From 44d3a6a75209370b1648e2dfedaa4c895923dae5 Mon Sep 17 00:00:00 2001 From: Daniel Darabos Date: Mon, 1 Sep 2014 22:14:28 -0700 Subject: [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 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 --- ec2/spark_ec2.py | 12 +++++++++++- 1 file changed, 11 insertions(+), 1 deletion(-) (limited to 'ec2') 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 -- cgit v1.2.3