666ghj/MiroFish · error · AssertionError

production edge pagination did not match raw cursor traversa

Error message

production edge pagination did not match raw cursor traversal

What it means

The edge counterpart of the node cross-check: the UUID set from fetch_all_edges must exactly match the set from _raw_pages over graph.edge.with_raw_response.get_by_graph_id. A mismatch means the production edge paginator and the manual raw-cursor walk observed different edges — dropped pages, duplicates, or edge churn between the two traversals.

Source

Thrown at backend/scripts/validate_zep_cloud_integration.py:559

            "status": client.batch.get(batch_id=batch_id).status,
            "item_count": len(listed_items),
            "item_pages_at_size_3": batch_pages,
            "episode_uuids": baseline_episode_uuids,
        }
        print(f"[zep-deep] batch completed pages={batch_pages}", flush=True)

        baseline_nodes = fetch_all_nodes(client, graph_id, page_size=2)
        baseline_edges = fetch_all_edges(client, graph_id, page_size=2)
        raw_nodes, node_pages = _raw_pages(
            client.graph.node.with_raw_response.get_by_graph_id, graph_id, page_size=2
        )
        raw_edges, edge_pages = _raw_pages(
            client.graph.edge.with_raw_response.get_by_graph_id, graph_id, page_size=2
        )
        if {_uuid(item) for item in baseline_nodes} != {_uuid(item) for item in raw_nodes}:
            raise AssertionError("production node pagination did not match raw cursor traversal")
        if {_uuid(item) for item in baseline_edges} != {_uuid(item) for item in raw_edges}:
            raise AssertionError("production edge pagination did not match raw cursor traversal")
        if len(baseline_nodes) <= 2 or len(baseline_edges) <= 2:
            raise AssertionError("the corpus did not produce enough artifacts to exercise pagination")

        baseline_names = {_uuid(node): node.name for node in baseline_nodes}
        baseline_ceo = client.graph.search(
            graph_id=graph_id,
            query="截至2026年4月底，谁担任澜舟科技首席执行官？",
            scope="edges",
            reranker="cross_encoder",
            limit=10,
        )
        result["baseline"] = {
            "node_count": len(baseline_nodes),
            "edge_count": len(baseline_edges),
            "node_pages_at_size_2": node_pages,
            "edge_pages_at_size_2": edge_pages,
            "invalidated_edge_count": sum(bool(edge.invalid_at) for edge in baseline_edges),
            "ceo_search": _search_view(baseline_ceo, baseline_names),

View on GitHub (pinned to b5b53acc57)

Solutions

Wait for every episode to reach processed=True and stop all writers before running the validation.
Upgrade zep-cloud and rerun on a quiesced graph.
If reproducible, diff both edge UUID sets to identify dropped/extra edges, then escalate to Zep support.

Defensive patterns

Strategy: retry

Try / catch

try:
    assert_equal_uuid_sets(baseline_edges, raw_edges)
except AssertionError as e:
    if "edge pagination did not match" in str(e):
        time.sleep(10)  # let async edge derivation settle
        baseline_edges = fetch_all_edges(client, graph_id, page_size=2)
        raw_edges, _ = _raw_pages(client.graph.edge.with_raw_response.get_by_graph_id, graph_id, page_size=2)
        assert_equal_uuid_sets(baseline_edges, raw_edges)
    else:
        raise

Prevention

Wait for episode processing AND edge derivation to settle before comparing edge listings.
Never run the updater concurrently with pagination consistency checks.
Diff both edge UUID sets when a mismatch repeats on an idle graph, then escalate.

When it happens

Trigger: Edges being created/invalidated concurrently (Zep derives edges asynchronously from episodes still processing); one pagination path mis-handling the zep-next-cursor for edges.

Common situations: A still-processing episode adds edges mid-check; the updater runs during validation; SDK version drift affecting edge pagination only.

Related errors

AI-assisted analysis of 666ghj/MiroFish@b5b53acc57 (2026-08-14). Data as JSON: /api/errors/4edc9ee9dd337ddb. Report an issue: GitHub.