feat: retry transient failures during upstream sync

application/cmd/cre_main.py

@@ -38,6 +38,50 @@
 app = None
 
 
+def fetch_upstream_json(
+    path: str,
+    timeout: Optional[float] = None,
+    max_attempts: Optional[int] = None,
+    backoff_seconds: Optional[float] = None,
+) -> Dict[str, Any]:
+    base_url = os.environ.get("CRE_UPSTREAM_API_URL", "https://opencre.org/rest/v1")
+    timeout = timeout or float(os.environ.get("CRE_UPSTREAM_TIMEOUT_SECONDS", "30"))
+    max_attempts = max_attempts or int(os.environ.get("CRE_UPSTREAM_MAX_ATTEMPTS", "4"))
+    backoff_seconds = backoff_seconds or float(
+        os.environ.get("CRE_UPSTREAM_RETRY_BACKOFF_SECONDS", "2")
+    )
+    url = f"{base_url}{path}"
+    last_error: Optional[Exception] = None
+
+    for attempt in range(1, max_attempts + 1):
+        try:
+            response = requests.get(url, timeout=timeout)
+            if response.status_code == 200:
+                return response.json()
+
+            status_error = RuntimeError(
+                f"cannot connect to upstream status code {response.status_code}"
+            )
+            if response.status_code < 500 and response.status_code != 429:
+                raise status_error
+            last_error = status_error
+        except requests.exceptions.RequestException as exc:
+            last_error = exc
+
+        if attempt < max_attempts:
+            logger.warning(
+                "upstream fetch failed for %s on attempt %s/%s, retrying",
+                url,
+                attempt,
+                max_attempts,
+            )
+            time.sleep(backoff_seconds * attempt)
+
+    if last_error:
+        raise RuntimeError(f"upstream fetch failed for {url}") from last_error
+    raise RuntimeError(f"upstream fetch failed for {url}")
+
+
 def register_node(node: defs.Node, collection: db.Node_collection) -> db.Node:
     """
     for each link find if either the root node or the link have a CRE,
@@ -591,15 +635,7 @@ def download_graph_from_upstream(cache: str) -> None:
     collection = db_connect(path=cache).with_graph()
 
     def download_cre_from_upstream(creid: str):
-        cre_response = requests.get(
-            os.environ.get("CRE_UPSTREAM_API_URL", "https://opencre.org/rest/v1")
-            + f"/id/{creid}"
-        )
-        if cre_response.status_code != 200:
-            raise RuntimeError(
-                f"cannot connect to upstream status code {cre_response.status_code}"
-            )
-        data = cre_response.json()
+        data = fetch_upstream_json(f"/id/{creid}")
         credict = data["data"]
         cre = defs.Document.from_dict(credict)
         if cre.id in imported_cres:
@@ -611,15 +647,7 @@ def download_cre_from_upstream(creid: str):
             if link.document.doctype == defs.Credoctypes.CRE:
                 download_cre_from_upstream(link.document.id)
 
-    root_cres_response = requests.get(
-        os.environ.get("CRE_UPSTREAM_API_URL", "https://opencre.org/rest/v1")
-        + "/root_cres"
-    )
-    if root_cres_response.status_code != 200:
-        raise RuntimeError(
-            f"cannot connect to upstream status code {root_cres_response.status_code}"
-        )
-    data = root_cres_response.json()
+    data = fetch_upstream_json("/root_cres")
     for root_cre in data["data"]:
         cre = defs.Document.from_dict(root_cre)
         register_cre(cre, collection)

application/tests/cre_main_test.py

@@ -6,6 +6,7 @@
 from typing import Any, Dict, List
 from unittest import mock
 from unittest.mock import Mock, patch
+import requests
 from rq import Queue, job
 from application.utils import redis
 from application.prompt_client import prompt_client as prompt_client
@@ -470,6 +471,23 @@ def test_register_cre(self) -> None:
             ],
         )
 
+    @patch("application.cmd.cre_main.time.sleep")
+    @patch("application.cmd.cre_main.requests.get")
+    def test_fetch_upstream_json_retries_transient_failures(
+        self, mock_get, mock_sleep
+    ) -> None:
+        transient_error = requests.exceptions.ConnectionError("reset by peer")
+        success_response = Mock()
+        success_response.status_code = 200
+        success_response.json.return_value = {"data": []}
+        mock_get.side_effect = [transient_error, success_response]
+
+        data = main.fetch_upstream_json("/root_cres")
+
+        self.assertEqual(data, {"data": []})
+        self.assertEqual(mock_get.call_count, 2)
+        mock_sleep.assert_called_once()
+
     @patch.object(main, "db_connect")
     @patch.object(Queue, "enqueue_call")
     @patch.object(redis, "wait_for_jobs")