path: root/vcluster

#!/bin/bash

print_help() {
    cat <<EOF >&2
Usage:

$0 init <spark_source>
$0 start <n>
$0 stop
$0 shell
$0 list

EOF
}

fail() {
    echo "$1" >&2
    exit 1
}

init() {
    spark_source="$1"
    stage_dir=$(mktemp -d)

    [ -n "$spark_source" ] || fail $(print_help)
    [ -e "$spark_source" ] || fail "No such file or directory: $spark_source"

    # build spark
    (cd "$spark_source" && ./dev/make-distribution.sh --name custom-spark -Phadoop-2.7)
    cp -r "$spark_source/dist" "$stage_dir"

    # prepare common docker image
    cat <<EOF > "$stage_dir/Dockerfile.spark"
FROM debian:jessie-backports

MAINTAINER Jakob Odersky <jakob@odersky.com>

# install base utilities
RUN \
    apt-get update && \
    apt-get install -y \
        wget \
        curl \
        git \
	openssl \
	ssh \
	openjdk-8-jdk \
	python \
	apt-transport-https \
	nano && \
    apt-get clean && \
    rm -rf /var/lib/apt/lists/*

COPY "$stage_dir"/dist /opt/spark

RUN adduser --system --group --home /run/spark spark

ENV SPARK_LOG_DIR /run/spark
ENV SPARK_NO_DAEMONIZE true

EXPOSE 6066
EXPOSE 7077
EXPOSE 8080

USER spark
WORKDIR /run/spark

EOF
    docker build -f "$stage_dir/Dockerfile.spark" -t jodersky/spark:latest "$stage_dir"

    # prepare master docker image
    cat <<EOF > "$stage_dir/Dockerfile.master"
FROM jodersky/spark:latest
MAINTAINER Jakob Odersky <jakob@odersky.com>
ENTRYPOINT ["/opt/spark/sbin/start-master.sh"]

EOF
    docker build -f "$stage_dir/Dockerfile.master" -t jodersky/spark-master:latest "$stage_dir"

    # prepare worker docker image
    cat <<EOF > "$stage_dir/Dockerfile.worker"
FROM jodersky/spark:latest
MAINTAINER Jakob Odersky <jakob@odersky.com>
ENV SPARK_WORKER_DIR /run/spark
ENTRYPOINT ["/opt/spark/sbin/start-slave.sh"]

EOF
    docker build -f "$stage_dir/Dockerfile.worker" -t jodersky/spark-worker:latest "$stage_dir"
}


while [ $# -gt 0 ]; do
    case "$1" in
	init)
	    shift
	    init "$1"
	    ;;
	
	start)
	    shift
	    workers="${1:-5}"
	    docker run \
		   --label spark=master \
		   --detach \
		   jodersky/spark-master

	    cores_total=$(grep -c ^processor /proc/cpuinfo)
	    cores_worker=$(( (cores_total + workers - 1) / workers)) # round up

	    mem_total=$(cat /proc/meminfo | grep MemTotal | awk '{print $2}') # in kb
	    mem_worker=$((mem_total / workers))

	    for i in $(seq 1 "$workers"); do
		docker run \
		       --label spark=worker \
		       --detach \
		       jodersky/spark-worker \
		       --cores "$cores_worker" \
		       --memory "$mem_worker"k \
		       spark://172.17.0.2:7077 
	    done
	    ;;

	shell)
	    docker run \
		   --label spark=shell \
		   --interactive \
		   --tty \
		   jodersky/spark /opt/spark/bin/spark-shell \
		   --master=spark://172.17.0.2:7077
	    ;;
	
	stop)
	    containers=$(docker ps -q --format="{{.ID}}" --filter label=spark)
	    if [[ -n "$containers" ]]; then
		docker kill $containers
		docker rm $containers
	    fi
	    ;;

	list)
	    docker ps -q --format="{{.ID}} {{.Labels}} {{.Command}}" --filter label=spark --no-trunc
	    ;;
	
	help|-h|--help|*)
	    print_help
	;;
    esac
    shift
done