From 5fbd73a038c4b008e811bd476abab2f43cedc8f7 Mon Sep 17 00:00:00 2001
From: codex <codex@local>
Date: Sat, 13 Jun 2026 20:46:45 +0000
Subject: [PATCH] fix(safe): retry Safe API transport errors instead of
 crashing

get_safe_transactions() retried on HTTP 429/5xx but let transport-level
failures (connection reset, read timeout, DNS) propagate uncaught,
crashing the safe monitor with a bare requests.ConnectionError that
surfaces as a "[yearn] main crashed" Telegram alert.

Wrap the request in the existing backoff/retry loop and add a 10s
timeout, mirroring get_safe_current_nonce()'s graceful handling. After
exhausting retries it returns [] (existing fall-through) instead of
aborting the whole run.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---
 protocols/safe/main.py  | 16 ++++++++++++++-
 tests/test_safe_main.py | 43 ++++++++++++++++++++++++++++++++++++++++-
 2 files changed, 57 insertions(+), 2 deletions(-)

diff --git a/protocols/safe/main.py b/protocols/safe/main.py
index ca604f2..a13ff14 100644
--- a/protocols/safe/main.py
+++ b/protocols/safe/main.py
@@ -59,7 +59,21 @@ def get_safe_transactions(
     }
 
     for attempt in range(max_retries):
-        response = requests.get(endpoint, params=params, headers=headers)
+        try:
+            response = requests.get(endpoint, params=params, headers=headers, timeout=10)
+        except requests.exceptions.RequestException as e:
+            # Transient transport failure (connection reset, read timeout, DNS).
+            # Retry with backoff instead of letting it bubble up and crash the run.
+            wait_time = 2**attempt
+            logger.warning(
+                "Request error talking to Safe API (%s), waiting %ss before retry (attempt %s/%s)...",
+                e,
+                wait_time,
+                attempt + 1,
+                max_retries,
+            )
+            time.sleep(wait_time)
+            continue
 
         if response.status_code == 200:
             return response.json()["results"]
diff --git a/tests/test_safe_main.py b/tests/test_safe_main.py
index b593126..7db0f62 100644
--- a/tests/test_safe_main.py
+++ b/tests/test_safe_main.py
@@ -1,7 +1,9 @@
 import importlib
 import os
 import unittest
-from unittest.mock import patch
+from unittest.mock import Mock, patch
+
+import requests
 
 
 class TestSafePendingTransactions(unittest.TestCase):
@@ -34,6 +36,45 @@ def test_current_nonce_advances_nonce_cache(self):
         mock_write.assert_called_once_with(safe_address, 22)
         self.assertEqual(pending, [{"nonce": 23}])
 
+    def test_get_safe_transactions_retries_on_connection_error(self):
+        safe_main = self._import_safe_main()
+
+        ok_response = Mock(status_code=200)
+        ok_response.json.return_value = {"results": [{"nonce": 5}]}
+
+        with (
+            patch.object(
+                safe_main.requests,
+                "get",
+                side_effect=[
+                    requests.exceptions.ConnectionError("Connection reset by peer"),
+                    ok_response,
+                ],
+            ) as mock_get,
+            patch.object(safe_main.time, "sleep") as mock_sleep,
+        ):
+            result = safe_main.get_safe_transactions("0xSafe", "arbitrum-main")
+
+        self.assertEqual(result, [{"nonce": 5}])
+        self.assertEqual(mock_get.call_count, 2)
+        mock_sleep.assert_called_once()
+
+    def test_get_safe_transactions_returns_empty_after_exhausting_retries(self):
+        safe_main = self._import_safe_main()
+
+        with (
+            patch.object(
+                safe_main.requests,
+                "get",
+                side_effect=requests.exceptions.ConnectionError("Connection reset by peer"),
+            ) as mock_get,
+            patch.object(safe_main.time, "sleep"),
+        ):
+            result = safe_main.get_safe_transactions("0xSafe", "arbitrum-main", max_retries=3)
+
+        self.assertEqual(result, [])
+        self.assertEqual(mock_get.call_count, 3)
+
     def test_executed_rows_are_filtered_even_when_api_returns_them(self):
         safe_main = self._import_safe_main()
         safe_address = "0xSafe"