diff --git a/ec2/spark-ec2 b/ec2/spark-ec2 index 31f9771223e51e4439fe1d752a6d75cd22f35065..4aa908242eeaa6fde925a834ff79a911fd072fc1 100755 --- a/ec2/spark-ec2 +++ b/ec2/spark-ec2 @@ -18,5 +18,9 @@ # limitations under the License. # -cd "`dirname $0`" -PYTHONPATH="./third_party/boto-2.4.1.zip/boto-2.4.1:$PYTHONPATH" python ./spark_ec2.py "$@" +# Preserve the user's CWD so that relative paths are passed correctly to +#+ the underlying Python script. +SPARK_EC2_DIR="$(dirname $0)" + +PYTHONPATH="${SPARK_EC2_DIR}/third_party/boto-2.4.1.zip/boto-2.4.1:$PYTHONPATH" \ + python "${SPARK_EC2_DIR}/spark_ec2.py" "$@" diff --git a/ec2/spark_ec2.py b/ec2/spark_ec2.py index 50f88f735650e37976471c5ec0ca7a2168739a0f..a5396c23759154dcb70bba0baee0fdb6283a14d0 100755 --- a/ec2/spark_ec2.py +++ b/ec2/spark_ec2.py @@ -40,6 +40,7 @@ from boto.ec2.blockdevicemapping import BlockDeviceMapping, BlockDeviceType, EBS from boto import ec2 DEFAULT_SPARK_VERSION = "1.1.0" +SPARK_EC2_DIR = os.path.dirname(os.path.realpath(__file__)) MESOS_SPARK_EC2_BRANCH = "v4" # A URL prefix from which to fetch AMI information @@ -593,7 +594,14 @@ def setup_cluster(conn, master_nodes, slave_nodes, opts, deploy_ssh_key): ) print "Deploying files to master..." - deploy_files(conn, "deploy.generic", opts, master_nodes, slave_nodes, modules) + deploy_files( + conn=conn, + root_dir=SPARK_EC2_DIR + "/" + "deploy.generic", + opts=opts, + master_nodes=master_nodes, + slave_nodes=slave_nodes, + modules=modules + ) print "Running setup on master..." setup_spark_cluster(master, opts) @@ -730,6 +738,8 @@ def get_num_disks(instance_type): # cluster (e.g. lists of masters and slaves). Files are only deployed to # the first master instance in the cluster, and we expect the setup # script to be run on that instance to copy them to other nodes. +# +# root_dir should be an absolute path to the directory with the files we want to deploy. def deploy_files(conn, root_dir, opts, master_nodes, slave_nodes, modules): active_master = master_nodes[0].public_dns_name