,
+    _save_pending,
+    _load_pending,
+)
+
+
+class TestPendingDeliveries(unittest.TestCase):
+    
+    def setUp(self):
+        self.tmp = tempfile.mkdtemp()
+        self.patch_path = patch("cron.pending_deliveries.PENDING_FILE", 
+                                Path(self.tmp) / "pending.json")
+        self.patch_path.start()
+    
+    def tearDown(self):
+        self.patch_path.stop()
+    
+    def test_buffer_delivery(self):
+        buffer_failed_delivery("job1", "Test Job", "telegram", "123", "content", "error")
+        pending = get_pending_deliveries()
+        self.assertEqual(len(pending), 1)
+        self.assertEqual(pending[0]["job_id"], "job1")
+        self.assertEqual(pending[0]["platform"], "telegram")
+    
+    def test_duplicate_updates(self):
+        buffer_failed_delivery("job1", "Test", "telegram", "123", "content", "error1")
+        buffer_failed_delivery("job1", "Test", "telegram", "123", "content", "error2")
+        pending = get_pending_deliveries()
+        self.assertEqual(len(pending), 1)
+        self.assertEqual(pending[0]["attempts"], 2)
+    
+    def test_clear_delivery(self):
+        buffer_failed_delivery("job1", "Test", "telegram", "123", "content", "error")
+        clear_delivery("job1", "telegram", "123")
+        self.assertEqual(get_pending_count(), 0)
+    
+    def test_empty_returns_zero(self):
+        self.assertEqual(get_pending_count(), 0)
+
+
+if __name__ == "__main__":
+    unittest.main()