模拟

后端 API

本地执行与 Pod 执行通过 llm_d_lens.simulation.progress.report_request_progress 共享请求进度报告。Pod 报告仍会等待就绪；本地报告则保留其原始开始时间。后端适配器通过 backends.analytics.artifact_error_message 共享产物错误格式化逻辑，同时各自保留自身的错误/超时字段优先级、状态映射和指标单位。任务 API 与后端环境准备逻辑保持不变。

模拟相关路由位于 llm_d_lens/simulation/router.py，前缀为 /api/simulation。

列出模拟后端

列出 Lens 已知的模拟（Simulation）后端，以及每个后端在当前环境中是否可用。

REST API
Python API

GET /api/simulation/backends

请求：无请求体。

响应：

{
  "backends": [
    {
      "name": "aiperf",
      "display_name": "AIPerf",
      "api_version": 1,
      "available": true,
      "version": "0.12.0",
      "unavailable_reason": null,
      "scenarios": [
        {
          "name": "chat",
          "display_name": "Chat",
          "description": "Conversational and general assistant traffic.",
          "supported_backends": []
        },
        {
          "name": "api-calling",
          "display_name": "Tool & API Use",
          "description": "Tool-use, agent, and business API traffic.",
          "supported_backends": []
        }
      ],
      "capabilities": {
        "prompt_kinds": ["trace"],
        "arrival_patterns": [],
        "supports_duration": true,
        "supports_request_count": false,
        "supports_request_rate": false,
        "supports_streaming": true,
        "supports_per_request_results": true,
        "supports_cancellation": true,
        "supports_trace_range": true,
        "scale_factor_trace_formats": ["baseten_trace"],
        "trace_format_capabilities": {
          "mooncake_trace": {"supports_trace_range": true, "supports_scale_factor": false}
        },
        "advanced_options": [],
        "trace_formats": [
          "mooncake_trace",
          "bailian_trace",
          "baseten_trace",
          "burst_gpt_trace",
          "weka_public_dataset"
        ]
      },
      "managed_installation": false
    }
  ]
}

from llm_d_lens.simulation.backends import list_backends

for backend in list_backends():
    print(backend.name, backend.available, backend.version)

发现端点模型

探测某个模型端点，发现它报告了哪些模型。可选的 api_key 以 Bearer 令牌发送，不包含在响应中。连接失败、认证失败或无效模型列表返回 HTTP 400。

REST API
Python API

POST /api/simulation/models/discover

请求体：ModelDiscoveryRequest

endpoint_url (string，必填)
api_key（string，可选，最多 4096 个字符；仅用于本次请求）

{
  "endpoint_url": "http://127.0.0.1:38000/v1"
}

响应：

{
  "endpoint": "http://127.0.0.1:38000/v1",
  "models": [
    "Qwen/Qwen3-8B",
    "Qwen/Qwen3-8B-Instruct"
  ]
}

import asyncio

from llm_d_lens.simulation.endpoints import discover_models

async def main() -> None:
    models = await discover_models("http://127.0.0.1:38000/v1", api_key="your-model-access-token")
    print(models["endpoint"], models["models"])

asyncio.run(main())

列出模拟场景

列出 Lens 暴露的模拟场景目录，例如聊天（chat）、API 调用（API-calling）和代码生成（coding）。

REST API
Python API

GET /api/simulation/scenarios

请求：无请求体。

响应：

{
  "scenarios": [
    {
      "name": "chat",
      "display_name": "Chat",
      "description": "Conversational and general assistant traffic.",
      "supported_backends": ["aiperf", "trace-replayer"]
    },
    {
      "name": "coding",
      "display_name": "Code Generation",
      "description": "Code generation and software engineering traffic.",
      "supported_backends": ["aiperf"]
    }
  ]
}

from llm_d_lens.simulation.backends import list_scenarios

for scenario in list_scenarios():
    print(scenario.name, scenario.supported_backends)

列出跟踪数据集

列出已注册的跟踪（trace）数据集，以及能够回放每个数据集的后端。

REST API
Python API

GET /api/simulation/trace-datasets

请求：无请求体。

响应：

{
  "data_dir": "/home/prism/.local/share/prism/data/datasets",
  "datasets": [
    {
      "name": "mooncake-arxiv",
      "filename": "mooncake_trace.jsonl",
      "scenario": "chat",
      "trace_format": "mooncake_trace",
      "description": "Mooncake chat trace derived from Arxiv summarization prompts.",
      "downloaded": false,
      "path": "/home/prism/.local/share/prism/data/datasets/mooncake_trace.jsonl",
      "artifact_uri": "artifact://dataset/mooncake-arxiv/mooncake_trace.jsonl",
      "size_bytes": null,
      "license": "Apache-2.0",
      "sha256": "f0c15fb14b60d0afca8b1f7c1ff29378f3df54318e6d112b47ad3b2a2f5881ef",
      "source_repository": "https://github.com/kvcache-ai/Mooncake.git",
      "supported_backends": ["aiperf", "trace-replayer"]
    }
  ]
}

from llm_d_lens.simulation.backends import list_backends
from llm_d_lens.simulation.traces import trace_registry

backends = list_backends()
for dataset in trace_registry.list():
    supported = [
        backend.name
        for backend in backends
        if any(scenario.name == dataset["scenario"] for scenario in backend.scenarios)
        and dataset["trace_format"] in backend.capabilities.get("trace_formats", [])
    ]
    print(dataset["name"], supported)

下载跟踪数据集

将已注册的跟踪数据集下载到 Lens 的本地跟踪存储中。

REST API
Python API

POST /api/simulation/trace-datasets/download

请求体：TraceDatasetDownloadRequest

dataset (string，必填)
force (boolean，默认 false)

{
  "dataset": "mooncake-arxiv",
  "force": false
}

响应：

{
  "artifact_uri": "artifact://dataset/mooncake-arxiv/mooncake_trace.jsonl",
  "path": "/home/prism/.local/share/prism/data/datasets/mooncake_trace.jsonl",
  "trace_format": "mooncake_trace",
  "sha256": "f0c15fb14b60d0afca8b1f7c1ff29378f3df54318e6d112b47ad3b2a2f5881ef",
  "metadata_path": "/home/prism/.local/share/prism/data/datasets/mooncake_trace.jsonl.metadata.json",
  "size_bytes": 1048576
}

import asyncio

from llm_d_lens.simulation.traces import BaseTrace

async def main() -> None:
    result = await BaseTrace.download("mooncake-arxiv", force=False)
    print(result["path"], result["sha256"])

asyncio.run(main())

获取跟踪数据集时间线

获取某个跟踪数据集的预聚合时间线视图。

REST API
Python API

GET /api/simulation/trace-datasets/{dataset_name}/timeline

请求：无请求体。

路径参数：dataset_name（string，必填）
查询参数：bins（integer，默认 800，范围 50-2000）

响应：

{
  "dataset": "mooncake-arxiv",
  "trace_format": "mooncake_trace",
  "request_count": 12000,
  "duration_seconds": 3600.0,
  "source_size_bytes": 1048576,
  "bin_width_seconds": 4.5,
  "bins": [
    {"start_seconds": 0.0, "end_seconds": 4.5, "request_count": 12},
    {"start_seconds": 4.5, "end_seconds": 9.0, "request_count": 18}
  ]
}

import asyncio

from llm_d_lens.simulation.traces import get_trace_timeline

async def main() -> None:
    timeline = await get_trace_timeline("mooncake-arxiv", 800)
    print(timeline["dataset"], timeline["request_count"])

asyncio.run(main())

列出模拟任务

列出模拟（Simulation）任务，可按状态、场景、后端、模型、数据集、端点和创建时间区间等可选条件进行过滤。

REST API
Python API

GET /api/simulation/tasks

请求：无请求体。

查询参数：status、scenario、search、backend、model、dataset、trace_format、endpoint、created_after、created_before、offset、limit。

响应：

{
  "tasks": [
    {
      "id": "1a2b3c4d",
      "name": "Trace simulation",
      "description": "Replay a chat trace against a ready deployment",
      "scenario": "chat",
      "status": "running",
      "endpoint_mode": "deployment",
      "endpoint_namespace": "prism-sim",
      "endpoint_service": "qwen-router",
      "endpoint_deployment_execution_id": "00000000-0000-4000-8000-000000000001",
      "endpoint_deployment_run_id": null,
      "endpoint_deployment_case_id": null,
      "endpoint_cluster_id": "cluster-1",
      "endpoint_cluster_name": "dev-cluster",
      "endpoint_deployment_name": "qwen3-guide",
      "endpoint_url": "http://127.0.0.1:38000/v1",
      "model_name": "Qwen/Qwen3-8B",
      "simulation": {
        "backend": "trace-replayer",
        "backend_options": {},
        "duration_seconds": 60.0,
        "num_requests": null,
        "stream": true,
        "warmup_enabled": false,
        "grace_period_seconds": 30.0
      },
      "prompt": {
        "type": "trace",
        "dataset": {"name": "mooncake-arxiv", "scenario": "chat", "tokenizer": null},
        "trace": {
          "path": "/home/prism/.local/share/prism/data/datasets/mooncake_trace.jsonl",
          "format": "mooncake_trace",
          "fixed_schedule": true,
          "timeout_seconds": 3600.0,
          "synthesis_speedup_ratio": 1.0,
          "start_seconds": 0.0,
          "end_seconds": null
        }
      },
      "task_dir": "/home/prism/.local/share/prism/data/artifacts/simulations/1a2b3c4d",
      "artifact_uri": null,
      "artifact_error": null,
      "logs_truncated": false,
      "artifacts_incomplete": false,
      "backend_version_inferred": false,
      "backend_version": null,
      "configuration_ids": ["config-1"],
      "progress_percent": 42.5,
      "progress_message": "Simulation running",
      "logs": [],
      "result": null,
      "error_message": null,
      "created_at": "2026-09-23T09:18:24.287000Z",
      "started_at": "2026-09-23T09:18:25.000000Z",
      "execution_started_at": "2026-09-23T09:18:27.000000Z",
      "completed_at": null,
      "owner_pid": null,
      "owner_instance_id": null,
      "live_summary": {
        "completed_requests": 420,
        "successful_requests": 418
      }
    }
  ],
  "total": 1,
  "offset": 0,
  "limit": 10,
  "counts": {
    "queued": 0,
    "running": 1,
    "completed": 0,
    "failed": 0,
    "cancelled": 0
  },
  "facets": {
    "backends": ["trace-replayer"],
    "models": ["Qwen/Qwen3-8B"],
    "datasets": ["mooncake-arxiv"],
    "trace_formats": ["mooncake_trace"],
    "endpoints": ["http://127.0.0.1:38000/v1"]
  }
}

import asyncio

from llm_d_lens.simulation.service import list_tasks

async def main() -> None:
    page = await list_tasks(status="running", limit=20)
    print(page.total, [task.id for task in page.tasks])

asyncio.run(main())

创建模拟任务

对于 Endpoint URL 目标（endpoint_mode != "deployment"），服务在保存或调度任务之前，会使用提供的 api_key 检查 /v1/models。校验失败返回 HTTP 400；直接调用 Python 接口时抛出 SimulationConfigurationError。手动填写模型名称不能绕过此检查。绑定部署的 Model Service 任务仍使用集群执行路径。

创建一个模拟（Simulation）任务，对目标模型端点回放某个跟踪数据集。

REST API
Python API

POST /api/simulation/tasks

请求体：SimulationTaskCreateRequest

name, description, scenario, backend, endpoint_mode
cluster_session_id, endpoint_namespace, endpoint_service, endpoint_deployment_execution_id, endpoint_cluster_id, endpoint_cluster_name, endpoint_deployment_name
model_service_group_id (string，可选) — 用于指定一个已发布、持续进行健康探测的模型服务（Model Service），而非原始的部署执行实例。设置该字段后，后端会将其解析为该分组当前活跃成员及其已发布名称，并强制将 endpoint_mode 设为 "deployment"，覆盖任何传入的 endpoint_deployment_execution_id/model_name。需要同时提供 api_key。
api_key (string，可选) — 模型访问令牌；当设置了 model_service_group_id 时为必填。
endpoint_url, model_name, trace_dataset, trace_path
duration_seconds, scale_factor, trace_start_seconds, trace_end_seconds, trace_timeout_seconds, fixed_schedule, stream, backend_options

{
  "name": "Trace simulation",
  "description": "Replay a chat trace against a ready deployment",
  "scenario": "chat",
  "backend": "trace-replayer",
  "endpoint_mode": "deployment",
  "cluster_session_id": "123e4567-e89b-12d3-a456-426614174000",
  "endpoint_namespace": "prism-sim",
  "endpoint_service": "qwen-router",
  "endpoint_deployment_execution_id": "00000000-0000-4000-8000-000000000001",
  "endpoint_cluster_id": "cluster-1",
  "endpoint_cluster_name": "dev-cluster",
  "endpoint_deployment_name": "qwen3-guide",
  "endpoint_url": "http://127.0.0.1:38000/v1",
  "model_name": "Qwen/Qwen3-8B",
  "trace_dataset": "mooncake-arxiv",
  "trace_path": "/home/prism/.local/share/prism/data/datasets/mooncake_trace.jsonl",
  "duration_seconds": 60,
  "scale_factor": 1,
  "trace_start_seconds": 0,
  "trace_end_seconds": null,
  "trace_timeout_seconds": 3600,
  "fixed_schedule": true,
  "stream": true,
  "backend_options": {}
}

若要改为指定一个已发布的模型服务，请设置 model_service_group_id 和 api_key；后端会丢弃传入的 endpoint_deployment_execution_id/endpoint_url/model_name，并根据该分组的活跃成员重新解析这些值（一个语法上合法的占位 URL 仍可满足请求校验）：

{
  "name": "Trace simulation",
  "scenario": "chat",
  "backend": "trace-replayer",
  "endpoint_mode": "deployment",
  "model_service_group_id": "01998e6b-4b2e-7c7e-8c6d-8e2a3f0c9b11",
  "api_key": "lens-mk-…",
  "endpoint_url": "http://model-service.internal",
  "model_name": "qwen3-8b",
  "trace_dataset": "mooncake-arxiv",
  "trace_path": "/home/prism/.local/share/prism/data/datasets/mooncake_trace.jsonl"
}

响应：

{
  "task_id": "1a2b3c4d",
  "status": "queued",
  "task": {
    "id": "1a2b3c4d",
    "name": "Trace simulation",
    "description": "Replay a chat trace against a ready deployment",
    "scenario": "chat",
    "status": "queued",
    "endpoint_mode": "deployment",
    "endpoint_namespace": "prism-sim",
    "endpoint_service": "qwen-router",
    "endpoint_deployment_execution_id": "00000000-0000-4000-8000-000000000001",
    "endpoint_deployment_run_id": null,
    "endpoint_deployment_case_id": null,
    "endpoint_cluster_id": "cluster-1",
    "endpoint_cluster_name": "dev-cluster",
    "endpoint_deployment_name": "qwen3-guide",
    "endpoint_url": "http://127.0.0.1:38000/v1",
    "model_name": "Qwen/Qwen3-8B",
    "simulation": {
      "backend": "trace-replayer",
      "backend_options": {},
      "duration_seconds": 60.0,
      "num_requests": null,
      "stream": true,
      "warmup_enabled": false,
      "grace_period_seconds": 30.0
    },
    "prompt": {
      "type": "trace",
      "dataset": {"name": "mooncake-arxiv", "scenario": "chat", "tokenizer": null},
      "trace": {
        "path": "/home/prism/.local/share/prism/data/datasets/mooncake_trace.jsonl",
        "format": "mooncake_trace",
        "fixed_schedule": true,
        "timeout_seconds": 3600.0,
        "synthesis_speedup_ratio": 1.0,
        "start_seconds": 0.0,
        "end_seconds": null
      }
    },
    "task_dir": "/home/prism/.local/share/prism/data/artifacts/simulations/1a2b3c4d",
    "artifact_uri": null,
    "artifact_error": null,
    "logs_truncated": false,
    "artifacts_incomplete": false,
    "backend_version_inferred": false,
    "backend_version": null,
    "configuration_ids": ["config-1"],
    "progress_percent": 0.0,
    "progress_message": "Queued",
    "logs": [],
    "result": null,
    "error_message": null,
    "created_at": "2026-09-23T09:18:24.287000Z",
    "started_at": null,
    "execution_started_at": null,
    "completed_at": null,
    "owner_pid": null,
    "owner_instance_id": null
  }
}

import asyncio

from llm_d_lens.simulation.models import SimulationTaskCreateRequest
from llm_d_lens.simulation.service import create_task


async def main() -> None:
    task = await create_task(
        SimulationTaskCreateRequest(
            scenario="chat",
            backend="trace-replayer",
            endpoint_mode="deployment",
            endpoint_deployment_execution_id="00000000-0000-4000-8000-000000000001",
            endpoint_url="http://127.0.0.1:38000/v1",
            model_name="Qwen/Qwen3-8B",
            trace_dataset="mooncake-arxiv",
            trace_path="/home/prism/.local/share/prism/data/datasets/mooncake_trace.jsonl",
        )
    )
    print(task.id, task.status)

    # Or target a published Model Service (resolves to its active member and
    # published name; requires api_key, and endpoint_url/model_name are discarded):
    task = await create_task(
        SimulationTaskCreateRequest(
            scenario="chat",
            backend="trace-replayer",
            endpoint_mode="deployment",
            model_service_group_id="01998e6b-4b2e-7c7e-8c6d-8e2a3f0c9b11",
            api_key="lens-mk-…",
            endpoint_url="http://model-service.internal",
            model_name="qwen3-8b",
            trace_dataset="mooncake-arxiv",
            trace_path="/home/prism/.local/share/prism/data/datasets/mooncake_trace.jsonl",
        )
    )
    print(task.id, task.status)


asyncio.run(main())

获取模拟任务

根据 id 获取某个模拟任务，包括进度、日志以及（如果存在的话）结果产物。

REST API
Python API

GET /api/simulation/tasks/{task_id}

请求：无请求体。

路径参数：task_id（string，必填）

响应：

{
  "id": "1a2b3c4d",
  "name": "Trace simulation",
  "description": "Replay a chat trace against a ready deployment",
  "scenario": "chat",
  "status": "completed",
  "endpoint_mode": "deployment",
  "endpoint_namespace": "prism-sim",
  "endpoint_service": "qwen-router",
  "endpoint_deployment_execution_id": "00000000-0000-4000-8000-000000000001",
  "endpoint_deployment_run_id": null,
  "endpoint_deployment_case_id": null,
  "endpoint_cluster_id": "cluster-1",
  "endpoint_cluster_name": "dev-cluster",
  "endpoint_deployment_name": "qwen3-guide",
  "endpoint_url": "http://127.0.0.1:38000/v1",
  "model_name": "Qwen/Qwen3-8B",
  "simulation": {
    "backend": "trace-replayer",
    "backend_options": {},
    "duration_seconds": 60.0,
    "num_requests": null,
    "stream": true,
    "warmup_enabled": false,
    "grace_period_seconds": 30.0
  },
  "prompt": {
    "type": "trace",
    "dataset": {"name": "mooncake-arxiv", "scenario": "chat", "tokenizer": null},
    "trace": {
      "path": "/home/prism/.local/share/prism/data/datasets/mooncake_trace.jsonl",
      "format": "mooncake_trace",
      "fixed_schedule": true,
      "timeout_seconds": 3600.0,
      "synthesis_speedup_ratio": 1.0,
      "start_seconds": 0.0,
      "end_seconds": null
    }
  },
  "task_dir": "/home/prism/.local/share/prism/data/artifacts/simulations/1a2b3c4d",
  "artifact_uri": "artifact://simulation/1a2b3c4d",
  "artifact_error": null,
  "logs_truncated": false,
  "artifacts_incomplete": false,
  "backend_version_inferred": false,
  "backend_version": "0.2.0",
  "configuration_ids": ["config-1"],
  "progress_percent": 100.0,
  "progress_message": "Simulation completed",
  "logs": ["Starting trace-replayer", "Simulation completed"],
  "result": {
    "run_id": "trace-replayer-1",
    "backend": "trace-replayer",
    "backend_version": "0.2.0",
    "summary": {
      "throughput_tps": 1180.4,
      "success_rate": 99.8,
      "completion_timeline": []
    },
    "artifacts": [
      {
        "uri": "artifact://simulation/1a2b3c4d/summary.json",
        "kind": "summary-json",
        "path": "/home/prism/.local/share/prism/data/artifacts/simulations/1a2b3c4d/summary.json",
        "media_type": "application/json"
      }
    ],
    "per_request": [
      {
        "request_id": 1,
        "status_code": 200,
        "successful": true,
        "error": null,
        "started_at_seconds": 0.0,
        "completed_at_seconds": 0.42,
        "latency_ms": 420.0,
        "ttft_ms": 110.0,
        "tpot_ms": 8.3,
        "input_tokens": 512.0,
        "output_tokens": 128.0,
        "metadata": {}
      }
    ],
    "backend_metrics": {},
    "warnings": []
  },
  "error_message": null,
  "created_at": "2026-09-23T09:18:24.287000Z",
  "started_at": "2026-09-23T09:18:25.000000Z",
  "execution_started_at": "2026-09-23T09:18:27.000000Z",
  "completed_at": "2026-09-23T09:19:32.000000Z",
  "owner_pid": null,
  "owner_instance_id": null
}

import asyncio

from llm_d_lens.simulation.service import get_task

async def main() -> None:
    task = await get_task("1a2b3c4d")
    if task is not None:
        print(task.id, task.status, task.progress_percent)

asyncio.run(main())

下载模拟产物

根据任务 id 和产物类型，以文件附件形式下载已保存的模拟产物。

REST API
Python API

GET /api/simulation/tasks/{task_id}/artifacts/{kind}

请求：无请求体。

路径参数：task_id（string，必填）、kind（string，必填）

响应：文件附件，从已保存的产物路径流式传输。不返回 JSON 响应体。任务的 result.artifacts 列表会告知哪些 kind 值可用。

import asyncio

from llm_d_lens.simulation.router import download_simulation_artifact

async def main() -> None:
    response = await download_simulation_artifact("1a2b3c4d", "summary-json")
    print(response.media_type)

asyncio.run(main())

获取模拟任务的响应状态码问题

获取某个模拟任务中返回特定 HTTP 状态码的请求。

REST API
Python API

GET /api/simulation/tasks/{task_id}/response-code-issues

请求：无请求体。

路径参数：task_id（string，必填）
查询参数：status_code（string，必填；缺失状态码时使用 unknown）、offset（integer，默认 0）、limit（integer，默认 100，最大 500）

响应：

{
  "task_id": "1a2b3c4d",
  "status_code": 500,
  "issues": [
    {
      "request_number": 12,
      "request_id": 12,
      "status_code": 500,
      "error": "Transport error: upstream reset the connection",
      "started_at_seconds": 14.2,
      "latency_ms": 982.5,
      "input_tokens": 512.0,
      "output_tokens": 0.0
    }
  ],
  "total": 1,
  "offset": 0,
  "limit": 100
}

import asyncio

from llm_d_lens.simulation.service import get_response_code_issues

async def main() -> None:
    issues = await get_response_code_issues("1a2b3c4d", 500, offset=0, limit=20)
    print(issues.task_id, issues.total)

asyncio.run(main())

停止模拟任务

停止一个正在运行的模拟任务。

REST API
Python API

POST /api/simulation/tasks/{task_id}/stop

请求：无请求体。

路径参数：task_id（string，必填）

响应：

{
  "task_id": "1a2b3c4d",
  "status": "cancelling"
}

import asyncio

from llm_d_lens.simulation.service import stop_task

async def main() -> None:
    await stop_task("1a2b3c4d")

asyncio.run(main())

重新运行模拟任务

将之前的模拟任务重新排队运行，可选择将其重新绑定到新的端点或部署执行实例。

REST API
Python API

POST /api/simulation/tasks/{task_id}/rerun

请求体：SimulationTaskRerunRequest（可选）

路径参数：task_id（string，必填）
可选的请求体字段：endpoint_url、endpoint_deployment_execution_id、endpoint_cluster_id、endpoint_cluster_name、endpoint_deployment_name
model_service_group_id (string，可选) — 重新绑定到一个已发布、经过健康探测的模型服务，而非原始的部署执行实例；其解析方式与创建任务时相同，当两者同时提供时优先于 endpoint_deployment_execution_id。

{
  "endpoint_url": "http://127.0.0.1:39000/v1",
  "endpoint_deployment_execution_id": "00000000-0000-4000-8000-000000000002",
  "endpoint_cluster_id": "cluster-1",
  "endpoint_cluster_name": "dev-cluster",
  "endpoint_deployment_name": "qwen3-guide-rerun"
}

响应：

{
  "task_id": "5e6f7a8b",
  "status": "queued",
  "task": {
    "id": "5e6f7a8b",
    "name": "Trace simulation · rerun 5e6f7a8b",
    "description": "Replay a chat trace against a ready deployment",
    "scenario": "chat",
    "status": "queued",
    "endpoint_mode": "deployment",
    "endpoint_namespace": "prism-sim",
    "endpoint_service": "qwen-router",
    "endpoint_deployment_execution_id": "00000000-0000-4000-8000-000000000002",
    "endpoint_deployment_run_id": null,
    "endpoint_deployment_case_id": null,
    "endpoint_cluster_id": "cluster-1",
    "endpoint_cluster_name": "dev-cluster",
    "endpoint_deployment_name": "qwen3-guide-rerun",
    "endpoint_url": "http://127.0.0.1:39000/v1",
    "model_name": "Qwen/Qwen3-8B",
    "simulation": {
      "backend": "trace-replayer",
      "backend_options": {},
      "duration_seconds": 60.0,
      "num_requests": null,
      "stream": true,
      "warmup_enabled": false,
      "grace_period_seconds": 30.0
    },
    "prompt": {
      "type": "trace",
      "dataset": {"name": "mooncake-arxiv", "scenario": "chat", "tokenizer": null},
      "trace": {
        "path": "/home/prism/.local/share/prism/data/datasets/mooncake_trace.jsonl",
        "format": "mooncake_trace",
        "fixed_schedule": true,
        "timeout_seconds": 3600.0,
        "synthesis_speedup_ratio": 1.0,
        "start_seconds": 0.0,
        "end_seconds": null
      }
    },
    "task_dir": "/home/prism/.local/share/prism/data/artifacts/simulations/5e6f7a8b",
    "artifact_uri": null,
    "artifact_error": null,
    "logs_truncated": false,
    "artifacts_incomplete": false,
    "backend_version_inferred": false,
    "backend_version": null,
    "configuration_ids": ["config-2"],
    "progress_percent": 0.0,
    "progress_message": "Queued",
    "logs": [],
    "result": null,
    "error_message": null,
    "created_at": "2026-09-23T09:25:24.287000Z",
    "started_at": null,
    "execution_started_at": null,
    "completed_at": null,
    "owner_pid": null,
    "owner_instance_id": null
  }
}

import asyncio

from llm_d_lens.simulation.models import SimulationTaskRerunRequest
from llm_d_lens.simulation.service import rerun_task


async def main() -> None:
    task = await rerun_task(
        "1a2b3c4d",
        override=SimulationTaskRerunRequest(
            endpoint_deployment_execution_id="00000000-0000-4000-8000-000000000002",
            endpoint_url="http://127.0.0.1:39000/v1",
        ),
    )
    print(task.id, task.name)

    # Or re-bind to a published Model Service (preferred over
    # endpoint_deployment_execution_id when both are supplied):
    task = await rerun_task(
        "1a2b3c4d",
        override=SimulationTaskRerunRequest(model_service_group_id="01998e6b-4b2e-7c7e-8c6d-8e2a3f0c9b11"),
    )
    print(task.id, task.name)


asyncio.run(main())

删除模拟任务

删除一条模拟任务记录。

REST API
Python API

DELETE /api/simulation/tasks/{task_id}

请求：无请求体。

路径参数：task_id（string，必填）

响应：

{
  "task_id": "1a2b3c4d",
  "status": "deleted"
}

import asyncio

from llm_d_lens.simulation.service import delete_task

async def main() -> None:
    await delete_task("1a2b3c4d")

asyncio.run(main())

在 FastAPI 的 OpenAPI schema 中可查看完整的请求/响应结构（使用 npm run docs:api 重新生成）。

后端 API​

列出模拟后端​

发现端点模型​

列出模拟场景​

列出跟踪数据集​

下载跟踪数据集​

获取跟踪数据集时间线​

列出模拟任务​

创建模拟任务​

获取模拟任务​

下载模拟产物​

获取模拟任务的响应状态码问题​

停止模拟任务​

重新运行模拟任务​

删除模拟任务​

后端 API

列出模拟后端

发现端点模型

列出模拟场景

列出跟踪数据集

下载跟踪数据集

获取跟踪数据集时间线

列出模拟任务

创建模拟任务

获取模拟任务

下载模拟产物

获取模拟任务的响应状态码问题

停止模拟任务

重新运行模拟任务

删除模拟任务