[Bug] get_task_status returns 'pending' for both 'submitted' and 'running' states

[codeasier]

---

name: Bug Report
about: Create a report to help us improve AgentScope Runtime
title: '[Bug] get_task_status returns "pending" for both "submitted" and "running" states'
labels: 'bug'
assignees: ''

Security Issues: If this involves security vulnerabilities, report via Alibaba Security Response Center (ASRC) instead of creating a public issue.

Bug Description

The get_task_status() method in task_engine_mixin.py maps both submitted and running internal task states to a single pending response status. This prevents API consumers (e.g., CLI tools) from distinguishing between a task that is queued vs. one that is actively executing.

Affected Component

• Engine
• Sandbox
• Tools
• Common
• Documentation
• Other:___________

Reproduction Steps

1. Submit a background task via POST /agent/process/task (which internally calls submit_stream_query_task)
2. The task is stored in self.active_tasks with status: "submitted"
3. An async task starts executing via asyncio.create_task(self.execute_stream_query_task(...)), which updates the internal status to "running"
4. Poll the task status via GET /agent/process/task/{task_id} (which calls get_task_status)
5. Bug: Both states return {"status": "pending"}

Relevant code in agentscope_runtime/engine/deployers/utils/service_utils/routing/task_engine_mixin.py (memory mode branch, ~line 348-362):

def get_task_status(self, task_id: str):
    if self.celery_app:
        # Celery branch works correctly (PENDING/SUCCESS/FAILURE)
        ...
    else:
        # In-memory mode branch — BUG IS HERE
        if task_id not in self.active_tasks:
            return {"error": f"Task {task_id} not found"}

        task_info = self.active_tasks[task_id]
        task_status = task_info.get("status", "unknown")

        if task_status in ["submitted", "running"]:
            return {"status": "pending", "result": None}  # ← Both mapped to "pending"
        elif task_status == "completed":
            return {"status": "finished", "result": task_info.get("result")}
        elif task_status == "failed":
            return {"status": "error", "result": task_info.get("error")}
        else:
            return {"status": task_status, "result": None}

Meanwhile, the submit endpoint (agent_app.py ~line 568) and the execute method (execute_stream_query_task ~line 288) correctly set the internal status to "submitted" and "running" respectively:

# agent_app.py — submit
self.active_tasks[task_id] = {
    "status": "submitted",  # Correctly set
    ...
}

# task_engine_mixin.py — execute_stream_query_task
self.active_tasks[task_id].update({
    "status": "running",  # Correctly set
    ...
})

The internal state machine works correctly; the bug is only in the get_task_status() response mapping.

Expected vs Actual Behavior

Expected: API should return distinct statuses matching the internal state:

• submitted → {"status": "submitted"}
• running → {"status": "running"}
• completed → {"status": "finished"}
• failed → {"status": "error"}

This would allow consumers (CLI, UI, orchestrators) to show appropriate progress indicators like "Task submitted, waiting to start..." vs. "Task is running...".

Actual: Both submitted and running are collapsed into pending, so consumers cannot tell if a task is queued or actively executing.

Error Messages

No error — the API returns a valid response, but with incorrect state information. The CLI displays:

[TASK_ID: dc554d72-bf5b-416c-bc75-b6e077b10355]
[STATUS: pending]

⏸️  Task is pending in queue...

Even when the task has been actively running for several minutes (confirmed by session logs showing 68 rounds of tool calls executing).

Environment

• AgentScope Runtime Version: 1.1.3
• Python Version: 3.13.12
• OS: Linux 6.14.0-37-generic (x86_64)
• Installation: pip (via qwenpaw dependency)
• Run Mode: In-memory (non-Celery)

Additional Context

Real-world impact: This bug caused an orchestration failure in a multi-agent system. The calling agent (default) submitted a background task to a worker agent (spec-runner). Because get_task_status always returned pending, the caller assumed the worker was not executing the task and proceeded to execute it independently — violating the delegation contract.

Suggested fix: In task_engine_mixin.py, change the in-memory branch of get_task_status:

if task_status == "submitted":
    return {"status": "submitted", "result": None}
elif task_status == "running":
    return {"status": "running", "result": None}
elif task_status == "completed":
    return {"status": "finished", "result": task_info.get("result")}
elif task_status == "failed":
    return {"status": "error", "result": task_info.get("error", "Unknown error")}
else:
    return {"status": task_status, "result": None}

Evidence: Task session file at /home/codeasier/.copaw/workspaces/spec-runner/sessions/default--to--spec-runner--1776164674678--4599535c_*.json (150KB, 68 messages) confirms the task was actively executing for ~5 minutes while the API continuously returned pending.