Apache Bigtop is managing its own setup of CI system. We have 1 master + 4 slaves total. There's 3 * s one 1T storages storage assigned to master, and another 2 * 1T 800G go to two of the slaves (06 and 07), respectively. Other two nodes (02 and 03) got 40G got 200G for each node to operate therefore they are reserved to run non storage sensitive jobs such as provisioner.

Table of Contents

Setup a Jenkins master

...

If you already have a managed Jenkins/Hudson, or you prefer your all installation, skip and go to Setup Bigtop packages build matrix
...
And the needed Plugin(s):
git plugin

Setup

...

Jenkins slaves

Assuming that you are going to add or replace an Amazon EC2 instance which have already been created with Amazon Linux 2 AMI, execute the following commands in that node:

No Format

// Install Docker and Docker Compose (execute as ec2-user)
$ sudo yum update -y
$ sudo yum install -y docker git java ruby
$ sudo amazon-linux-extras install -y docker  # to make sure the latest version of docker is installed
$ sudo service docker start
$ sudo curl -L "https://github.com/docker/compose/releases/download/1.27.4/docker-compose-$(uname -s)-$(uname -m)" -o /usr/local/bin/docker-compose
$ sudo chmod +x /usr/local/bin/docker-compose

// Create jenkins user and setup SSH auth so that Jenkins master can connect to this node
$ sudo adduser jenkins
$ sudo usermod -aG docker jenkins
$ sudo -i
# su - jenkins
$ mkdir .ssh
$ cat << EOF > .ssh/authorized_keys
> (MASTER_PUBLIC_KEY_HERE)
> EOF
$ chmod 700 .ssh
$ chmod 400 .ssh/authorized_keys

// Make sure Docker and Docker Compose work for jenkins user, and login to
// Docker Hub so that Docker-related Jenkins jobs can push generated images
$ docker run hello-world
$ docker-compose version
$ docker login --username bigtop

Then add the node or update its configuration (e.g., FQDN) via Jenkins. In the latter case, you may have to disconnect and relaunch agent so that Jenkins recognizes the change.

NOTE: Some AMI (e.g., Amazon Linux 2) uses 1000 as a UID for the default user, but it conflicts our Docker image, which assigns Jenkins to UID 1000, and causes permission issue when building packages inside the container. In this case, reassign Jenkins to 1000 and the default user to another UID.

Setup Bigtop packages build matrix

Create a new job:

New Item -> Multi-configuration project

In Source Code Management section, fill in your repo and the branch, for example:

No Format
Repository URL: https://gitbox.apache.org/repos/asf/bigtop.git Branch Specifier: master

In the Configuration Matrix section, add a user defined axis with following name and values:

No Format
Name: BUILD_ENVIRONMENTS Values: centos-6 centos-7 fedora-25 ubuntu-16.04 debian-8 opensuse-42.1

Add another user defined axis:

No Format

Name: COMPONENTS

Values: 
bigtop-groovy
bigtop-jsvc
bigtop-tomcat
bigtop-utils
crunch
datafu
flume
giraph
hadoop
hama
hbase
hive
hue
ignite-hadoop
kafka
kite
mahout
oozie
phoenix
pig
solr
spark
sqoop
sqoop2
tachyon
tez
ycsb
zookeeper

Finally, add a shell build step with the following script:

No Format
docker run --rm -v `pwd`:/ws --workdir /ws bigtop/slaves:trunk-$BUILD_ENVIRONMENTS \ bash -l -c './gradlew allclean $COMPONENTS-pkg'

The result will be looked like this:

https://ci.bigtop.apache.org/view/Packages/job/Bigtop-trunk-packages/

However, do aware that full matrix build of packages is time consuming, and they need roughly 50GB disk space for each build.

Setup Bigtop deployments build matrix

To setup a deployments build matrix, prepare a job like above in packages build matrix and add these defined axis:

No Format
Name: OS Values: centos6 centos7 debian8 ubuntu_xenial

See the sample configs in bigtop/provisioner/docker.

No Format

Name: COMPONENTS
 
Values: 
alluxio
apex
crunch
flink
flume
giraph
ignite_hadoop
hbase
hcat
hive
httpfs
hue
mahout
mapred-app
oozie
pig
qfs
solrcloud
spark
sqoop
sqoop2
tez
yarn
zookeeper
ycsb
gpdb
ambari

See the available components for deployment in bigtop-deploy/puppet/hieradata/site.yaml.

Use the following shell script for builds:

No Format

# destroy previous cluster
./gradlew docker-provisioner-destroy
 
# setup configuration file
CONFIG="config_${OS}_Bigtop-trunk-deployments.yaml"
sed "s/components.*/components: [hdfs, yarn, mapreduce, ${COMPONENTS}]/g" \
  provisioner/docker/config_${OS}.yaml > provisioner/docker/${CONFIG}


# provision
# Since the puppet deployment will always return 0, 
# we need to save the log and grep error to determine status
./gradlew -Pconfig=${CONFIG} -Pnum_instances=1 docker-provisioner > tmp.log 2>&1
cat tmp.log


# destroy provisioned cluster
./gradlew docker-provisioner-destroy


# Fail the build if Errors occur in tmp.log
grep Error tmp.log && exit 1
exit 0

The result will be looked like this:

https://ci.bigtop.apache.org/view/Provisioner/job/Bigtop-1.2.1-deployments/

You can replace the repo by your own repo by adding sed commend so that your own packages can be tested using the same deployment recipes.

The components is also configurable, choose what you'd like to deploy and test as what you want.

Setup Bigtop smoke tests build matrix

Create a new job:

New Item -> Multi-configuration project

In Source Code Management section, fill in your repo and the branch, for example:

No Format
Repository URL: https://gitbox.apache.org/repos/asf/bigtop.git Branch Specifier: master

In the Configuration Matrix section, add axises with following name and values:

User-defined axis:

No Format

Name: COMPONENTS
 
Values: 
hdfs.alluxio@alluxio
ambari.bigtop-utils@ambari
hdfs.yarn.apex@apex
flume@flume
hdfs.hbase@hbase
hdfs.yarn@hdfs.hcfs
hdfs.yarn.hive@hive
hdfs.yarn.ignite_hadoop@ignite-hadoop
hdfs.yarn.mahout@mahout
hdfs.yarn@mapreduce
hdfs.yarn.qfs@qfs
hdfs.zookeeper.solrcloud@solr
hdfs.yarn.spark@spark
sqoop@sqoop
hdfs.yarn@yarn
zookeeper@zookeeper

For each value, the format is defined as DEPLOY_COMPONENTS@TEST_COMPONENETS, where DEPLOY_COMPONENTS cam be period separated string.

For example,

hdfs.alluxio@alluxio

Component hdfs and alluxio will be deployed, and alluxio will be tested.

Note: Why use period instead of comma as separator? Because Jenkins does not allow comma to be used in user defined axis.

Slaves aixs:

You got check boxes to check. To add options, go to Manage Jenkins → Manage Nodes and add labels for your slaves. The labels will be available here.

The following labels are bigtop's configuration for your reference.

No Format
Name: OS Values: centos-7-x86_64-deploy fedora-26-x86_64-deploy debian-9-amd64-deploy ubuntu-16.04-amd64-deploy debian-9-arm64-deploy ubuntu-16.04-arm64-deploy fedora-26-aarch64-deploy

In build section, add execute shell:

Code Block

language	bash

#!/bin/bash -x

# Setup configuration file
CONFIG="config_${OS}_Bigtop-smoke-tests.yaml"
DEPLOY_COMPONENTS=`echo ${COMPONENTS} | cut -d '@' -f 1 | sed "s/\\./,/g"`
TEST_COMPONENTS=`echo ${COMPONENTS} | cut -d '@' -f 2 | sed "s/\\./,/g"`
OS_FOR_URL=`echo ${OS} | awk -F"-" '{ print $1"-"$2"-"$3}' | sed "s@-@/@g"`

# Adjust mem based on test target
if [[ $DEPLOY_COMPONENTS == *"mahout"* ]]; then
  MEM=6g
fi

# The combination of DISTRO X ARCH is complicated. Resolved in the following logic
OS_WO_ARCH=`echo ${OS} | awk -F"-" '{ print $1"-"$2}'`
ARCH=`echo ${OS} | awk -F"-" '{ print $3}'`
case "${ARCH}" in
  aarch64|arm64)
    ARCH_SUFFIX="-aarch64"
    ;;
  ppc64le|ppc64el)
    ARCH_SUFFIX="-ppc64le"
    ;;
  x86_64|amd64)
    ARCH_SUFFIX=""
    ;;
  *)
    echo "Unsupported arch [${ARCH}]"
    exit 1
    ;;
esac 

# Determine distro for config.yaml
case "${OS}" in
  centos-*|fedora-*|opensuse-*)
    DISTRO=centos
    ;;
  debian-*|ubuntu-*)
    DISTRO=debian
    ;;
  *)
    echo "Unsupported distro [${OS}]"
    exit 1
    ;;
esac

cat > provisioner/docker/${CONFIG} <<-__EOT__
docker:
        memory_limit: "${MEM:-4g}"
        image: "bigtop/puppet:trunk-${OS_WO_ARCH}"

# Package CI is currently broken
# repo: https://ci.bigtop.apache.org/view/Packages/job/Bigtop-trunk-repos/OS=${OS},label=docker-slave/ws/output
repo: http://repos.bigtop.apache.org/releases/1.3.0/${OS_FOR_URL}
distro: ${DISTRO}
components: [${DEPLOY_COMPONENTS}]
enable_local_repo: false
smoke_test_components: [${TEST_COMPONENTS}]
__EOT__

cat

Create a new job:

New Item -> Multi-configuration project

In Source Code Management section, fill in your repo and the branch, for example:

No Format
Repository URL: https://gitbox.apache.org/repos/asf/bigtop.git Branch Specifier: master

In the Configuration Matrix section, add a user defined axis with following name and values:

No Format
Name: BUILD_ENVIRONMENTS Values: centos-6 centos-7 fedora-25 ubuntu-16.04 debian-8 opensuse-42.1

Add another user defined axis:

No Format

Name: COMPONENTS

Values: 
bigtop-groovy
bigtop-jsvc
bigtop-tomcat
bigtop-utils
crunch
datafu
flume
giraph
hadoop
hama
hbase
hive
hue
ignite-hadoop
kafka
kite
mahout
oozie
phoenix
pig
solr
spark
sqoop
sqoop2
tachyon
tez
ycsb
zookeeper

Finally, add a shell build step with the following script:

No Format
docker run --rm -v `pwd`:/ws --workdir /ws bigtop/slaves:trunk-$BUILD_ENVIRONMENTS \ bash -l -c './gradlew allclean $COMPONENTS-pkg'

The result will be looked like this:

https://ci.bigtop.apache.org/view/Packages/job/Bigtop-trunk-packages/

However, do aware that full matrix build of packages is time consuming, and they need roughly 50GB disk space for each build.

Setup Bigtop deployments build matrix

To setup a deployments build matrix, prepare a job like above in packages build matrix and add these defined axis:

No Format
Name: OS Values: centos6 centos7 debian8 ubuntu_xenial

See the sample configs in bigtop/provisioner/docker.

No Format

Name: COMPONENTS
 
Values: 
alluxio
apex
crunch
flink
flume
giraph
ignite_hadoop
hbase
hcat
hive
httpfs
hue
mahout
mapred-app
oozie
pig
qfs
solrcloud
spark
sqoop
sqoop2
tez
yarn
zookeeper
ycsb
gpdb
ambari

See the available components for deployment in bigtop-deploy/puppet/hieradata/site.yaml.

Use the following shell script for builds:

No Format

# destroy previous cluster
./gradlew docker-provisioner-destroy
 
# setup configuration file
CONFIG="config_${OS}_Bigtop-trunk-deployments.yaml"
sed "s/components.*/components: [hdfs, yarn, mapreduce, ${COMPONENTS}]/g" \
  provisioner/docker/config_${OS}.yaml > provisioner/docker/${CONFIG}


# provision
# Since the puppet deployment will always return 0, 
# we need to save the log and grep error to determine status
./gradlew -Pconfig=${CONFIG} -Pnum_instances=1 docker-provisioner > tmp.log 2>&1
cat tmp.log


# destroy provisioned cluster
./gradlew docker-provisioner-destroy


# Fail the build if Errors occur in tmp.log
grep Error tmp.log && exit 1
exit 0 Workaround that files written inside containers are own by root
docker run --rm -v ${PWD}:/bigtop-home "bigtop/puppet:trunk-${OS_WO_ARCH}${ARCH_SUFFIX}" bash -c "chown -R ${UID}:${UID} /bigtop-home"

# Provision
(cd provisioner/docker && ./docker-hadoop.sh -d -C ${CONFIG} -c 3 -s)

# Destroy provisinoed cluster
(cd provisioner/docker && ./docker-hadoop.sh -d)

# Workaround that files written inside containers are own by root
docker run --rm -v ${PWD}:/bigtop-home "bigtop/puppet:trunk-${OS_WO_ARCH}${ARCH_SUFFIX}" bash -c "chown -R ${UID}:${UID} /bigtop-home"

The result will be looked like this:

https://ci.bigtop.apache.org/view/ProvisionerTest/job/Bigtop-1.2.1-deployments/

You can replace the repo by your own repo by adding sed commend so that your own packages can be tested using the same deployment recipes.

...

-trunk-smoke-tests/

There're some limitations for the current implementation hence the script is a bit complicated.

Smoke tests running inside docker write files as root, hence files written by smoke tests can't be edited by Jenkins. Currently we work around this by spinning up a docker to chown files back to be owned by jenkins.
Provisioner should be able to decide which distro (centos or debian), however currently is externalized for user to configure it.

Advanced part: Setup a SSL secured Jenkins master

...

Most convient way to run it is as a docker container, so pull the image:

No Format
# docker pull quay.io/letsencrypt/letsencrypt:latest

...

This enables the reverse proxy mode of port 433 to port 8080 and setting Jenkins specific parameters.

And the location of the certificates have to be aligned with letsencrypt in /etc/http/conf.d/ssl.conf

...

The last thing is to change jenkins to port 8080 and start apache httpd. Please note, we are running the latest Jenkins LTS to keep up with the security updates, etc.

No Format
# docker run -d --name jenkins-master-2.6 -p 8080:8080 -v /home/jenkins:/var/jenkins_home jenkins/jenkins:lts # systemctl start httpd

...

No Format

# service httpd stop
# docker stop jenkins-master
# docker run --rm -i -t -p 80:80 -p 443:443  -v "/etc/letsencrypt:/etc/letsencrypt"  quay.io/letsencrypt/letsencrypt:latest  renewcertbot/certbot renew
# docker start jenkins-master
# service httpd start

Page tree

Versions Compared

Old Version 47

New Version Current

Key

Setup a Jenkins master

If you already have a managed Jenkins/Hudson, or you prefer your all installation, skip and go to Setup Bigtop packages build matrix
...
And the needed Plugin(s):
git plugin

Setup

Jenkins slaves

Setup Bigtop packages build matrix

Setup Bigtop deployments build matrix

Setup Bigtop smoke tests build matrix

Setup Bigtop deployments build matrix

Advanced part: Setup a SSL secured Jenkins master

Page tree

Page History

Versions Compared

Old Version 47

New Version Current

Key

Setup a Jenkins master

If you already have a managed Jenkins/Hudson, or you prefer your all installation, skip and go to Setup Bigtop packages build matrix...And the needed Plugin(s):git plugin

Setup

Jenkins slaves

Setup Bigtop packages build matrix

Setup Bigtop deployments build matrix

Setup Bigtop smoke tests build matrix

Setup Bigtop deployments build matrix

Advanced part: Setup a SSL secured Jenkins master

If you already have a managed Jenkins/Hudson, or you prefer your all installation, skip and go to Setup Bigtop packages build matrix
...
And the needed Plugin(s):
git plugin