From 13e871f2c0ff986016ac4f34e1b273f7745668d0 Mon Sep 17 00:00:00 2001
From: David Traina <44659830+DavidTraina@users.noreply.github.com>
Date: Thu, 5 Nov 2026 00:39:30 -0600
Subject: [PATCH] fix(batch-evaluation): use a 60 s timeout for observation
 page fetches

The observations v2 API can scan for up to 45 s before returning a page,
but page fetches used the client-wide default timeout of 5 s. Timeouts are
not retried, so the run stopped, and resuming repeated the same request.
Send a per-request timeout above the server budget without changing the
global timeout, and drop the 'after N retries' wording from the fetch
failure log, since timeouts and 422s are never retried.
---
 langfuse/batch_evaluation.py        | 12 ++++++++----
 tests/unit/test_batch_evaluation.py |  5 ++++-
 2 files changed, 12 insertions(+), 5 deletions(-)

diff --git a/langfuse/batch_evaluation.py b/langfuse/batch_evaluation.py
index 4d7bf3371..3e7540a31 100644
--- a/langfuse/batch_evaluation.py
+++ b/langfuse/batch_evaluation.py
@@ -38,6 +38,9 @@
 metadata, model, usage, prompt, metrics, trace_context.
 """
 
+# The v2 observations API scans for up to 45 s before it returns a page.
+_OBSERVATIONS_FETCH_TIMEOUT_SECONDS = 60
+
 
 def _with_time_fields(fields: Optional[str]) -> str:
     """Add the ``time`` group, whose ``updatedAt`` picks the newest duplicate row."""
@@ -962,9 +965,7 @@ def build_resume_token() -> BatchEvaluationResumeToken:
                     fields=fields,
                 )
             except Exception as e:
-                logger.error(
-                    "Failed to fetch batch after %s retries: %s", max_retries, e
-                )
+                logger.error("Failed to fetch batch: %s", e)
 
                 return self._build_result(
                     total_items_fetched=total_items_fetched,
@@ -1117,7 +1118,10 @@ async def _fetch_batch_with_retry(
             limit=limit,
             cursor=cursor,
             filter=filter,
-            request_options={"max_retries": max_retries},
+            request_options={
+                "max_retries": max_retries,
+                "timeout_in_seconds": _OBSERVATIONS_FETCH_TIMEOUT_SECONDS,
+            },
         )
 
         return list(response.data), response.meta.cursor
diff --git a/tests/unit/test_batch_evaluation.py b/tests/unit/test_batch_evaluation.py
index f6c29f4f6..3f3265ba1 100644
--- a/tests/unit/test_batch_evaluation.py
+++ b/tests/unit/test_batch_evaluation.py
@@ -255,7 +255,10 @@ async def test_fetch_failure_returns_resume_token_for_failed_page():
 
     assert resumed.completed is True
     assert set(resumed.item_evaluations) == {"obs-3", "obs-2", "obs-1", "obs-0"}
-    assert api.calls[-1]["request_options"] == {"max_retries": 3}
+    assert api.calls[-1]["request_options"] == {
+        "max_retries": 3,
+        "timeout_in_seconds": 60,
+    }
 
 
 @pytest.mark.asyncio