[SPARK-3772] Allow ipython to be used by Pyspark workers; IPython support improvements:

This pull request addresses a few issues related to PySpark's IPython support:

- Fix the remaining uses of the '-u' flag, which IPython doesn't support (see SPARK-3772).
- Change PYSPARK_PYTHON_OPTS to PYSPARK_DRIVER_PYTHON_OPTS, so that the old name is reserved in case we ever want to allow the worker Python options to be customized (this variable was introduced in apache#2554 and hasn't landed in a release yet, so this doesn't break any compatibility).
- Introduce a PYSPARK_DRIVER_PYTHON option that allows the driver to use `ipython` while the workers use a different Python version.
- Attempt to use Python 2.7 by default if PYSPARK_PYTHON is not specified.
- Retain the old semantics for IPYTHON=1 and IPYTHON_OPTS (to avoid breaking existing example programs).

There are more details in a block comment in `bin/pyspark`.

Author: Josh Rosen <[email protected]>

Closes apache#2651 from JoshRosen/SPARK-3772 and squashes the following commits:

7b8eb86 [Josh Rosen] More changes to PySpark python executable configuration:
c4f5778 [Josh Rosen] [SPARK-3772] Allow ipython to be used by Pyspark workers; IPython fixes:

Author: JoshRosen

bin/pyspark

@@ -50,22 +50,47 @@ fi
 
 . "$FWDIR"/bin/load-spark-env.sh
 
-# Figure out which Python executable to use
+# 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.
+#
+# In Spark 1.2, we removed the documentation of the IPYTHON and IPYTHON_OPTS variables and added
+# PYSPARK_DRIVER_PYTHON and PYSPARK_DRIVER_PYTHON_OPTS to allow IPython to be used for the driver.
+# Now, users can simply set PYSPARK_DRIVER_PYTHON=ipython to use IPython and set
+# PYSPARK_DRIVER_PYTHON_OPTS to pass options when starting the Python driver
+# (e.g. PYSPARK_DRIVER_PYTHON_OPTS='notebook').  This supports full customization of the IPython
+# and executor Python executables.
+#
+# For backwards-compatibility, we retain the old IPYTHON and IPYTHON_OPTS variables.
+
+# Determine the Python executable to use if PYSPARK_PYTHON or PYSPARK_DRIVER_PYTHON isn't set:
+if hash python2.7 2>/dev/null; then
+  # Attempt to use Python 2.7, if installed:
+  DEFAULT_PYTHON="python2.7"
+else
+  DEFAULT_PYTHON="python"
+fi
+
+# Determine the Python executable to use for the driver:
+if [[ -n "$IPYTHON_OPTS" || "$IPYTHON" == "1" ]]; then
+  # If IPython options are specified, assume user wants to run IPython
+  # (for backwards-compatibility)
+  PYSPARK_DRIVER_PYTHON_OPTS="$PYSPARK_DRIVER_PYTHON_OPTS $IPYTHON_OPTS"
+  PYSPARK_DRIVER_PYTHON="ipython"
+elif [[ -z "$PYSPARK_DRIVER_PYTHON" ]]; then
+  PYSPARK_DRIVER_PYTHON="${PYSPARK_PYTHON:-"$DEFAULT_PYTHON"}"
+fi
+
+# Determine the Python executable to use for the executors:
 if [[ -z "$PYSPARK_PYTHON" ]]; then
-  if [[ "$IPYTHON" = "1" || -n "$IPYTHON_OPTS" ]]; then
-    # for backward compatibility
-    PYSPARK_PYTHON="ipython"
+  if [[ $PYSPARK_DRIVER_PYTHON == *ipython* && $DEFAULT_PYTHON != "python2.7" ]]; then
+    echo "IPython requires Python 2.7+; please install python2.7 or set PYSPARK_PYTHON" 1>&2
+    exit 1
   else
-    PYSPARK_PYTHON="python"
+    PYSPARK_PYTHON="$DEFAULT_PYTHON"
   fi
 fi
 export PYSPARK_PYTHON
 
-if [[ -z "$PYSPARK_PYTHON_OPTS" && -n "$IPYTHON_OPTS" ]]; then
-  # for backward compatibility
-  PYSPARK_PYTHON_OPTS="$IPYTHON_OPTS"
-fi
-
 # Add the PySpark classes to the Python path:
 export PYTHONPATH="$SPARK_HOME/python/:$PYTHONPATH"
 export PYTHONPATH="$SPARK_HOME/python/lib/py4j-0.8.2.1-src.zip:$PYTHONPATH"
@@ -93,9 +118,9 @@ if [[ -n "$SPARK_TESTING" ]]; then
   unset YARN_CONF_DIR
   unset HADOOP_CONF_DIR
   if [[ -n "$PYSPARK_DOC_TEST" ]]; then
-    exec "$PYSPARK_PYTHON" -m doctest $1
+    exec "$PYSPARK_DRIVER_PYTHON" -m doctest $1
   else
-    exec "$PYSPARK_PYTHON" $1
+    exec "$PYSPARK_DRIVER_PYTHON" $1
   fi
   exit
 fi
@@ -111,5 +136,5 @@ if [[ "$1" =~ \.py$ ]]; then
 else
   # PySpark shell requires special handling downstream
   export PYSPARK_SHELL=1
-  exec "$PYSPARK_PYTHON" $PYSPARK_PYTHON_OPTS
+  exec "$PYSPARK_DRIVER_PYTHON" $PYSPARK_DRIVER_PYTHON_OPTS
 fi

core/src/main/scala/org/apache/spark/api/python/PythonWorkerFactory.scala

@@ -108,10 +108,12 @@ private[spark] class PythonWorkerFactory(pythonExec: String, envVars: Map[String
       serverSocket = new ServerSocket(0, 1, InetAddress.getByAddress(Array(127, 0, 0, 1)))
 
       // Create and start the worker
-      val pb = new ProcessBuilder(Seq(pythonExec, "-u", "-m", "pyspark.worker"))
+      val pb = new ProcessBuilder(Seq(pythonExec, "-m", "pyspark.worker"))
       val workerEnv = pb.environment()
       workerEnv.putAll(envVars)
       workerEnv.put("PYTHONPATH", pythonPath)
+      // This is equivalent to setting the -u flag; we use it because ipython doesn't support -u:
+      workerEnv.put("PYTHONUNBUFFERED", "YES")
       val worker = pb.start()
 
       // Redirect worker stdout and stderr
@@ -149,10 +151,12 @@ private[spark] class PythonWorkerFactory(pythonExec: String, envVars: Map[String
 
       try {
         // Create and start the daemon
-        val pb = new ProcessBuilder(Seq(pythonExec, "-u", "-m", "pyspark.daemon"))
+        val pb = new ProcessBuilder(Seq(pythonExec, "-m", "pyspark.daemon"))
         val workerEnv = pb.environment()
         workerEnv.putAll(envVars)
         workerEnv.put("PYTHONPATH", pythonPath)
+        // This is equivalent to setting the -u flag; we use it because ipython doesn't support -u:
+        workerEnv.put("PYTHONUNBUFFERED", "YES")
         daemon = pb.start()
 
         val in = new DataInputStream(daemon.getInputStream)

core/src/main/scala/org/apache/spark/deploy/PythonRunner.scala

@@ -34,7 +34,8 @@ object PythonRunner {
     val pythonFile = args(0)
     val pyFiles = args(1)
     val otherArgs = args.slice(2, args.length)
-    val pythonExec = sys.env.get("PYSPARK_PYTHON").getOrElse("python") // TODO: get this from conf
+    val pythonExec =
+      sys.env.getOrElse("PYSPARK_DRIVER_PYTHON", sys.env.getOrElse("PYSPARK_PYTHON", "python"))
 
     // Format python file paths before adding them to the PYTHONPATH
     val formattedPythonFile = formatPath(pythonFile)
@@ -57,6 +58,7 @@ object PythonRunner {
     val builder = new ProcessBuilder(Seq(pythonExec, formattedPythonFile) ++ otherArgs)
     val env = builder.environment()
     env.put("PYTHONPATH", pythonPath)
+    // This is equivalent to setting the -u flag; we use it because ipython doesn't support -u:
     env.put("PYTHONUNBUFFERED", "YES") // value is needed to be set to a non-empty string
     env.put("PYSPARK_GATEWAY_PORT", "" + gatewayServer.getListeningPort)
     builder.redirectErrorStream(true) // Ugly but needed for stdout and stderr to synchronize

docs/programming-guide.md

@@ -211,17 +211,17 @@ For a complete list of options, run `pyspark --help`. Behind the scenes,
 
 It is also possible to launch the PySpark shell in [IPython](http://ipython.org), the
 enhanced Python interpreter. PySpark works with IPython 1.0.0 and later. To
-use IPython, set the `PYSPARK_PYTHON` variable to `ipython` when running `bin/pyspark`:
+use IPython, set the `PYSPARK_DRIVER_PYTHON` variable to `ipython` when running `bin/pyspark`:
 
 {% highlight bash %}
-$ PYSPARK_PYTHON=ipython ./bin/pyspark
+$ PYSPARK_DRIVER_PYTHON=ipython ./bin/pyspark
 {% endhighlight %}
 
-You can customize the `ipython` command by setting `PYSPARK_PYTHON_OPTS`. For example, to launch
+You can customize the `ipython` command by setting `PYSPARK_DRIVER_PYTHON_OPTS`. For example, to launch
 the [IPython Notebook](http://ipython.org/notebook.html) with PyLab plot support:
 
 {% highlight bash %}
-$ PYSPARK_PYTHON=ipython PYSPARK_PYTHON_OPTS="notebook --pylab inline" ./bin/pyspark
+$ PYSPARK_DRIVER_PYTHON=ipython PYSPARK_DRIVER_PYTHON_OPTS="notebook --pylab inline" ./bin/pyspark
 {% endhighlight %}
 
 </div>