来源：诸葛子房的博客

前文介绍jdbc interpreter和interpreter模块交互代码，本篇文章主要分析Zengine调用Interpreter模块代码。

介绍完这篇文章之后，我们即可将paragraph run的流程串起来(后面会将整个流程进行串讲)

同样，来看下这个测试类

zeppelin-zengine/src/test/java/org/apache/zeppelin/interpreter/remote/RemoteInterpreterTest.java

@Testpublic void testFIFOScheduler() throws InterruptedException, InterpreterException {    LOGGER.info("===testFIFOScheduler====");    interpreterSetting.getOption().setPerUser(InterpreterOption.SHARED);// by default SleepInterpreter would use FIFOScheduler    LOGGER.info("===getInterpreter====");final Interpreter interpreter1 = interpreterSetting.getInterpreter("user1", note1Id, "sleep");    LOGGER.info("===createDummyInterpreterContext====");final InterpreterContext context1 = createDummyInterpreterContext();// run this dummy interpret method first to launch the RemoteInterpreterProcess to avoid the// time overhead of launching the process.    LOGGER.info("111");    LOGGER.info("=====name:{}=======",interpreter1.getClassName());    System.out.println(interpreter1.getClassName());    interpreter1.interpret("10101", context1);    LOGGER.info("222");    Thread thread1 = new Thread() {@Overridepublic void run() {try {          assertEquals(Code.SUCCESS, interpreter1.interpret("100", context1).code());        } catch (InterpreterException e) {          e.printStackTrace();          fail();        }      }    };    Thread thread2 = new Thread() {@Overridepublic void run() {try {          assertEquals(Code.SUCCESS, interpreter1.interpret("100", context1).code());        } catch (InterpreterException e) {          e.printStackTrace();          fail();        }      }    };long start = System.currentTimeMillis();    thread1.start();    thread2.start();    thread1.join();    thread2.join();long end = System.currentTimeMillis();    assertTrue((end - start) >= 200);  }

可以看下这个测试方法，这边加了一些日志

RemoteInterpreterTest 继承 AbstractInterpreterTest 里面的抽象类，会先执行setUp方法对读取配置文件信息interpreter 进行初始化

核心主要是执行RemoteInterpreter里面的 interpret方法

  @Override  public InterpreterResult interpret(final String st, final InterpreterContext context)      throws InterpreterException {    LOGGER.info("st:\n{}", st);if (LOGGER.isDebugEnabled()) {      LOGGER.debug("st:\n{}", st);    }
    final FormType form = getFormType();    RemoteInterpreterProcess interpreterProcess = null;try {      interpreterProcess = getOrCreateInterpreterProcess();    } catch (IOException e) {throw new InterpreterException(e);    }if (!interpreterProcess.isRunning()) {return new InterpreterResult(InterpreterResult.Code.ERROR,"Interpreter process is not running\n" + interpreterProcess.getErrorMessage());    }return interpreterProcess.callRemoteFunction(client -> {          RemoteInterpreterResult remoteResult = client.interpret(              sessionId, className, st, convert(context));Map remoteConfig = (Map) GSON.fromJson(              remoteResult.getConfig(), new TypeToken>() {              }.getType());          context.getConfig().clear();if (remoteConfig != null) {            context.getConfig().putAll(remoteConfig);          }          GUI currentGUI = context.getGui();          GUI currentNoteGUI = context.getNoteGui();if (form == FormType.NATIVE) {            GUI remoteGui = GUI.fromJson(remoteResult.getGui());            GUI remoteNoteGui = GUI.fromJson(remoteResult.getNoteGui());            currentGUI.clear();            currentGUI.setParams(remoteGui.getParams());            currentGUI.setForms(remoteGui.getForms());            currentNoteGUI.setParams(remoteNoteGui.getParams());            currentNoteGUI.setForms(remoteNoteGui.getForms());          } else if (form == FormType.SIMPLE) {            final Map currentForms = currentGUI.getForms();            final Map currentParams = currentGUI.getParams();            final GUI remoteGUI = GUI.fromJson(remoteResult.getGui());            final Map remoteForms = remoteGUI.getForms();            final Map remoteParams = remoteGUI.getParams();            currentForms.putAll(remoteForms);            currentParams.putAll(remoteParams);          }
return convert(remoteResult);        }    );
  }

其中getOrCreateInterpreterProcess()一路点下去最终是去调用zeppelin-zengine/src/main/java/org/apache/zeppelin/interpreter/remote/ExecRemoteInterpreterProcess.java 里面的start 方法，通过 commons-exec命令执行shell 或者cmd 脚本(bin/interpreter.sh) 启动一个独立的进程，shell 脚本里面具体执行的类(org.apache.zeppelin.interpreter.remote.RemoteInterpreterServer)，和前一篇文章interpreter 原理分析相呼应

@Overridepublic void start(String userName) throws IOException {// start server process    CommandLine cmdLine = CommandLine.parse(interpreterRunner);    cmdLine.addArgument("-d", false);    cmdLine.addArgument(getInterpreterDir(), false);    cmdLine.addArgument("-c", false);    cmdLine.addArgument(getIntpEventServerHost(), false);    cmdLine.addArgument("-p", false);    cmdLine.addArgument(String.valueOf(intpEventServerPort), false);    cmdLine.addArgument("-r", false);    cmdLine.addArgument(getInterpreterPortRange(), false);    cmdLine.addArgument("-i", false);    cmdLine.addArgument(getInterpreterGroupId(), false);if (isUserImpersonated() && !userName.equals("anonymous")) {      cmdLine.addArgument("-u", false);      cmdLine.addArgument(userName, false);    }    cmdLine.addArgument("-l", false);    cmdLine.addArgument(getLocalRepoDir(), false);    cmdLine.addArgument("-g", false);    cmdLine.addArgument(getInterpreterSettingName(), false);
    interpreterProcessLauncher = new InterpreterProcessLauncher(cmdLine, getEnv());    interpreterProcessLauncher.launch();    interpreterProcessLauncher.waitForReady(getConnectTimeout());if (interpreterProcessLauncher.isLaunchTimeout()) {throw new IOException(String.format("Interpreter Process creation is time out in %d seconds", getConnectTimeout() / 1000) + "\n"              + "You can increase timeout threshold via "              + "setting zeppelin.interpreter.connect.timeout of this interpreter.\n"              + interpreterProcessLauncher.getErrorMessage());    }
if (!interpreterProcessLauncher.isRunning()) {throw new IOException("Fail to launch interpreter process:\n" + interpreterProcessLauncher.getErrorMessage());    }
if (isHadoopClientAvailable()) {String launchOutput = interpreterProcessLauncher.getProcessLaunchOutput();      Matcher m = YARN_APP_PATTER.matcher(launchOutput);if (m.find()) {String appId = m.group(1);        LOGGER.info("Detected yarn app: {}, add it to YarnAppMonitor", appId);        YarnAppMonitor.get().addYarnApp(ConverterUtils.toApplicationId(appId), this);      }    }  }

而实际调用thrift server 端服务的client 端代码

上述图片是代码运行的log，可以帮助我们定位代码的运行顺序

({FIFO-RemoteInterpreter-python-shared_process-shared_session-1} ProcessLauncher.java[launch]:96) - Process is launched: [.\\bin\interpreter.cmd, -d, ./\interpreter/python, -c, 10.4.144.223, -p, 52945, -r, :, -i, python-shared_process, -l, ./\local-repo\python, -g, python]
({FIFO-RemoteInterpreter-md-shared_process-shared_session-1} ProcessLauncher.java[launch]:96) - Process is launched: [.\\bin\interpreter.cmd, -d, ./\interpreter/md, -c, 10.4.144.223, -p, 52945, -r, :, -i, md-shared_process, -l, ./\local-repo\md, -g, md]
({FIFO-RemoteInterpreter-jdbc-shared_process-shared_session-1} ProcessLauncher.java[launch]:96) - Process is launched: [.\\bin\interpreter.cmd, -d, ./\interpreter/jdbc, -c, 10.4.144.223, -p, 52945, -r, :, -i, jdbc-shared_process, -l, ./\local-repo\jdbc, -g, jdbc]

‍

Apache Zeppelin系列教程第六篇——Zengine调用Interpreter原理分析