[SWE-Paddle] Add task PaddlePaddle__Paddle_18687

swe-paddle/tasks/PaddlePaddle__Paddle-18687/README.md

@@ -0,0 +1,44 @@
+# PaddlePaddle__Paddle-18687
+
+This directory converts Paddle PR #18687 into a SWE-Paddle community task candidate.
+
+## Source
+
+| Field | Value |
+| --- | --- |
+| Repo | `PaddlePaddle/Paddle` |
+| PR | [#18687](https://github.com/PaddlePaddle/Paddle/pull/18687) |
+| PR title | `add parameter server launch` |
+| Base commit | `d07ad4c6059db28c5f384a25190385742d9ba718` |
+| Merged at | `2019-07-22T14:11:50Z` |
+| Task type | `feature_enhancement` |
+| Resource | CPU |
+
+## Summary
+
+Add a local parameter-server launcher that starts the requested server and worker processes with consistent roles, endpoints, identifiers, and training-script arguments.
+
+## Why This Is A Good SWE-Paddle Candidate
+
+- The change exercises command parsing, environment construction, role assignment, process orchestration, and failure propagation as one focused feature.
+- Its 151-line Python production change is substantial while remaining isolated to one launcher file.
+- Deterministic CPU tests can validate the process contract without starting training, loading model weights, or requiring external services.
+- A P2P case protects the existing collective launcher's training-argument forwarding behavior.
+
+## Files
+
+- `proposal.md`: candidate proposal for maintainer triage.
+- `instruction.md`: self-contained problem statement for the coding agent.
+- `solution/code.patch`: gold patch from the merged PR.
+- `tests/test.patch`: behavior tests exposing the target feature.
+- `tests/test.sh`: minimal target test command.
+- `environment/README.md`: environment notes for reproduction.
+- `README.md`: task overview and verification entrypoint.
+
+## Verification
+
+```bash
+bash tests/test.sh
+```
+
+Expected behavior: applying `tests/test.patch` to `base_commit` should keep the existing launcher regression case passing while the parameter-server cases fail; applying both `tests/test.patch` and `solution/code.patch` should pass all target tests.

swe-paddle/tasks/PaddlePaddle__Paddle-18687/environment/README.md

@@ -0,0 +1,27 @@
+# Environment Notes
+
+This candidate is part of the SWE-Paddle community task set.
+
+## Expected Environment
+
+- Repository: `PaddlePaddle/Paddle`
+- Base commit: `d07ad4c6059db28c5f384a25190385742d9ba718`
+- Resource: CPU
+- GPU required: no
+- Build path: no source build is required; the tests execute checkout Python control flow with controlled subprocess doubles.
+
+## Run Order
+
+1. Check out `PaddlePaddle/Paddle` at the base commit.
+2. Apply `tests/test.patch`.
+3. Run `bash tests/test.sh`; the target behavior should fail before the fix.
+4. Apply `solution/code.patch`.
+5. Run `bash tests/test.sh` again; the target behavior should pass after the gold patch.
+
+## Minimal Test Command
+
+```bash
+bash tests/test.sh
+```
+
+The verifier is responsible for deriving stable F2P and P2P node IDs from repeated runs.

swe-paddle/tasks/PaddlePaddle__Paddle-18687/instruction.md

@@ -0,0 +1,19 @@
+# 为参数服务器分布式训练增加统一的本地启动入口
+
+## 详细描述
+
+在本地调试参数服务器（Parameter Server）分布式训练时，需要同时启动 server 和 worker 进程，并为各进程配置角色、编号和通信信息。随着进程数量增加，手动维护这些配置容易出现不一致，也不便于统一传递训练脚本参数和处理子进程异常。
+
+需要提供一个与现有分布式启动方式一致的命令入口。开发者只需指定 server 数量、worker 数量、起始端口和训练脚本，启动器即可一次拉起完整任务，并将训练脚本后的参数传递给各个子进程。
+
+## 验收说明
+
+- 按指定数量启动 server 和 worker，并为同一角色分配不重复的编号；所有子进程应拿到一致的 server 地址和 worker 总数。
+- 训练脚本参数应原样传递，启动器应等待所有子进程结束，并在子进程以非零状态退出时报告失败。
+- 子进程环境不应继承会干扰本地通信的代理设置，现有 collective launch 的参数转发行为保持不变。
+
+## 技术要求
+
+- 熟悉 Python 子进程和环境变量处理。
+- 熟悉 PaddlePaddle 分布式 launch 流程。
+- 了解参数服务器训练中的 server、worker 与 endpoint 配置。

swe-paddle/tasks/PaddlePaddle__Paddle-18687/proposal.md

@@ -0,0 +1,55 @@
+# Task Proposal: PaddlePaddle__Paddle-18687
+
+## 1. 来源信息
+
+- Instance ID：`PaddlePaddle__Paddle-18687`
+- PR 链接：https://github.com/PaddlePaddle/Paddle/pull/18687
+- PR 标题：`add parameter server launch`
+- `base_commit`：`d07ad4c6059db28c5f384a25190385742d9ba718`
+- merged 时间：`2019-07-22T14:11:50Z`
+- 你的身份：熟悉该模块的 contributor
+- 后续联系人：TBD
+
+## 2. 问题一句话
+
+为本地参数服务器训练提供统一启动入口，根据 server/worker 数量生成一致的进程环境并管理完整任务生命周期。
+
+## 3. 为什么适合作为 SWE-Paddle 样本
+
+- **真实性**：来源 PR 解决本地启动参数服务器任务时需要手工组织多个进程的问题。
+- **代表性**：覆盖命令行解析、角色环境生成、endpoint 编排、子进程启动和失败传播。
+- **边界清楚**：production change 仅新增一个 Python launcher，不涉及算子、模型权重或 GPU 执行。
+- **非平凡性**：需要同时保证两类角色的数量、编号、共享配置、参数转发与进程退出语义。
+- **环境友好性**：测试通过 controlled `Popen` doubles 验证真实控制流，可在 CPU 环境稳定运行。
+
+## 4. 任务类型和标签
+
+- 任务类型：`feature_enhancement`
+- 执行后端：`cpu`
+- 设备范围：`cpu_only`
+- 模块标签：`[distributed, parameter_server, launcher, multiprocessing]`
+
+## 5. 验证思路
+
+- 目标测试命令：`bash tests/test.sh`
+- 目标测试文件：`test/swe_paddle/test_pr18687_launch_ps.py`
+- 修复前预期：现有 collective launcher 的参数转发测试通过；参数服务器 launcher 不存在，因此参数解析、角色编排和失败传播用例失败。
+- 修复后预期：参数服务器命令行、server/worker 环境、命令参数、等待行为及非零退出码传播全部通过。
+- P2P 候选：现有 `paddle.distributed.launch` 仍能把训练脚本及其余参数完整解析出来。
+
+## 6. 环境与资源
+
+- 资源需求：CPU
+- Paddle 来源：`PaddlePaddle/Paddle` source checkout at `base_commit`
+- 是否能提供 Docker：暂无
+- patch 类型：Python-only
+- 环境建议：无需 source build；使用 Python、pytest 和 controlled subprocess doubles 即可运行。
+- 最小测试命令：`bash tests/test.sh`
+- 是否有 oracle 日志：由 SWE-Paddle verifier 结果另行维护
+
+## 7. 风险自查
+
+- 泄露风险：instruction 只描述启动器的外部行为，不给出 Gold patch 的具体代码结构。
+- 环境风险：不启动真实训练进程，不依赖 Paddle wheel、网络、外部服务或 GPU。
+- flaky 风险：进程行为由确定性的 doubles 记录，不依赖端口竞争和真实多进程时序。
+- 拆分风险：来源 PR 只有一个 production file，所有验证围绕同一个参数服务器启动入口。

swe-paddle/tasks/PaddlePaddle__Paddle-18687/solution/code.patch

@@ -0,0 +1,157 @@
+diff --git a/python/paddle/distributed/launch_ps.py b/python/paddle/distributed/launch_ps.py
+new file mode 100644
+index 0000000000000000000000000000000000000000..ded2e49c3e63638710d74322afef1ed12ff53c6c
+--- /dev/null
++++ b/python/paddle/distributed/launch_ps.py
+@@ -0,0 +1,151 @@
++# Copyright (c) 2019 PaddlePaddle Authors. All Rights Reserved.
++#
++# Licensed under the Apache License, Version 2.0 (the "License");
++# you may not use this file except in compliance with the License.
++# You may obtain a copy of the License at
++#
++#     http://www.apache.org/licenses/LICENSE-2.0
++#
++# Unless required by applicable law or agreed to in writing, software
++# distributed under the License is distributed on an "AS IS" BASIS,
++# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
++# See the License for the specific language governing permissions and
++# limitations under the License.
++
++from __future__ import print_function
++from __future__ import unicode_literals
++import subprocess
++import sys
++import os
++import copy
++from argparse import ArgumentParser, REMAINDER
++
++
++def parse_args():
++    # Optional arguments for the launch helper
++    parser = ArgumentParser(description="Distributed training")
++    parser.add_argument(
++        "--cluster_node_ips",
++        type=str,
++        default="127.0.0.1",
++        help="Paddle cluster nodes ips, such as 192.168.0.16,192.168.0.17..")
++
++    parser.add_argument(
++        "--node_ip",
++        type=str,
++        default="127.0.0.1",
++        help="The current node ip. ")
++
++    parser.add_argument(
++        "--start_port",
++        type=int,
++        default=6170,
++        help="The trainer's start port on a single node")
++
++    parser.add_argument(
++        "--print_config",
++        type=bool,
++        default=True,
++        help="Print the config or not")
++
++    parser.add_argument(
++        "--worker_num", type=int, default=2, help="number of workers")
++
++    parser.add_argument(
++        "--server_num", type=int, default=2, help="number of servers")
++
++    parser.add_argument(
++        "--log_dir",
++        default="logs",
++        type=str,
++        help="The path for each process's log.If it's not setted, the log will printed to default pipe."
++    )
++
++    # positional
++    parser.add_argument(
++        "training_script",
++        type=str,
++        help="The full path to the single GPU training "
++        "program/script to be launched in parallel, "
++        "followed by all the arguments for the "
++        "training script")
++
++    # rest from the training program
++    parser.add_argument('training_script_args', nargs=REMAINDER)
++    return parser.parse_args()
++
++
++def start_procs(args):
++    worker_num = args.worker_num
++    server_num = args.server_num
++    start_port = args.start_port
++    default_env = os.environ.copy()
++    current_env = copy.copy(default_env)
++    current_env.pop("http_proxy", None)
++    current_env.pop("https_proxy", None)
++    procs = []
++    cmds = []
++    log_fns = []
++    ports = range(start_port, start_port + server_num, 1)
++    endpoints = ",".join(["127.0.0.1:" + str(x) for x in ports])
++    for i in range(server_num):
++        current_env.update({
++            "TRAINER_NUM": str(worker_num),
++            "CURRENT_ID": str(i),
++            "ENDPOINTS": endpoints,
++            "TRAINING_ROLE": "PSERVER"
++        })
++        cmd = [sys.executable, "-u", args.training_script
++               ] + args.training_script_args
++        cmds.append(cmd)
++        print(cmd)
++        if args.log_dir is not None:
++            os.system("mkdir -p {}".format(args.log_dir))
++            fn = open("%s/serverlog.%d" % (args.log_dir, i), "w")
++            log_fns.append(fn)
++            proc = subprocess.Popen(cmd, env=current_env, stdout=fn, stderr=fn)
++        else:
++            proc = subprocess.Popen(cmd, env=current_env)
++        procs.append(proc)
++
++    for i in range(worker_num):
++        current_env.update({
++            "ENDPOINTS": endpoints,
++            "TRAINER_NUM": str(worker_num),
++            "TRAINING_ROLE": "TRAINER",
++            "CURRENT_ID": str(i)
++        })
++        cmd = [sys.executable, "-u", args.training_script
++               ] + args.training_script_args
++        print(cmd)
++        cmds.append(cmd)
++        if args.log_dir is not None:
++            os.system("mkdir -p {}".format(args.log_dir))
++            fn = open("%s/workerlog.%d" % (args.log_dir, i), "w")
++            log_fns.append(fn)
++            proc = subprocess.Popen(cmd, env=current_env, stdout=fn, stderr=fn)
++        else:
++            proc = subprocess.Popen(cmd, env=current_env)
++        procs.append(proc)
++
++    for i in range(0, len(procs)):
++        proc = procs[i]
++
++        proc.wait()
++        if len(log_fns) > 0:
++            log_fns[i].close()
++
++        if proc.returncode != 0:
++            raise subprocess.CalledProcessError(
++                returncode=procs[i].returncode, cmd=cmds[i])
++
++
++def launch():
++    args = parse_args()
++    if args.print_config:
++        start_procs(args)
++
++
++# server num, worker num        
++if __name__ == "__main__":
++    launch()