jacksoncode.github.io/tutorials/python/66.html at master · jacksoncode/jacksoncode.github.io

Repository navigation

History

100

101

102

103

104

105

106

107

108

109

110

111

112

113

114

115

116

117

118

119

120

121

122

123

124

125

126

127

128

129

130

131

132

133

134

135

136

137

138

139

140

141

142

143

144

145

146

147

148

149

150

151

152

153

154

155

156

157

158

159

160

161

162

163

164

165

166

167

168

169

170

171

172

173

174

175

176

177

178

179

180

181

182

183

184

185

186

187

188

189

190

191

192

193

194

195

196

197

198

199

200

201

202

203

204

205

206

207

208

209

210

211

212

213

214

215

216

217

218

219

220

221

222

223

224

225

226

227

228

229

230

231

232

233

234

235

236

237

238

239

240

241

242

243

244

245

246

247

248

249

250

251

252

253

254

255

256

257

258

259

260

261

262

263

264

265

266

267

268

269

270

271

272

273

<!DOCTYPE html>

<head>

<title>14.5. 分布式进程 - CodeClub</title>

<style>

.tutorial-container { max-width: 960px; margin: 70px auto 40px; padding: 0 20px; }

.tutorial-sidebar { position: fixed; top: 70px; left: 20px; width: 260px; max-height: calc(100vh - 90px); overflow-y: auto; background: var(--bg-secondary, #fff); border-radius: 8px; padding: 16px; box-shadow: 0 2px 8px rgba(0,0,0,0.08); }

.tutorial-sidebar h3 { font-size: 16px; margin-bottom: 12px; color: var(--text-primary, #1a1a2e); }

.tutorial-sidebar a { display: block; padding: 6px 10px; color: var(--text-secondary, #666); text-decoration: none; font-size: 14px; border-radius: 4px; }

.tutorial-sidebar a:hover, .tutorial-sidebar a.active { background: var(--primary, #3b82f6); color: #fff; }

.tutorial-sidebar .sub-link { padding-left: 24px; font-size: 13px; }

.tutorial-content { margin-left: 300px; background: var(--bg-secondary, #fff); border-radius: 8px; padding: 32px; box-shadow: 0 2px 8px rgba(0,0,0,0.08); }

.tutorial-content h1 { font-size: 28px; margin-bottom: 20px; color: var(--text-primary, #1a1a2e); border-bottom: 2px solid var(--primary, #3b82f6); padding-bottom: 12px; }

.tutorial-content h2 { font-size: 22px; margin: 24px 0 16px; color: var(--text-primary, #1a1a2e); }

.tutorial-content h3 { font-size: 18px; margin: 20px 0 12px; }

.tutorial-content p { line-height: 1.8; margin: 12px 0; color: var(--text-secondary, #333); }

.tutorial-content pre { background: #1e1e2e; color: #cdd6f4; padding: 16px; border-radius: 8px; overflow-x: auto; margin: 16px 0; }

.tutorial-content code { font-family: 'Fira Code', 'Cascadia Code', monospace; font-size: 14px; }

.tutorial-content p code { background: #f1f5f9; padding: 2px 6px; border-radius: 4px; color: #e83e8c; }

.tutorial-content blockquote { border-left: 4px solid var(--primary, #3b82f6); padding: 12px 20px; margin: 16px 0; background: #f8fafc; border-radius: 0 8px 8px 0; }

.tutorial-content ul, .tutorial-content ol { padding-left: 24px; margin: 12px 0; }

.tutorial-content li { margin: 6px 0; line-height: 1.7; }

.tutorial-content img { max-width: 100%; border-radius: 8px; margin: 16px 0; }

.tutorial-nav { display: flex; justify-content: space-between; margin-top: 32px; padding-top: 20px; border-top: 1px solid var(--border, #e5e7eb); }

.tutorial-nav a { color: var(--primary, #3b82f6); text-decoration: none; font-weight: 500; }

.tutorial-nav a:hover { text-decoration: underline; }

.tutorial-index { max-width: 960px; margin: 70px auto 40px; padding: 0 20px; }

.tutorial-card { background: var(--bg-secondary, #fff); border-radius: 12px; padding: 24px; margin: 16px 0; box-shadow: 0 2px 8px rgba(0,0,0,0.08); transition: transform 0.2s; }

.tutorial-card:hover { transform: translateY(-2px); }

.tutorial-card h3 { margin: 0 0 8px; color: var(--text-primary, #1a1a2e); }

.tutorial-card p { color: var(--text-secondary, #666); margin: 0 0 12px; }

.tutorial-card .badge { display: inline-block; padding: 4px 10px; border-radius: 20px; font-size: 12px; font-weight: 600; margin-right: 8px; }

.badge-python { background: #3b82f6; color: #fff; }

.badge-git { background: #f97316; color: #fff; }

.badge-ai { background: #8b5cf6; color: #fff; }

@media (max-width: 768px) {

.tutorial-sidebar { display: none; }

.tutorial-content { margin-left: 0; }

}

</style>

</head>

<body>

<span>CodeClub</span>

</a>

</button>

</ul>

</div>

</nav>

<div class="tutorial-container" style="max-width: 960px; margin: 80px auto 40px; padding: 0 20px;"><div class="tutorial-sidebar"><h3>目录</h3><a href="0.html">1. 简介</a><a href="1.html">2. Python历史</a><a href="2.html">3. 安装Python</a><a href="3.html">3.1. Python解释器</a><a href="4.html">4. 第一个Python程序</a><a href="5.html">4.1. 使用文本编辑器</a><a href="6.html">4.2. 输入和输出</a><a href="7.html">5. Python基础</a><a href="8.html">5.1. 数据类型和变量</a><a href="9.html">5.2. 字符串和编码</a><a href="10.html">5.3. 使用list和tuple</a><a href="11.html">5.4. 条件判断</a><a href="12.html">5.5. 模式匹配</a><a href="13.html">5.6. 循环</a><a href="14.html">5.7. 使用dict和set</a><a href="15.html">6. 函数</a><a href="16.html">6.1. 调用函数</a><a href="17.html">6.2. 定义函数</a><a href="18.html">6.3. 函数的参数</a><a href="19.html">6.4. 递归函数</a><a href="20.html">7. 高级特性</a><a href="21.html">7.1. 切片</a><a href="22.html">7.2. 迭代</a><a href="23.html">7.3. 列表生成式</a><a href="24.html">7.4. 生成器</a><a href="25.html">7.5. 迭代器</a><a href="26.html">8. 函数式编程</a><a href="27.html">8.1. 高阶函数</a><a href="28.html">8.1.1. map/reduce</a><a href="29.html">8.1.2. filter</a><a href="30.html">8.1.3. sorted</a><a href="31.html">8.2. 返回函数</a><a href="32.html">8.3. 匿名函数</a><a href="33.html">8.4. 装饰器</a><a href="34.html">8.5. 偏函数</a><a href="35.html">9. 模块</a><a href="36.html">9.1. 使用模块</a><a href="37.html">9.2. 安装第三方模块</a><a href="38.html">10. 面向对象编程</a><a href="39.html">10.1. 类和实例</a><a href="40.html">10.2. 访问限制</a><a href="41.html">10.3. 继承和多态</a><a href="42.html">10.4. 获取对象信息</a><a href="43.html">10.5. 实例属性和类属性</a><a href="44.html">11. 面向对象高级编程</a><a href="45.html">11.1. 使用__slots__</a><a href="46.html">11.2. 使用@property</a><a href="47.html">11.3. 多重继承</a><a href="48.html">11.4. 定制类</a><a href="49.html">11.5. 使用枚举类</a><a href="50.html">11.6. 使用元类</a><a href="51.html">12. 错误、调试和测试</a><a href="52.html">12.1. 错误处理</a><a href="53.html">12.2. 调试</a><a href="54.html">12.3. 单元测试</a><a href="55.html">12.4. 文档测试</a><a href="56.html">13. IO编程</a><a href="57.html">13.1. 文件读写</a><a href="58.html">13.2. StringIO和BytesIO</a><a href="59.html">13.3. 操作文件和目录</a><a href="60.html">13.4. 序列化</a><a href="61.html">14. 进程和线程</a><a href="62.html">14.1. 多进程</a><a href="63.html">14.2. 多线程</a><a href="64.html">14.3. ThreadLocal</a><a href="65.html">14.4. 进程 vs. 线程</a><a href="66.html">14.5. 分布式进程</a><a href="67.html">15. 正则表达式</a><a href="68.html">16. 常用内建模块</a><a href="69.html">16.1. datetime</a><a href="70.html">16.2. collections</a><a href="71.html">16.3. argparse</a><a href="72.html">16.4. base64</a><a href="73.html">16.5. struct</a><a href="74.html">16.6. hashlib</a><a href="75.html">16.7. hmac</a><a href="76.html">16.8. itertools</a><a href="77.html">16.9. contextlib</a><a href="78.html">16.10. urllib</a><a href="79.html">16.11. XML</a><a href="80.html">16.12. HTMLParser</a><a href="81.html">16.13. venv</a><a href="82.html">17. 常用第三方模块</a><a href="83.html">17.1. Pillow</a><a href="84.html">17.2. requests</a><a href="85.html">17.3. chardet</a><a href="86.html">17.4. psutil</a><a href="87.html">18. 图形界面</a><a href="88.html">18.1. 海龟绘图</a><a href="89.html">19. 网络编程</a><a href="90.html">19.1. TCP/IP简介</a><a href="91.html">19.2. TCP编程</a><a href="92.html">19.3. UDP编程</a><a href="93.html">20. 电子邮件</a><a href="94.html">20.1. SMTP发送邮件</a><a href="95.html">20.2. POP3收取邮件</a><a href="96.html">21. 访问数据库</a><a href="97.html">21.1. 使用SQLite</a><a href="98.html">21.2. 使用MySQL</a><a href="99.html">21.3. 使用SQLAlchemy</a><a href="100.html">22. Web开发</a><a href="101.html">22.1. HTTP协议简介</a><a href="102.html">22.2. HTML简介</a><a href="103.html">22.3. WSGI接口</a><a href="104.html">22.4. 使用Web框架</a><a href="105.html">22.5. 使用模板</a><a href="106.html">23. 异步IO</a><a href="107.html">23.1. 协程</a><a href="108.html">23.2. 使用asyncio</a><a href="109.html">23.3. 使用aiohttp</a><a href="110.html">24. FAQ</a><a href="111.html">25. 期末总结</a></div><div class="tutorial-content"><h1>14.5. 分布式进程</h1><div class="tutorial-body"><p>Python的<code>multiprocessing</code>模块不但支持多进程，其中<code>managers</code>子模块还支持把多进程分布到多台机器上。一个服务进程可以作为调度者，将任务分布到其他多个进程中，依靠网络通信。由于<code>managers</code>模块封装很好，不必了解网络通信的细节，就可以很容易地编写分布式多进程程序。</p>

<p>举个例子：如果我们已经有一个通过<code>Queue</code>通信的多进程程序在同一台机器上运行，现在，由于处理任务的进程任务繁重，希望把发送任务的进程和处理任务的进程分布到两台机器上。怎么用分布式进程实现？</p>

<p>原有的<code>Queue</code>可以继续使用，但是，通过<code>managers</code>模块把<code>Queue</code>通过网络暴露出去，就可以让其他机器的进程访问<code>Queue</code>了。</p>

<p>我们先看服务进程，服务进程负责启动<code>Queue</code>，把<code>Queue</code>注册到网络上，然后往<code>Queue</code>里面写入任务：</p>

<pre class="hljs"><code class="language-python"><span class="hljs-comment"># task_master.py</span>

<span class="hljs-keyword">import</span> random, time, queue

<span class="hljs-keyword">from</span> multiprocessing.managers <span class="hljs-keyword">import</span> BaseManager

<span class="hljs-comment"># 发送任务的队列:</span>

task_queue = queue.Queue()

<span class="hljs-comment"># 接收结果的队列:</span>

result_queue = queue.Queue()

<span class="hljs-comment"># 从BaseManager继承的QueueManager:</span>

<span class="hljs-keyword">class</span> <span class="hljs-title class_">QueueManager</span>(<span class="hljs-title class_ inherited__">BaseManager</span>):

<span class="hljs-comment"># 把两个Queue都注册到网络上, callable参数关联了Queue对象:</span>

QueueManager.register(<span class="hljs-string">'get_task_queue'</span>, <span class="hljs-built_in">callable</span>=<span class="hljs-keyword">lambda</span>: task_queue)

QueueManager.register(<span class="hljs-string">'get_result_queue'</span>, <span class="hljs-built_in">callable</span>=<span class="hljs-keyword">lambda</span>: result_queue)

<span class="hljs-comment"># 绑定端口5000, 设置验证码'abc':</span>

manager = QueueManager(address=(<span class="hljs-string">''</span>, <span class="hljs-number">5000</span>), authkey=<span class="hljs-string">b'abc'</span>)

<span class="hljs-comment"># 启动Queue:</span>

manager.start()

<span class="hljs-comment"># 获得通过网络访问的Queue对象:</span>

task = manager.get_task_queue()

result = manager.get_result_queue()

<span class="hljs-comment"># 放几个任务进去:</span>

<span class="hljs-keyword">for</span> i <span class="hljs-keyword">in</span> <span class="hljs-built_in">range</span>(<span class="hljs-number">10</span>):

n = random.randint(<span class="hljs-number">0</span>, <span class="hljs-number">10000</span>)

<span class="hljs-built_in">print</span>(<span class="hljs-string">'Put task %d...'</span> % n)

task.put(n)

<span class="hljs-comment"># 从result队列读取结果:</span>

<span class="hljs-built_in">print</span>(<span class="hljs-string">'Try get results...'</span>)

<span class="hljs-keyword">for</span> i <span class="hljs-keyword">in</span> <span class="hljs-built_in">range</span>(<span class="hljs-number">10</span>):

r = result.get(timeout=<span class="hljs-number">10</span>)

<span class="hljs-built_in">print</span>(<span class="hljs-string">'Result: %s'</span> % r)

manager.shutdown()

<span class="hljs-built_in">print</span>(<span class="hljs-string">'master exit.'</span>)

</code></pre>

<p>请注意，当我们在一台机器上写多进程程序时，创建的<code>Queue</code>可以直接拿来用，但是，在分布式多进程环境下，添加任务到<code>Queue</code>不可以直接对原始的<code>task_queue</code>进行操作，那样就绕过了<code>QueueManager</code>的封装，必须通过<code>manager.get_task_queue()</code>获得的<code>Queue</code>接口添加。</p>

<p>然后，在另一台机器上启动任务进程（本机上启动也可以）：</p>

<pre class="hljs"><code class="language-python"><span class="hljs-comment"># task_worker.py</span>

<span class="hljs-keyword">import</span> time, sys, queue

<span class="hljs-keyword">from</span> multiprocessing.managers <span class="hljs-keyword">import</span> BaseManager

<span class="hljs-comment"># 创建类似的QueueManager:</span>

<span class="hljs-keyword">class</span> <span class="hljs-title class_">QueueManager</span>(<span class="hljs-title class_ inherited__">BaseManager</span>):

<span class="hljs-comment"># 由于这个QueueManager只从网络上获取Queue，所以注册时只提供名字:</span>

QueueManager.register(<span class="hljs-string">'get_task_queue'</span>)

QueueManager.register(<span class="hljs-string">'get_result_queue'</span>)

<span class="hljs-comment"># 连接到服务器，也就是运行task_master.py的机器:</span>

server_addr = <span class="hljs-string">'127.0.0.1'</span>

<span class="hljs-built_in">print</span>(<span class="hljs-string">'Connect to server %s...'</span> % server_addr)

<span class="hljs-comment"># 端口和验证码注意保持与task_master.py设置的完全一致:</span>

m = QueueManager(address=(server_addr, <span class="hljs-number">5000</span>), authkey=<span class="hljs-string">b'abc'</span>)

<span class="hljs-comment"># 从网络连接:</span>

m.connect()

<span class="hljs-comment"># 获取Queue的对象:</span>

task = m.get_task_queue()

result = m.get_result_queue()

<span class="hljs-comment"># 从task队列取任务,并把结果写入result队列:</span>

<span class="hljs-keyword">for</span> i <span class="hljs-keyword">in</span> <span class="hljs-built_in">range</span>(<span class="hljs-number">10</span>):

<span class="hljs-keyword">try</span>:

n = task.get(timeout=<span class="hljs-number">1</span>)

<span class="hljs-built_in">print</span>(<span class="hljs-string">'run task %d * %d...'</span> % (n, n))

r = <span class="hljs-string">'%d * %d = %d'</span> % (n, n, n*n)

time.sleep(<span class="hljs-number">1</span>)

result.put(r)

<span class="hljs-keyword">except</span> Queue.Empty:

<span class="hljs-built_in">print</span>(<span class="hljs-string">'task queue is empty.'</span>)

<span class="hljs-built_in">print</span>(<span class="hljs-string">'worker exit.'</span>)

</code></pre>

<p>任务进程要通过网络连接到服务进程，所以要指定服务进程的IP。</p>

<p>现在，可以试试分布式进程的工作效果了。先启动<code>task_master.py</code>服务进程：</p>

<pre class="hljs"><code class="language-plain">$ python3 task_master.py

Put task 3411...

Put task 1605...

Put task 1398...

Put task 4729...

Put task 5300...

Put task 7471...

Put task 68...

Put task 4219...

Put task 339...

Put task 7866...

Try get results...

</code></pre>

<p><code>task_master.py</code>进程发送完任务后，开始等待<code>result</code>队列的结果。现在启动<code>task_worker.py</code>进程：</p>

<pre class="hljs"><code class="language-plain">$ python3 task_worker.py

Connect to server 127.0.0.1...

run task 3411 * 3411...

run task 1605 * 1605...

run task 1398 * 1398...

run task 4729 * 4729...

run task 5300 * 5300...

run task 7471 * 7471...

run task 68 * 68...

run task 4219 * 4219...

run task 339 * 339...

run task 7866 * 7866...

worker exit.

</code></pre>

<p><code>task_worker.py</code>进程结束，在<code>task_master.py</code>进程中会继续打印出结果：</p>

<pre class="hljs"><code class="language-plain">Result: 3411 * 3411 = 11634921

Result: 1605 * 1605 = 2576025

Result: 1398 * 1398 = 1954404

Result: 4729 * 4729 = 22363441

Result: 5300 * 5300 = 28090000

Result: 7471 * 7471 = 55815841

Result: 68 * 68 = 4624

Result: 4219 * 4219 = 17799961

Result: 339 * 339 = 114921

Result: 7866 * 7866 = 61873956

</code></pre>

<p>这个简单的Master/Worker模型有什么用？其实这就是一个简单但真正的分布式计算，把代码稍加改造，启动多个worker，就可以把任务分布到几台甚至几十台机器上，比如把计算<code>n*n</code>的代码换成发送邮件，就实现了邮件队列的异步发送。</p>

<p>Queue对象存储在哪？注意到<code>task_worker.py</code>中根本没有创建Queue的代码，所以，Queue对象存储在<code>task_master.py</code>进程中：</p>

<pre class="ascii"><code> │

┌─────────────────────────────────────────┐ ┌──────────────────────────────────────┐

│task_master.py │ │ │task_worker.py │

│ │ │ │

│ task = manager.get_task_queue() │ │ │ task = manager.get_task_queue() │

│ result = manager.get_result_queue() │ │ result = manager.get_result_queue() │

│ │ │ │ │ │ │

│ │ │ │ │ │

│ ▼ │ │ │ │ │

│ ┌─────────────────────────────────┐ │ │ │ │

│ │QueueManager │ │ │ │ │ │

│ │ ┌────────────┐ ┌──────────────┐ │ │ │ │ │

│ │ │ task_queue │ │ result_queue │ │◀───┼──┼──┼──────────────┘ │

│ │ └────────────┘ └──────────────┘ │ │ │ │

│ └─────────────────────────────────┘ │ │ │ │

└─────────────────────────────────────────┘ └──────────────────────────────────────┘

│

Network

</code></pre>

<p>而<code>Queue</code>之所以能通过网络访问，就是通过<code>QueueManager</code>实现的。由于<code>QueueManager</code>管理的不止一个<code>Queue</code>，所以，要给每个<code>Queue</code>的网络调用接口起个名字，比如<code>get_task_queue</code>。</p>

<p><code>authkey</code>有什么用？这是为了保证两台机器正常通信，不被其他机器恶意干扰。如果<code>task_worker.py</code>的<code>authkey</code>和<code>task_master.py</code>的<code>authkey</code>不一致，肯定连接不上。</p>

<p>Python的分布式进程接口简单，封装良好，适合需要把繁重任务分布到多台机器的环境下。</p>

<p>注意Queue的作用是用来传递任务和接收结果，每个任务的描述数据量要尽量小。比如发送一个处理日志文件的任务，就不要发送几百兆的日志文件本身，而是发送日志文件存放的完整路径，由Worker进程再去共享的磁盘上读取文件。</p>

<p><a href="task_master.py">task_master.py</a></p>

<p><a href="task_worker.py">task_worker.py</a></p>

</div>

<a href="/books/python/process-thread/process-vs-thread/index.html"

onclick="gs_load_chapter('/books/python/process-thread/process-vs-thread/index.html',true);return false;"

class="inline-block my-2">

stroke="currentColor">

<path stroke-linecap="round" stroke-linejoin="round"

d="M18.75 19.5l-7.5-7.5 7.5-7.5m-6 15L5.25 12l7.5-7.5" />

</svg>

进程 vs. 线程

</a>

</div>

<a href="/books/python/reg-exp/index.html"

onclick="gs_load_chapter('/books/python/reg-exp/index.html',true);return false;"

class="inline-block my-2">

正则表达式

stroke="currentColor">

<path stroke-linecap="round" stroke-linejoin="round"

d="M11.25 4.5l7.5 7.5-7.5 7.5m-6-15l7.5 7.5-7.5 7.5" />

</svg>

</a></div><div style="display: flex; justify-content: space-between; margin-top: 32px; padding-top: 20px; border-top: 1px solid var(--border, #e5e7eb);"><a href="65.html" style="color: var(--primary, #3b82f6); text-decoration: none; font-weight: 500;">← 14.4. 进程 vs. 线程</a><a href="67.html" style="color: var(--primary, #3b82f6); text-decoration: none; font-weight: 500;">15. 正则表达式 →</a></div><p style="margin-top: 16px; font-size: 13px; color: var(--text-secondary, #999);">原文：<a href="https://liaoxuefeng.com/books/python/process-thread/process-manager/index.html" style="color: var(--primary, #3b82f6);">廖雪峰 - 14.5. 分布式进程</a></p></div></div>

</div>

</footer>

</body>

</html>

FilesExpand file tree

66.html

Latest commit

History

66.html

File metadata and controls