[SPARK-3342] Add SSDs to block device mapping

On `m3.2xlarge` instances the 2x80GB SSDs are inaccessible if not added to the block device mapping when the instance is created. They work when added with this patch. I have not tested this with other instance types, and I do not know much about this script and EC2 deployment in general. Maybe this code needs to depend on the instance type.

The requirement for this mapping is described in the AWS docs at:
http://docs.aws.amazon.com/AWSEC2/latest/UserGuide/InstanceStorage.html#InstanceStore_UsageScenarios

"For M3 instances, you must specify instance store volumes in the block
device mapping for the instance. When you launch an M3 instance, we
ignore any instance store volumes specified in the block device mapping
for the AMI."

Author: Daniel Darabos <darabos.daniel@gmail.com>

Closes #2081 from darabos/patch-1 and squashes the following commits:

1ceb2c8 [Daniel Darabos] Use %d string interpolation instead of {}.
a1854d7 [Daniel Darabos] Only specify ephemeral device mapping for M3.
e0d9e37 [Daniel Darabos] Create ephemeral device mapping based on get_num_disks().
6b116a6 [Daniel Darabos] Add SSDs to block device mapping

1 files changed, 11 insertions, 1 deletions

diff --git a/ec2/spark_ec2.py b/ec2/spark_ec2.py
index ae4c488f83..7e25df57ee 100755
--- a/ec2/spark_ec2.py
+++ b/ec2/spark_ec2.py
@@ -26,6 +26,7 @@ import os
 import pipes
 import random
 import shutil
+import string
 import subprocess
 import sys
 import tempfile
@@ -34,7 +35,7 @@ import urllib2
 from optparse import OptionParser
 from sys import stderr
 import boto
-from boto.ec2.blockdevicemapping import BlockDeviceMapping, EBSBlockDeviceType
+from boto.ec2.blockdevicemapping import BlockDeviceMapping, BlockDeviceType, EBSBlockDeviceType
 from boto import ec2
 
 # A URL prefix from which to fetch AMI information
@@ -355,6 +356,15 @@ def launch_cluster(conn, opts, cluster_name):
         device.delete_on_termination = True
         block_map["/dev/sdv"] = device
 
+    # AWS ignores the AMI-specified block device mapping for M3 (see SPARK-3342).
+    if opts.instance_type.startswith('m3.'):
+        for i in range(get_num_disks(opts.instance_type)):
+            dev = BlockDeviceType()
+            dev.ephemeral_name = 'ephemeral%d' % i
+            # The first ephemeral drive is /dev/sdb.
+            name = '/dev/sd' + string.letters[i + 1]
+            block_map[name] = dev
+
     # Launch slaves
     if opts.spot_price is not None:
         # Launch spot instances with the requested price