langflow-ai/langflow · error · HTTPException

Graph not found

Error message

Graph not found

What it means

Raised as HTTP 404 with detail 'Graph not found' when chat_service.get_cache(flow_id) raises KeyError, meaning there is no compiled Graph in the cache for that flow id. The per-vertex build route requires a prior successful full-flow build to have populated the in-memory graph cache; this early get_cache uses strict KeyError semantics.

Source

Thrown at src/backend/base/langflow/api/v1/chat.py:487

        flow_id=flow_id,
        flow_user_id=flow.user_id,
        workspace_id=flow.workspace_id,
        folder_id=flow.folder_id,
    )

    chat_service = get_chat_service()
    telemetry_service = get_telemetry_service()
    flow_id_str = str(flow_id)

    next_runnable_vertices = []
    top_level_vertices = []
    start_time = time.perf_counter()
    error_message = None
    run_id = None
    try:
        graph: Graph = await chat_service.get_cache(flow_id_str)
    except KeyError as exc:
        raise HTTPException(status_code=404, detail="Graph not found") from exc

    try:
        cache = await chat_service.get_cache(flow_id_str)
        if isinstance(cache, CacheMiss):
            # If there's no cache
            await logger.awarning(f"No cache found for {flow_id_str}. Building graph starting at {vertex_id}")
            async with session_scope() as session:
                graph = await build_graph_from_db(
                    flow_id=flow_id,
                    session=session,
                    chat_service=chat_service,
                )
            run_id = str(uuid.uuid4())
            graph.set_run_id(run_id)
        else:
            graph = cache.get("result")
            await graph.initialize_run()
            run_id = graph.run_id

View on GitHub (pinned to 976ec789d2)

Solutions

1. Run a full flow build (POST /api/v1/chat/build/{flow_id}) first, wait for it to succeed, then build individual vertices.
2. Pin requests to the same worker in dev (single worker) or use the Redis cache backend so all workers share graph cache.
3. If the cache was flushed intentionally, re-build the flow before vertex-level calls.

Example fix

// before
await client.post(f"/api/v1/chat/build/{flow_id}/vertices/{vid}")  # 404 Graph not found

// after
await client.post(f"/api/v1/chat/build/{flow_id}")  # populate graph cache
await client.post(f"/api/v1/chat/build/{flow_id}/vertices/{vid}")

# Ensure the graph is cached before vertex-level builds
build = await client.post(f"/api/v1/chat/build/{flow_id}")
build.raise_for_status()
# only now issue per-vertex requests

try:
    await client.post(f"/api/v1/chat/build/{flow_id}/vertices/{vid}")
except httpx.HTTPStatusError as e:
    if e.response.status_code == 404 and e.response.json()["detail"] == "Graph not found":
        await client.post(f"/api/v1/chat/build/{flow_id}")  # rebuild cache, then retry once
        return await client.post(f"/api/v1/chat/build/{flow_id}/vertices/{vid}")
    raise

When it happens

Trigger: Calling POST /build/{flow_id}/vertices/{vertex_id} before any successful POST /build/{flow_id} on that server process, or after a server restart / cache flush that dropped the graph cache.

Common situations: Restarting the backend between building the flow and building a single vertex; running multiple workers so the vertex request lands on a worker without the cached graph; cache TTL expiry under load.

Related errors

• Job not found: {exc!s}
• Job not found
• Ingestion run not found.
• Knowledge bases not found: {}
• No ingestion job found for the knowledge base {kb_name}

AI-assisted analysis of langflow-ai/langflow@976ec789d2 (2026-08-14). Data as JSON: /api/errors/dd6401ffcef7db74. Report an issue: GitHub.