From 3eaf3b91513d13074caafe73f99a3c25d0848166 Mon Sep 17 00:00:00 2001
From: Brad Stein <Brad.Stein@gmail.com>
Date: Wed, 21 Jan 2026 18:21:42 -0300
Subject: [PATCH] portal: log and retry ariadne calls

---
 backend/atlas_portal/ariadne_client.py | 45 ++++++++++++++++++++------
 backend/atlas_portal/settings.py       |  2 ++
 2 files changed, 37 insertions(+), 10 deletions(-)

diff --git a/backend/atlas_portal/ariadne_client.py b/backend/atlas_portal/ariadne_client.py
index 95b68d1..42aff29 100644
--- a/backend/atlas_portal/ariadne_client.py
+++ b/backend/atlas_portal/ariadne_client.py
@@ -1,5 +1,7 @@
 from __future__ import annotations
 
+import logging
+import time
 from typing import Any
 
 import httpx
@@ -7,6 +9,8 @@ from flask import jsonify, request
 
 from . import settings
 
+logger = logging.getLogger(__name__)
+
 
 class AriadneError(Exception):
     def __init__(self, message: str, status_code: int = 502) -> None:
@@ -39,17 +43,38 @@ def request_raw(
     if not enabled():
         raise AriadneError("ariadne not configured", 503)
 
-    try:
-        with httpx.Client(timeout=settings.ARIADNE_TIMEOUT_SEC) as client:
-            return client.request(
-                method,
-                _url(path),
-                headers=_auth_headers(),
-                json=payload,
-                params=params,
+    url = _url(path)
+    attempts = max(1, settings.ARIADNE_RETRY_COUNT)
+    for attempt in range(1, attempts + 1):
+        try:
+            with httpx.Client(timeout=settings.ARIADNE_TIMEOUT_SEC) as client:
+                resp = client.request(
+                    method,
+                    url,
+                    headers=_auth_headers(),
+                    json=payload,
+                    params=params,
+                )
+            if resp.status_code >= 500:
+                logger.warning(
+                    "ariadne error response",
+                    extra={"method": method, "path": path, "status": resp.status_code},
+                )
+            return resp
+        except httpx.RequestError as exc:
+            logger.warning(
+                "ariadne request failed",
+                extra={
+                    "method": method,
+                    "path": path,
+                    "attempt": attempt,
+                    "timeout_sec": settings.ARIADNE_TIMEOUT_SEC,
+                    "error": str(exc),
+                },
             )
-    except httpx.RequestError as exc:
-        raise AriadneError("ariadne unavailable", 502) from exc
+            if attempt >= attempts:
+                raise AriadneError("ariadne unavailable", 502) from exc
+            time.sleep(settings.ARIADNE_RETRY_BACKOFF_SEC * attempt)
 
 
 def proxy(
diff --git a/backend/atlas_portal/settings.py b/backend/atlas_portal/settings.py
index c704558..e2ff3f5 100644
--- a/backend/atlas_portal/settings.py
+++ b/backend/atlas_portal/settings.py
@@ -50,6 +50,8 @@ KEYCLOAK_ADMIN_REALM = os.getenv("KEYCLOAK_ADMIN_REALM", KEYCLOAK_REALM)
 
 ARIADNE_URL = os.getenv("ARIADNE_URL", "").strip()
 ARIADNE_TIMEOUT_SEC = float(os.getenv("ARIADNE_TIMEOUT_SEC", "10"))
+ARIADNE_RETRY_COUNT = int(os.getenv("ARIADNE_RETRY_COUNT", "2"))
+ARIADNE_RETRY_BACKOFF_SEC = float(os.getenv("ARIADNE_RETRY_BACKOFF_SEC", "0.2"))
 
 ACCOUNT_ALLOWED_GROUPS = [
     g.strip()