diff --git a/core/src/main/scala/org/apache/spark/api/python/PythonUtils.scala b/core/src/main/scala/org/apache/spark/api/python/PythonUtils.scala
index 1c92af45f869a..acbaba6791850 100644
--- a/core/src/main/scala/org/apache/spark/api/python/PythonUtils.scala
+++ b/core/src/main/scala/org/apache/spark/api/python/PythonUtils.scala
@@ -35,10 +35,6 @@ private[spark] object PythonUtils {
       pythonPath += Seq(sparkHome, "python", "lib", "py4j-0.8.2.1-src.zip").mkString(File.separator)
     }
     pythonPath ++= SparkContext.jarOfObject(this)
-    sys.env.get("PYSPARK_ARCHIVES_PATH") match {
-      case Some(path) => pythonPath += path
-      case None => // do nothing
-    }
     pythonPath.mkString(File.pathSeparator)
   }
 
diff --git a/core/src/main/scala/org/apache/spark/api/python/PythonWorkerFactory.scala b/core/src/main/scala/org/apache/spark/api/python/PythonWorkerFactory.scala
index e314408c067e9..3f6e0dd6050d2 100644
--- a/core/src/main/scala/org/apache/spark/api/python/PythonWorkerFactory.scala
+++ b/core/src/main/scala/org/apache/spark/api/python/PythonWorkerFactory.scala
@@ -49,7 +49,7 @@ private[spark] class PythonWorkerFactory(pythonExec: String, envVars: Map[String
 
   val pythonPath = PythonUtils.mergePythonPaths(
     PythonUtils.sparkPythonPath,
-    envVars.getOrElse("PYTHONPATH", ""),
+    envVars.getOrElse("PYTHONPATH", sys.env.getOrElse("PYSPARK_ARCHIVES_PATH", "")),
     sys.env.getOrElse("PYTHONPATH", ""))
 
   def create(): Socket = {
diff --git a/core/src/main/scala/org/apache/spark/deploy/PythonRunner.scala b/core/src/main/scala/org/apache/spark/deploy/PythonRunner.scala
index 53e18c4bcec23..569abcfff68f5 100644
--- a/core/src/main/scala/org/apache/spark/deploy/PythonRunner.scala
+++ b/core/src/main/scala/org/apache/spark/deploy/PythonRunner.scala
@@ -52,6 +52,7 @@ object PythonRunner {
     pathElements ++= formattedPyFiles
     pathElements += PythonUtils.sparkPythonPath
     pathElements += sys.env.getOrElse("PYTHONPATH", "")
+    pathElements += sys.env.getOrElse("PYSPARK_ARCHIVES_PATH", "")
     val pythonPath = PythonUtils.mergePythonPaths(pathElements: _*)
 
     // Launch Python process
diff --git a/docs/submitting-applications.md b/docs/submitting-applications.md
index 73afbe06df606..3ecbf2308cd44 100644
--- a/docs/submitting-applications.md
+++ b/docs/submitting-applications.md
@@ -22,9 +22,6 @@ For Python, you can use the `--py-files` argument of `spark-submit` to add `.py`
 files to be distributed with your application. If you depend on multiple Python files we recommend
 packaging them into a `.zip` or `.egg`.
 
-As Python can not read files from assembly jar which packaged by JDK1.7+, so packaging pyspark into a
-`.zip`(the name contains "pyspark") and use `--py-files` argument of `spark-submit` to distribute it.
-
 # Launching Applications with spark-submit
 
 Once a user application is bundled, it can be launched using the `bin/spark-submit` script.
diff --git a/yarn/src/main/scala/org/apache/spark/deploy/yarn/Client.scala b/yarn/src/main/scala/org/apache/spark/deploy/yarn/Client.scala
index c1effd3c8a718..b1ee4096f4b6c 100644
--- a/yarn/src/main/scala/org/apache/spark/deploy/yarn/Client.scala
+++ b/yarn/src/main/scala/org/apache/spark/deploy/yarn/Client.scala
@@ -247,6 +247,7 @@ private[spark] class Client(
     List(
       (SPARK_JAR, sparkJar(sparkConf), CONF_SPARK_JAR),
       (APP_JAR, args.userJar, CONF_SPARK_USER_JAR),
+      (PYSPARK_ARCHIVES, pysparkArchives(sparkConf), CONF_PYSPARK_ARCHIVES),
       ("log4j.properties", oldLog4jConf.orNull, null)
     ).foreach { case (destName, _localPath, confKey) =>
       val localPath: String = if (_localPath != null) _localPath.trim() else ""
@@ -386,6 +387,12 @@ private[spark] class Client(
     val appStagingDir = getAppStagingDir(appId)
     val localResources = prepareLocalResources(appStagingDir)
     val launchEnv = setupLaunchEnv(appStagingDir)
+    // From SPARK-1920 and SPARK-1520 we know PySpark on Yarn can not work when the assembly jar are
+    // package by JDK 1.7+, so we ship PySpark archives to executors as assembly jar, and add this
+    // path to PYTHONPATH.
+    for ((resPath, res) <- localResources if resPath.contains(PYSPARK_ARCHIVES)) {
+      launchEnv("PYSPARK_ARCHIVES_PATH") = resPath
+    }
     val amContainer = Records.newRecord(classOf[ContainerLaunchContext])
     amContainer.setLocalResources(localResources)
     amContainer.setEnvironment(launchEnv)
@@ -681,9 +688,10 @@ object Client extends Logging {
     new Client(args, sparkConf).run()
   }
 
-  // Alias for the Spark assembly jar and the user jar
+  // Alias for the Spark assembly jar, the user jar and PySpark archives
   val SPARK_JAR: String = "__spark__.jar"
   val APP_JAR: String = "__app__.jar"
+  val PYSPARK_ARCHIVES: String = "__pyspark__.zip"
 
   // URI scheme that identifies local resources
   val LOCAL_SCHEME = "local"
@@ -695,6 +703,9 @@ object Client extends Logging {
   val CONF_SPARK_JAR = "spark.yarn.jar"
   val ENV_SPARK_JAR = "SPARK_JAR"
 
+  // Location of any user-defined PySpark archives
+  val CONF_PYSPARK_ARCHIVES = "spark.pyspark.archives"
+
   // Internal config to propagate the location of the user's jar to the driver/executors
   val CONF_SPARK_USER_JAR = "spark.yarn.user.jar"
 
@@ -733,6 +744,19 @@ object Client extends Logging {
     }
   }
 
+  /**
+   * Find the user-defined PySpark archives if configured, or return default.
+   * The default pyspark.zip is in the same path with assembly jar.
+   */
+  private def pysparkArchives(conf: SparkConf): String = {
+    if (conf.contains(CONF_PYSPARK_ARCHIVES)) {
+      conf.get(CONF_PYSPARK_ARCHIVES)
+    } else {
+      val sparkJarPath = SparkContext.jarOfClass(this.getClass).head
+      sparkJarPath.substring(0, sparkJarPath.lastIndexOf('/')) + "/pyspark.zip"
+    }
+  }
+
   /**
    * Return the path to the given application's staging directory.
    */
diff --git a/yarn/src/main/scala/org/apache/spark/deploy/yarn/ExecutorRunnable.scala b/yarn/src/main/scala/org/apache/spark/deploy/yarn/ExecutorRunnable.scala
index a8341991f64ad..2325ffcb78772 100644
--- a/yarn/src/main/scala/org/apache/spark/deploy/yarn/ExecutorRunnable.scala
+++ b/yarn/src/main/scala/org/apache/spark/deploy/yarn/ExecutorRunnable.scala
@@ -75,12 +75,7 @@ class ExecutorRunnable(
 
     val localResources = prepareLocalResources
     ctx.setLocalResources(localResources)
-    // From SPARK-1920 and SPARK-1520 we know PySpark on Yarn can not work when the assembly jar are
-    // package by JDK 1.7+, so we ship PySpark archives to executors by Yarn with --py-files, and
-    // add this path to PYTHONPATH.
-    for ((resPath, res) <- localResources if resPath.contains("pyspark")) {
-      env("PYSPARK_ARCHIVES_PATH") = resPath
-    }
+
     ctx.setEnvironment(env)
 
     val credentials = UserGroupInformation.getCurrentUser().getCredentials()
@@ -304,6 +299,12 @@ class ExecutorRunnable(
     }
 
     System.getenv().filterKeys(_.startsWith("SPARK")).foreach { case (k, v) => env(k) = v }
+
+    // Add PySpark archives path
+    sys.env.get("PYSPARK_ARCHIVES_PATH") match {
+      case Some(pythonArchivesPath) => env("PYSPARK_ARCHIVES_PATH") = pythonArchivesPath
+      case None =>
+    }
     env
   }
 }