[SPARK-6324] [CORE] Centralize handling of script usage messages.

Reorganize code so that the launcher library handles most of the work
of printing usage messages, instead of having an awkward protocol between
the library and the scripts for that.

This mostly applies to SparkSubmit, since the launcher lib does not do
command line parsing for classes invoked in other ways, and thus cannot
handle failures for those. Most scripts end up going through SparkSubmit,
though, so it all works.

The change adds a new, internal command line switch, "--usage-error",
which prints the usage message and exits with a non-zero status. Scripts
can override the command printed in the usage message by setting an
environment variable - this avoids having to grep the output of
SparkSubmit to remove references to the "spark-submit" script.

The only sub-optimal part of the change is the special handling for the
spark-sql usage, which is now done in SparkSubmitArguments.

Author: Marcelo Vanzin <vanzin@cloudera.com>

Closes #5841 from vanzin/SPARK-6324 and squashes the following commits:

2821481 [Marcelo Vanzin] Merge branch 'master' into SPARK-6324
bf139b5 [Marcelo Vanzin] Filter output of Spark SQL CLI help.
c6609bf [Marcelo Vanzin] Fix exit code never being used when printing usage messages.
6bc1b41 [Marcelo Vanzin] [SPARK-6324] [core] Centralize handling of script usage messages.

15 files changed, 147 insertions, 192 deletions

diff --git a/bin/pyspark b/bin/pyspark
index 7cb19c51b4..f9dbddfa53 100755
--- a/bin/pyspark
+++ b/bin/pyspark
@@ -17,24 +17,10 @@
 # limitations under the License.
 #
 
-# Figure out where Spark is installed
 export SPARK_HOME="$(cd "`dirname "$0"`"/..; pwd)"
 
 source "$SPARK_HOME"/bin/load-spark-env.sh
-
-function usage() {
-  if [ -n "$1" ]; then
-    echo $1
-  fi
-  echo "Usage: ./bin/pyspark [options]" 1>&2
-  "$SPARK_HOME"/bin/spark-submit --help 2>&1 | grep -v Usage 1>&2
-  exit $2
-}
-export -f usage
-
-if [[ "$@" = *--help ]] || [[ "$@" = *-h ]]; then
-  usage
-fi
+export _SPARK_CMD_USAGE="Usage: ./bin/pyspark [options]"
 
 # In Spark <= 1.1, setting IPYTHON=1 would cause the driver to be launched using the `ipython`
 # executable, while the worker would still be launched using PYSPARK_PYTHON.
diff --git a/bin/pyspark2.cmd b/bin/pyspark2.cmd
index 09b4149c2a..45e9e3def5 100644
--- a/bin/pyspark2.cmd
+++ b/bin/pyspark2.cmd
@@ -21,6 +21,7 @@ rem Figure out where the Spark framework is installed
 set SPARK_HOME=%~dp0..
 
 call %SPARK_HOME%\bin\load-spark-env.cmd
+set _SPARK_CMD_USAGE=Usage: bin\pyspark.cmd [options]
 
 rem Figure out which Python to use.
 if "x%PYSPARK_DRIVER_PYTHON%"=="x" (
diff --git a/bin/spark-class b/bin/spark-class
index c49d97ce5c..7bb1afe4b4 100755
--- a/bin/spark-class
+++ b/bin/spark-class
@@ -16,18 +16,12 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 #
-set -e
 
 # Figure out where Spark is installed
 export SPARK_HOME="$(cd "`dirname "$0"`"/..; pwd)"
 
 . "$SPARK_HOME"/bin/load-spark-env.sh
 
-if [ -z "$1" ]; then
-  echo "Usage: spark-class <class> [<args>]" 1>&2
-  exit 1
-fi
-
 # Find the java binary
 if [ -n "${JAVA_HOME}" ]; then
   RUNNER="${JAVA_HOME}/bin/java"
@@ -98,9 +92,4 @@ CMD=()
 while IFS= read -d '' -r ARG; do
   CMD+=("$ARG")
 done < <("$RUNNER" -cp "$LAUNCH_CLASSPATH" org.apache.spark.launcher.Main "$@")
-
-if [ "${CMD[0]}" = "usage" ]; then
-  "${CMD[@]}"
-else
-  exec "${CMD[@]}"
-fi
+exec "${CMD[@]}"
diff --git a/bin/spark-shell b/bin/spark-shell
index b3761b5e13..a6dc863d83 100755
--- a/bin/spark-shell
+++ b/bin/spark-shell
@@ -29,20 +29,7 @@ esac
 set -o posix
 
 export FWDIR="$(cd "`dirname "$0"`"/..; pwd)"
-
-usage() {
-  if [ -n "$1" ]; then
-    echo "$1"
-  fi
-  echo "Usage: ./bin/spark-shell [options]"
-  "$FWDIR"/bin/spark-submit --help 2>&1 | grep -v Usage 1>&2
-  exit "$2"
-}
-export -f usage
-
-if [[ "$@" = *--help ]] || [[ "$@" = *-h ]]; then
-  usage "" 0
-fi
+export _SPARK_CMD_USAGE="Usage: ./bin/spark-shell [options]"
 
 # SPARK-4161: scala does not assume use of the java classpath,
 # so we need to add the "-Dscala.usejavacp=true" flag manually. We
diff --git a/bin/spark-shell2.cmd b/bin/spark-shell2.cmd
index 00fd30fa38..251309d67f 100644
--- a/bin/spark-shell2.cmd
+++ b/bin/spark-shell2.cmd
@@ -18,12 +18,7 @@ rem limitations under the License.
 rem
 
 set SPARK_HOME=%~dp0..
-
-echo "%*" | findstr " \<--help\> \<-h\>" >nul
-if %ERRORLEVEL% equ 0 (
-  call :usage
-  exit /b 0
-)
+set _SPARK_CMD_USAGE=Usage: .\bin\spark-shell.cmd [options]
 
 rem SPARK-4161: scala does not assume use of the java classpath,
 rem so we need to add the "-Dscala.usejavacp=true" flag manually. We
@@ -37,16 +32,4 @@ if "x%SPARK_SUBMIT_OPTS%"=="x" (
 set SPARK_SUBMIT_OPTS="%SPARK_SUBMIT_OPTS% -Dscala.usejavacp=true"
 
 :run_shell
-call %SPARK_HOME%\bin\spark-submit2.cmd --class org.apache.spark.repl.Main %*
-set SPARK_ERROR_LEVEL=%ERRORLEVEL%
-if not "x%SPARK_LAUNCHER_USAGE_ERROR%"=="x" (
-  call :usage
-  exit /b 1
-)
-exit /b %SPARK_ERROR_LEVEL%
-
-:usage
-echo %SPARK_LAUNCHER_USAGE_ERROR%
-echo "Usage: .\bin\spark-shell.cmd [options]" >&2
-call %SPARK_HOME%\bin\spark-submit2.cmd --help 2>&1 | findstr /V "Usage" 1>&2
-goto :eof
+%SPARK_HOME%\bin\spark-submit2.cmd --class org.apache.spark.repl.Main %*
diff --git a/bin/spark-sql b/bin/spark-sql
index ca1729f4cf..4ea7bc6e39 100755
--- a/bin/spark-sql
+++ b/bin/spark-sql
@@ -17,41 +17,6 @@
 # limitations under the License.
 #
 
-#
-# Shell script for starting the Spark SQL CLI
-
-# Enter posix mode for bash
-set -o posix
-
-# NOTE: This exact class name is matched downstream by SparkSubmit.
-# Any changes need to be reflected there.
-export CLASS="org.apache.spark.sql.hive.thriftserver.SparkSQLCLIDriver"
-
-# Figure out where Spark is installed
 export FWDIR="$(cd "`dirname "$0"`"/..; pwd)"
-
-function usage {
-  if [ -n "$1" ]; then
-    echo "$1"
-  fi
-  echo "Usage: ./bin/spark-sql [options] [cli option]"
-  pattern="usage"
-  pattern+="\|Spark assembly has been built with Hive"
-  pattern+="\|NOTE: SPARK_PREPEND_CLASSES is set"
-  pattern+="\|Spark Command: "
-  pattern+="\|--help"
-  pattern+="\|======="
-
-  "$FWDIR"/bin/spark-submit --help 2>&1 | grep -v Usage 1>&2
-  echo
-  echo "CLI options:"
-  "$FWDIR"/bin/spark-class "$CLASS" --help 2>&1 | grep -v "$pattern" 1>&2
-  exit "$2"
-}
-export -f usage
-
-if [[ "$@" = *--help ]] || [[ "$@" = *-h ]]; then
-  usage "" 0
-fi
-
-exec "$FWDIR"/bin/spark-submit --class "$CLASS" "$@"
+export _SPARK_CMD_USAGE="Usage: ./bin/spark-sql [options] [cli option]"
+exec "$FWDIR"/bin/spark-submit --class org.apache.spark.sql.hive.thriftserver.SparkSQLCLIDriver "$@"
diff --git a/bin/spark-submit b/bin/spark-submit
index 0e0afe71a0..255378b0f0 100755
--- a/bin/spark-submit
+++ b/bin/spark-submit
@@ -22,16 +22,4 @@ SPARK_HOME="$(cd "`dirname "$0"`"/..; pwd)"
 # disable randomized hash for string in Python 3.3+
 export PYTHONHASHSEED=0
 
-# Only define a usage function if an upstream script hasn't done so.
-if ! type -t usage >/dev/null 2>&1; then
-  usage() {
-    if [ -n "$1" ]; then
-      echo "$1"
-    fi
-    "$SPARK_HOME"/bin/spark-class org.apache.spark.deploy.SparkSubmit --help
-    exit "$2"
-  }
-  export -f usage
-fi
-
 exec "$SPARK_HOME"/bin/spark-class org.apache.spark.deploy.SparkSubmit "$@"
diff --git a/bin/spark-submit2.cmd b/bin/spark-submit2.cmd
index d3fc4a5cc3..651376e526 100644
--- a/bin/spark-submit2.cmd
+++ b/bin/spark-submit2.cmd
@@ -24,15 +24,4 @@ rem disable randomized hash for string in Python 3.3+
 set PYTHONHASHSEED=0
 
 set CLASS=org.apache.spark.deploy.SparkSubmit
-call %~dp0spark-class2.cmd %CLASS% %*
-set SPARK_ERROR_LEVEL=%ERRORLEVEL%
-if not "x%SPARK_LAUNCHER_USAGE_ERROR%"=="x" (
-  call :usage
-  exit /b 1
-)
-exit /b %SPARK_ERROR_LEVEL%
-
-:usage
-echo %SPARK_LAUNCHER_USAGE_ERROR%
-call %SPARK_HOME%\bin\spark-class2.cmd %CLASS% --help
-goto :eof
+%~dp0spark-class2.cmd %CLASS% %*
diff --git a/bin/sparkR b/bin/sparkR
index 8c918e2b09..464c29f369 100755
--- a/bin/sparkR
+++ b/bin/sparkR
@@ -17,23 +17,7 @@
 # limitations under the License.
 #
 
-# Figure out where Spark is installed
 export SPARK_HOME="$(cd "`dirname "$0"`"/..; pwd)"
-
 source "$SPARK_HOME"/bin/load-spark-env.sh
-
-function usage() {
-  if [ -n "$1" ]; then
-    echo $1
-  fi
-  echo "Usage: ./bin/sparkR [options]" 1>&2
-  "$SPARK_HOME"/bin/spark-submit --help 2>&1 | grep -v Usage 1>&2
-  exit $2
-}
-export -f usage
-
-if [[ "$@" = *--help ]] || [[ "$@" = *-h ]]; then
-  usage
-fi
-
+export _SPARK_CMD_USAGE="Usage: ./bin/sparkR [options]"
 exec "$SPARK_HOME"/bin/spark-submit sparkr-shell-main "$@"
diff --git a/core/src/main/scala/org/apache/spark/deploy/SparkSubmit.scala b/core/src/main/scala/org/apache/spark/deploy/SparkSubmit.scala
index 8cf4d58847..3aa3f948e8 100644
--- a/core/src/main/scala/org/apache/spark/deploy/SparkSubmit.scala
+++ b/core/src/main/scala/org/apache/spark/deploy/SparkSubmit.scala
@@ -82,13 +82,13 @@ object SparkSubmit {
   private val CLASS_NOT_FOUND_EXIT_STATUS = 101
 
   // Exposed for testing
-  private[spark] var exitFn: () => Unit = () => System.exit(1)
+  private[spark] var exitFn: Int => Unit = (exitCode: Int) => System.exit(exitCode)
   private[spark] var printStream: PrintStream = System.err
   private[spark] def printWarning(str: String): Unit = printStream.println("Warning: " + str)
   private[spark] def printErrorAndExit(str: String): Unit = {
     printStream.println("Error: " + str)
     printStream.println("Run with --help for usage help or --verbose for debug output")
-    exitFn()
+    exitFn(1)
   }
   private[spark] def printVersionAndExit(): Unit = {
     printStream.println("""Welcome to
@@ -99,7 +99,7 @@ object SparkSubmit {
       /_/
                         """.format(SPARK_VERSION))
     printStream.println("Type --help for more information.")
-    exitFn()
+    exitFn(0)
   }
 
   def main(args: Array[String]): Unit = {
@@ -160,7 +160,7 @@ object SparkSubmit {
             // detect exceptions with empty stack traces here, and treat them differently.
             if (e.getStackTrace().length == 0) {
               printStream.println(s"ERROR: ${e.getClass().getName()}: ${e.getMessage()}")
-              exitFn()
+              exitFn(1)
             } else {
               throw e
             }
@@ -700,7 +700,7 @@ object SparkSubmit {
   /**
    * Return whether the given main class represents a sql shell.
    */
-  private def isSqlShell(mainClass: String): Boolean = {
+  private[deploy] def isSqlShell(mainClass: String): Boolean = {
     mainClass == "org.apache.spark.sql.hive.thriftserver.SparkSQLCLIDriver"
   }
 
diff --git a/core/src/main/scala/org/apache/spark/deploy/SparkSubmitArguments.scala b/core/src/main/scala/org/apache/spark/deploy/SparkSubmitArguments.scala
index cc6a7bd9f4..b7429a901e 100644
--- a/core/src/main/scala/org/apache/spark/deploy/SparkSubmitArguments.scala
+++ b/core/src/main/scala/org/apache/spark/deploy/SparkSubmitArguments.scala
@@ -17,12 +17,15 @@
 
 package org.apache.spark.deploy
 
+import java.io.{ByteArrayOutputStream, PrintStream}
+import java.lang.reflect.InvocationTargetException
 import java.net.URI
 import java.util.{List => JList}
 import java.util.jar.JarFile
 
 import scala.collection.JavaConversions._
 import scala.collection.mutable.{ArrayBuffer, HashMap}
+import scala.io.Source
 
 import org.apache.spark.deploy.SparkSubmitAction._
 import org.apache.spark.launcher.SparkSubmitArgumentsParser
@@ -412,6 +415,9 @@ private[deploy] class SparkSubmitArguments(args: Seq[String], env: Map[String, S
       case VERSION =>
         SparkSubmit.printVersionAndExit()
 
+      case USAGE_ERROR =>
+        printUsageAndExit(1)
+
       case _ =>
         throw new IllegalArgumentException(s"Unexpected argument '$opt'.")
     }
@@ -449,11 +455,14 @@ private[deploy] class SparkSubmitArguments(args: Seq[String], env: Map[String, S
     if (unknownParam != null) {
       outStream.println("Unknown/unsupported param " + unknownParam)
     }
-    outStream.println(
+    val command = sys.env.get("_SPARK_CMD_USAGE").getOrElse(
       """Usage: spark-submit [options] <app jar | python file> [app arguments]
         |Usage: spark-submit --kill [submission ID] --master [spark://...]
-        |Usage: spark-submit --status [submission ID] --master [spark://...]
-        |
+        |Usage: spark-submit --status [submission ID] --master [spark://...]""".stripMargin)
+    outStream.println(command)
+
+    outStream.println(
+      """
         |Options:
         |  --master MASTER_URL         spark://host:port, mesos://host:port, yarn, or local.
         |  --deploy-mode DEPLOY_MODE   Whether to launch the driver program locally ("client") or
@@ -525,6 +534,65 @@ private[deploy] class SparkSubmitArguments(args: Seq[String], env: Map[String, S
         |                              delegation tokens periodically.
       """.stripMargin
     )
-    SparkSubmit.exitFn()
+
+    if (SparkSubmit.isSqlShell(mainClass)) {
+      outStream.println("CLI options:")
+      outStream.println(getSqlShellOptions())
+    }
+
+    SparkSubmit.exitFn(exitCode)
   }
+
+  /**
+   * Run the Spark SQL CLI main class with the "--help" option and catch its output. Then filter
+   * the results to remove unwanted lines.
+   *
+   * Since the CLI will call `System.exit()`, we install a security manager to prevent that call
+   * from working, and restore the original one afterwards.
+   */
+  private def getSqlShellOptions(): String = {
+    val currentOut = System.out
+    val currentErr = System.err
+    val currentSm = System.getSecurityManager()
+    try {
+      val out = new ByteArrayOutputStream()
+      val stream = new PrintStream(out)
+      System.setOut(stream)
+      System.setErr(stream)
+
+      val sm = new SecurityManager() {
+        override def checkExit(status: Int): Unit = {
+          throw new SecurityException()
+        }
+
+        override def checkPermission(perm: java.security.Permission): Unit = {}
+      }
+      System.setSecurityManager(sm)
+
+      try {
+        Class.forName(mainClass).getMethod("main", classOf[Array[String]])
+          .invoke(null, Array(HELP))
+      } catch {
+        case e: InvocationTargetException =>
+          // Ignore SecurityException, since we throw it above.
+          if (!e.getCause().isInstanceOf[SecurityException]) {
+            throw e
+          }
+      }
+
+      stream.flush()
+
+      // Get the output and discard any unnecessary lines from it.
+      Source.fromString(new String(out.toByteArray())).getLines
+        .filter { line =>
+          !line.startsWith("log4j") && !line.startsWith("usage")
+        }
+        .mkString("\n")
+    } finally {
+      System.setSecurityManager(currentSm)
+      System.setOut(currentOut)
+      System.setErr(currentErr)
+    }
+  }
+
 }
diff --git a/core/src/test/scala/org/apache/spark/deploy/SparkSubmitSuite.scala b/core/src/test/scala/org/apache/spark/deploy/SparkSubmitSuite.scala
index 46369457f0..46ea28d0f1 100644
--- a/core/src/test/scala/org/apache/spark/deploy/SparkSubmitSuite.scala
+++ b/core/src/test/scala/org/apache/spark/deploy/SparkSubmitSuite.scala
@@ -62,7 +62,7 @@ class SparkSubmitSuite
     SparkSubmit.printStream = printStream
 
     @volatile var exitedCleanly = false
-    SparkSubmit.exitFn = () => exitedCleanly = true
+    SparkSubmit.exitFn = (_) => exitedCleanly = true
 
     val thread = new Thread {
       override def run() = try {
diff --git a/launcher/src/main/java/org/apache/spark/launcher/Main.java b/launcher/src/main/java/org/apache/spark/launcher/Main.java
index 929b29a49e..62492f9baf 100644
--- a/launcher/src/main/java/org/apache/spark/launcher/Main.java
+++ b/launcher/src/main/java/org/apache/spark/launcher/Main.java
@@ -53,21 +53,33 @@ class Main {
     List<String> args = new ArrayList<String>(Arrays.asList(argsArray));
     String className = args.remove(0);
 
-    boolean printLaunchCommand;
-    boolean printUsage;
+    boolean printLaunchCommand = !isEmpty(System.getenv("SPARK_PRINT_LAUNCH_COMMAND"));
     AbstractCommandBuilder builder;
-    try {
-      if (className.equals("org.apache.spark.deploy.SparkSubmit")) {
+    if (className.equals("org.apache.spark.deploy.SparkSubmit")) {
+      try {
         builder = new SparkSubmitCommandBuilder(args);
-      } else {
-        builder = new SparkClassCommandBuilder(className, args);
+      } catch (IllegalArgumentException e) {
+        printLaunchCommand = false;
+        System.err.println("Error: " + e.getMessage());
+        System.err.println();
+
+        MainClassOptionParser parser = new MainClassOptionParser();
+        try {
+          parser.parse(args);
+        } catch (Exception ignored) {
+          // Ignore parsing exceptions.
+        }
+
+        List<String> help = new ArrayList<String>();
+        if (parser.className != null) {
+          help.add(parser.CLASS);
+          help.add(parser.className);
+        }
+        help.add(parser.USAGE_ERROR);
+        builder = new SparkSubmitCommandBuilder(help);
       }
-      printLaunchCommand = !isEmpty(System.getenv("SPARK_PRINT_LAUNCH_COMMAND"));
-      printUsage = false;
-    } catch (IllegalArgumentException e) {
-      builder = new UsageCommandBuilder(e.getMessage());
-      printLaunchCommand = false;
-      printUsage = true;
+    } else {
+      builder = new SparkClassCommandBuilder(className, args);
     }
 
     Map<String, String> env = new HashMap<String, String>();
@@ -78,13 +90,7 @@ class Main {
     }
 
     if (isWindows()) {
-      // When printing the usage message, we can't use "cmd /v" since that prevents the env
-      // variable from being seen in the caller script. So do not call prepareWindowsCommand().
-      if (printUsage) {
-        System.out.println(join(" ", cmd));
-      } else {
-        System.out.println(prepareWindowsCommand(cmd, env));
-      }
+      System.out.println(prepareWindowsCommand(cmd, env));
     } else {
       // In bash, use NULL as the arg separator since it cannot be used in an argument.
       List<String> bashCmd = prepareBashCommand(cmd, env);
@@ -135,33 +141,30 @@ class Main {
   }
 
   /**
-   * Internal builder used when command line parsing fails. This will behave differently depending
-   * on the platform:
-   *
-   * - On Unix-like systems, it will print a call to the "usage" function with two arguments: the
-   *   the error string, and the exit code to use. The function is expected to print the command's
-   *   usage and exit with the provided exit code. The script should use "export -f usage" after
-   *   declaring a function called "usage", so that the function is available to downstream scripts.
-   *
-   * - On Windows it will set the variable "SPARK_LAUNCHER_USAGE_ERROR" to the usage error message.
-   *   The batch script should check for this variable and print its usage, since batch scripts
-   *   don't really support the "export -f" functionality used in bash.
+   * A parser used when command line parsing fails for spark-submit. It's used as a best-effort
+   * at trying to identify the class the user wanted to invoke, since that may require special
+   * usage strings (handled by SparkSubmitArguments).
    */
-  private static class UsageCommandBuilder extends AbstractCommandBuilder {
+  private static class MainClassOptionParser extends SparkSubmitOptionParser {
 
-    private final String message;
+    String className;
 
-    UsageCommandBuilder(String message) {
-      this.message = message;
+    @Override
+    protected boolean handle(String opt, String value) {
+      if (opt == CLASS) {
+        className = value;
+      }
+      return false;
     }
 
     @Override
-    public List<String> buildCommand(Map<String, String> env) {
-      if (isWindows()) {
-        return Arrays.asList("set", "SPARK_LAUNCHER_USAGE_ERROR=" + message);
-      } else {
-        return Arrays.asList("usage", message, "1");
-      }
+    protected boolean handleUnknown(String opt) {
+      return false;
+    }
+
+    @Override
+    protected void handleExtraArgs(List<String> extra) {
+
     }
 
   }
diff --git a/launcher/src/main/java/org/apache/spark/launcher/SparkSubmitCommandBuilder.java b/launcher/src/main/java/org/apache/spark/launcher/SparkSubmitCommandBuilder.java
index 7d387d406e..3e5a2820b6 100644
--- a/launcher/src/main/java/org/apache/spark/launcher/SparkSubmitCommandBuilder.java
+++ b/launcher/src/main/java/org/apache/spark/launcher/SparkSubmitCommandBuilder.java
@@ -77,6 +77,7 @@ class SparkSubmitCommandBuilder extends AbstractCommandBuilder {
   }
 
   private final List<String> sparkArgs;
+  private final boolean printHelp;
 
   /**
    * Controls whether mixing spark-submit arguments with app arguments is allowed. This is needed
@@ -87,10 +88,11 @@ class SparkSubmitCommandBuilder extends AbstractCommandBuilder {
 
   SparkSubmitCommandBuilder() {
     this.sparkArgs = new ArrayList<String>();
+    this.printHelp = false;
   }
 
   SparkSubmitCommandBuilder(List<String> args) {
-    this();
+    this.sparkArgs = new ArrayList<String>();
     List<String> submitArgs = args;
     if (args.size() > 0 && args.get(0).equals(PYSPARK_SHELL)) {
       this.allowsMixedArguments = true;
@@ -104,14 +106,16 @@ class SparkSubmitCommandBuilder extends AbstractCommandBuilder {
       this.allowsMixedArguments = false;
     }
 
-    new OptionParser().parse(submitArgs);
+    OptionParser parser = new OptionParser();
+    parser.parse(submitArgs);
+    this.printHelp = parser.helpRequested;
   }
 
   @Override
   public List<String> buildCommand(Map<String, String> env) throws IOException {
-    if (PYSPARK_SHELL_RESOURCE.equals(appResource)) {
+    if (PYSPARK_SHELL_RESOURCE.equals(appResource) && !printHelp) {
       return buildPySparkShellCommand(env);
-    } else if (SPARKR_SHELL_RESOURCE.equals(appResource)) {
+    } else if (SPARKR_SHELL_RESOURCE.equals(appResource) && !printHelp) {
       return buildSparkRCommand(env);
     } else {
       return buildSparkSubmitCommand(env);
@@ -311,6 +315,8 @@ class SparkSubmitCommandBuilder extends AbstractCommandBuilder {
 
   private class OptionParser extends SparkSubmitOptionParser {
 
+    boolean helpRequested = false;
+
     @Override
     protected boolean handle(String opt, String value) {
       if (opt.equals(MASTER)) {
@@ -341,6 +347,9 @@ class SparkSubmitCommandBuilder extends AbstractCommandBuilder {
           allowsMixedArguments = true;
           appResource = specialClasses.get(value);
         }
+      } else if (opt.equals(HELP) || opt.equals(USAGE_ERROR)) {
+        helpRequested = true;
+        sparkArgs.add(opt);
       } else {
         sparkArgs.add(opt);
         if (value != null) {
@@ -360,6 +369,7 @@ class SparkSubmitCommandBuilder extends AbstractCommandBuilder {
         appArgs.add(opt);
         return true;
       } else {
+        checkArgument(!opt.startsWith("-"), "Unrecognized option: %s", opt);
         sparkArgs.add(opt);
         return false;
       }
diff --git a/launcher/src/main/java/org/apache/spark/launcher/SparkSubmitOptionParser.java b/launcher/src/main/java/org/apache/spark/launcher/SparkSubmitOptionParser.java
index 2290000876..b88bba883a 100644
--- a/launcher/src/main/java/org/apache/spark/launcher/SparkSubmitOptionParser.java
+++ b/launcher/src/main/java/org/apache/spark/launcher/SparkSubmitOptionParser.java
@@ -61,6 +61,7 @@ class SparkSubmitOptionParser {
   // Options that do not take arguments.
   protected final String HELP = "--help";
   protected final String SUPERVISE = "--supervise";
+  protected final String USAGE_ERROR = "--usage-error";
   protected final String VERBOSE = "--verbose";
   protected final String VERSION = "--version";
 
@@ -120,6 +121,7 @@ class SparkSubmitOptionParser {
   final String[][] switches = {
     { HELP, "-h" },
     { SUPERVISE },
+    { USAGE_ERROR },
     { VERBOSE, "-v" },
     { VERSION },
   };