Add timeout and retries for DSN config fetch from GitHub API

src/sentry_config.py

@@ -3,8 +3,8 @@
 import base64
 import logging
 import os
+import time
 from configparser import ConfigParser
-from functools import lru_cache
 
 import requests
 
@@ -15,6 +15,33 @@
 SENTRY_CONFIG_API_URL = (
     "https://api.github.com/repos/{owner}/.sentry/contents/sentry_config.ini"
 )
+GITHUB_API_TIMEOUT_SECONDS = float(os.environ.get("GITHUB_API_TIMEOUT_SECONDS", "10"))
+GITHUB_API_MAX_RETRIES = int(os.environ.get("GITHUB_API_MAX_RETRIES", "2"))
+GITHUB_API_RETRY_BACKOFF_SECONDS = float(
+    os.environ.get("GITHUB_API_RETRY_BACKOFF_SECONDS", "0.5")
+)
+
+
+def _github_get_with_retries(url: str, headers: dict[str, str]) -> requests.Response:
+    total_attempts = GITHUB_API_MAX_RETRIES + 1
+    for attempt in range(1, total_attempts + 1):
+        try:
+            return requests.get(
+                url,
+                headers=headers,
+                timeout=GITHUB_API_TIMEOUT_SECONDS,
+            )
+        except (requests.exceptions.ConnectionError, requests.exceptions.Timeout):
+            if attempt == total_attempts:
+                raise
+            wait_seconds = GITHUB_API_RETRY_BACKOFF_SECONDS * (2 ** (attempt - 1))
+            logger.warning(
+                "GitHub API request failed; retrying in %.1fs (%s/%s)",
+                wait_seconds,
+                attempt + 1,
+                total_attempts,
+            )
+            time.sleep(wait_seconds)
 
 
 def fetch_dsn_for_github_org(org: str, token: str) -> str:
@@ -27,7 +54,7 @@
         api_url = SENTRY_CONFIG_API_URL.replace("{owner}", org)
 
         # - Get meta about sentry_config.ini file
-        resp = requests.get(api_url, headers=headers)
+        resp = _github_get_with_retries(api_url, headers)
         resp.raise_for_status()
         meta = resp.json()
 

tests/test_sentry_config_file.py

@@ -1,10 +1,16 @@
 from __future__ import annotations
 
+from unittest.mock import Mock
+from unittest.mock import patch
 from unittest import TestCase
 
+import requests
 import responses
 
 from src.sentry_config import fetch_dsn_for_github_org
+from src.sentry_config import GITHUB_API_MAX_RETRIES
+from src.sentry_config import GITHUB_API_RETRY_BACKOFF_SECONDS
+from src.sentry_config import GITHUB_API_TIMEOUT_SECONDS
 from src.sentry_config import SENTRY_CONFIG_API_URL as api_url
 
 expected_dsn = (
@@ -55,3 +61,39 @@ def test_file_missing(self) -> None:
 
     def test_bad_contents(self) -> None:
         pass
+
+    @patch("src.sentry_config.time.sleep")
+    @patch("src.sentry_config.requests.get")
+    def test_fetch_retries_connection_error_once(
+        self,
+        mock_get: Mock,
+        mock_sleep: Mock,
+    ) -> None:
+        response = Mock()
+        response.raise_for_status.return_value = None
+        response.json.return_value = sentry_config_file_meta
+        mock_get.side_effect = [
+            requests.exceptions.ConnectionError("connection reset"),
+            response,
+        ]
+
+        assert fetch_dsn_for_github_org(org, token) == expected_dsn
+        assert mock_get.call_count == 2
+        for call in mock_get.call_args_list:
+            assert call.kwargs["timeout"] == GITHUB_API_TIMEOUT_SECONDS
+        mock_sleep.assert_called_once_with(GITHUB_API_RETRY_BACKOFF_SECONDS)
+
+    @patch("src.sentry_config.time.sleep")
+    @patch("src.sentry_config.requests.get")
+    def test_fetch_raises_after_connection_error_retries_exhausted(
+        self,
+        mock_get: Mock,
+        mock_sleep: Mock,
+    ) -> None:
+        mock_get.side_effect = requests.exceptions.ConnectionError("connection reset")
+
+        with self.assertRaises(requests.exceptions.ConnectionError):
+            fetch_dsn_for_github_org(org, token)
+
+        assert mock_get.call_count == GITHUB_API_MAX_RETRIES + 1
+        assert mock_sleep.call_count == GITHUB_API_MAX_RETRIES