From e58d4f1cc5f8d2763ac0a3d987199e71d09017bb Mon Sep 17 00:00:00 2001
From: harshitanand <harshitanand@users.noreply.github.com>
Date: Fri, 10 Apr 2026 13:51:52 +0530
Subject: [PATCH 1/2] fix: add max_items limit to fetch_all_edges to prevent
 unbounded memory growth (#513)

fetch_all_nodes already had a max_items guard (default 2000) but
fetch_all_edges had no such safeguard, allowing unbounded memory growth
on graphs with large numbers of edges.

Add _MAX_EDGES = 5000 constant and mirror the same loop-guard pattern
from fetch_all_nodes: cap the result list, emit a warning log, and
break pagination once the limit is reached.
---
 backend/app/utils/zep_paging.py | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)

diff --git a/backend/app/utils/zep_paging.py b/backend/app/utils/zep_paging.py
index 943cd1ae..8e40226c 100644
--- a/backend/app/utils/zep_paging.py
+++ b/backend/app/utils/zep_paging.py
@@ -19,6 +19,7 @@ logger = get_logger('mirofish.zep_paging')
 
 _DEFAULT_PAGE_SIZE = 100
 _MAX_NODES = 2000
+_MAX_EDGES = 5000
 _DEFAULT_MAX_RETRIES = 3
 _DEFAULT_RETRY_DELAY = 2.0  # seconds, doubles each retry
 
@@ -106,10 +107,11 @@ def fetch_all_edges(
     client: Zep,
     graph_id: str,
     page_size: int = _DEFAULT_PAGE_SIZE,
+    max_items: int = _MAX_EDGES,
     max_retries: int = _DEFAULT_MAX_RETRIES,
     retry_delay: float = _DEFAULT_RETRY_DELAY,
 ) -> list[Any]:
-    """分页获取图谱所有边，返回完整列表。每页请求自带重试。"""
+    """分页获取图谱所有边，最多返回 max_items 条（默认 5000）。每页请求自带重试。"""
     all_edges: list[Any] = []
     cursor: str | None = None
     page_num = 0
@@ -132,6 +134,10 @@ def fetch_all_edges(
             break
 
         all_edges.extend(batch)
+        if len(all_edges) >= max_items:
+            all_edges = all_edges[:max_items]
+            logger.warning(f"Edge count reached limit ({max_items}), stopping pagination for graph {graph_id}")
+            break
         if len(batch) < page_size:
             break
 

From 7d6f6bb1354b119ca952be32bfe347314dd2ec7a Mon Sep 17 00:00:00 2001
From: 666ghj <670939375@qq.com>
Date: Wed, 22 Jul 2026 01:22:47 +0800
Subject: [PATCH 2/2] fix: preserve edge paging call compatibility

---
 backend/app/utils/zep_paging.py       |  2 +-
 backend/tests/test_zep_edge_paging.py | 44 +++++++++++++++++++++++++++
 2 files changed, 45 insertions(+), 1 deletion(-)
 create mode 100644 backend/tests/test_zep_edge_paging.py

diff --git a/backend/app/utils/zep_paging.py b/backend/app/utils/zep_paging.py
index 8e40226c..22c20526 100644
--- a/backend/app/utils/zep_paging.py
+++ b/backend/app/utils/zep_paging.py
@@ -107,9 +107,9 @@ def fetch_all_edges(
     client: Zep,
     graph_id: str,
     page_size: int = _DEFAULT_PAGE_SIZE,
-    max_items: int = _MAX_EDGES,
     max_retries: int = _DEFAULT_MAX_RETRIES,
     retry_delay: float = _DEFAULT_RETRY_DELAY,
+    max_items: int = _MAX_EDGES,
 ) -> list[Any]:
     """分页获取图谱所有边，最多返回 max_items 条（默认 5000）。每页请求自带重试。"""
     all_edges: list[Any] = []
diff --git a/backend/tests/test_zep_edge_paging.py b/backend/tests/test_zep_edge_paging.py
new file mode 100644
index 00000000..d0e92ff4
--- /dev/null
+++ b/backend/tests/test_zep_edge_paging.py
@@ -0,0 +1,44 @@
+from types import SimpleNamespace
+
+from app.utils import zep_paging
+
+
+def _client():
+    edge_api = SimpleNamespace(get_by_graph_id=lambda *args, **kwargs: [])
+    return SimpleNamespace(graph=SimpleNamespace(edge=edge_api))
+
+
+def test_edge_cap_stops_pagination_at_requested_limit(monkeypatch):
+    pages = [
+        [SimpleNamespace(uuid_="e1"), SimpleNamespace(uuid_="e2")],
+        [SimpleNamespace(uuid_="e3"), SimpleNamespace(uuid_="e4")],
+    ]
+    calls = []
+
+    def fake_fetch(*args, **kwargs):
+        calls.append(kwargs)
+        return pages[len(calls) - 1]
+
+    monkeypatch.setattr(zep_paging, "_fetch_page_with_retry", fake_fetch)
+
+    result = zep_paging.fetch_all_edges(_client(), "graph", page_size=2, max_items=3)
+
+    assert [edge.uuid_ for edge in result] == ["e1", "e2", "e3"]
+    assert len(calls) == 2
+    assert calls[1]["uuid_cursor"] == "e2"
+
+
+def test_existing_positional_retry_arguments_keep_their_meaning(monkeypatch):
+    observed = {}
+
+    def fake_fetch(*args, **kwargs):
+        observed.update(kwargs)
+        return []
+
+    monkeypatch.setattr(zep_paging, "_fetch_page_with_retry", fake_fetch)
+
+    zep_paging.fetch_all_edges(_client(), "graph", 25, 7, 0.25)
+
+    assert observed["limit"] == 25
+    assert observed["max_retries"] == 7
+    assert observed["retry_delay"] == 0.25