Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Configurable workers #46

Merged
merged 4 commits into from
May 3, 2014
Merged

Configurable workers #46

Show file tree
Hide file tree
Changes from 3 commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
8e1f6d8
Added SPARK_WORKER_INSTANCES and SPARK_WORKER_CORES support
Apr 14, 2014
c8e3f22
Added SPARK_MASTER_OPTS configuration
May 1, 2014
149c767
Added spark_master_opts to README
May 1, 2014
65f0cfe
Avoid workers with no cores when over-subscribing workers
May 3, 2014
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
3 changes: 3 additions & 0 deletions README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -34,6 +34,9 @@ and can be used to install any pre-requisites.
{{mapred_local_dirs}}
{{spark_local_dirs}}
{{default_spark_mem}}
{{spark_worker_instances}}
{{spark_worker_cores}}
{{spark_master_opts}}

You can add new variables by modifying `deploy_templates.py`

Expand Down
14 changes: 14 additions & 0 deletions deploy_templates.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -12,6 +12,7 @@

# Find system memory in KB and compute Spark's default limit from that
mem_command = "cat /proc/meminfo | grep MemTotal | awk '{print $2}'"
cpu_command = "nproc"

master_ram_kb = int(
os.popen(mem_command).read().strip())
Expand All @@ -20,8 +21,14 @@

slave_mem_command = "ssh -t -o StrictHostKeyChecking=no %s %s" %\
(first_slave, mem_command)

slave_cpu_command = "ssh -t -o StrictHostKeyChecking=no %s %s" %\
(first_slave, cpu_command)

slave_ram_kb = int(os.popen(slave_mem_command).read().strip())

slave_cpus = int(os.popen(slave_cpu_command).read().strip())

system_ram_kb = min(slave_ram_kb, master_ram_kb)

system_ram_mb = system_ram_kb / 1024
Expand All @@ -42,6 +49,10 @@
# Make tachyon_mb as spark_mb for now.
tachyon_mb = spark_mb

worker_instances = int(os.getenv("SPARK_WORKER_INSTANCES", 1))
# Distribute equally cpu cores among worker instances
worker_cores = slave_cpus / worker_instances
Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It might be good to do:

max(slave_cpus / worker_instances, 1)

I some divergent cases people might try to over-subscribe the machines. It would be good to have Spark at least give 1 core to each worker in these cases.


template_vars = {
"master_list": os.getenv("MASTERS"),
"active_master": os.getenv("MASTERS").split("\n")[0],
Expand All @@ -50,6 +61,9 @@
"mapred_local_dirs": os.getenv("MAPRED_LOCAL_DIRS"),
"spark_local_dirs": os.getenv("SPARK_LOCAL_DIRS"),
"default_spark_mem": "%dm" % spark_mb,
"spark_worker_instances": "%d" % worker_instances,
"spark_worker_cores": "%d" % worker_cores,
"spark_master_opts": os.getenv("SPARK_MASTER_OPTS"),
"spark_version": os.getenv("SPARK_VERSION"),
"shark_version": os.getenv("SHARK_VERSION"),
"hadoop_major_version": os.getenv("HADOOP_MAJOR_VERSION"),
Expand Down
5 changes: 5 additions & 0 deletions templates/root/spark/conf/spark-env.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -20,12 +20,17 @@ export SPARK_MEM={{default_spark_mem}}
SPARK_JAVA_OPTS+=" -Dspark.local.dir={{spark_local_dirs}}"
export SPARK_JAVA_OPTS

export SPARK_MASTER_OPTS="{{spark_master_opts}}"

export HADOOP_HOME="/root/ephemeral-hdfs"
export SPARK_LIBRARY_PATH="/root/ephemeral-hdfs/lib/native/"
export SPARK_MASTER_IP={{active_master}}
export MASTER=`cat /root/spark-ec2/cluster-url`
export SPARK_CLASSPATH=$SPARK_CLASSPATH":/root/ephemeral-hdfs/conf"

export SPARK_WORKER_INSTANCES={{spark_worker_instances}}
export SPARK_WORKER_CORES={{spark_worker_cores}}

# Bind Spark's web UIs to this machine's public EC2 hostname:
export SPARK_PUBLIC_DNS=`wget -q -O - http://169.254.169.254/latest/meta-data/public-hostname`

Expand Down