vagrant-mesos

Spin up your Mesos Cluster with Vagrant!

This also spins up Marathon and installs mesos-docker executor to each mesos-slave. This means you can build your own private PaaS with vagrant up!!

• Using VirtualBox
  • Mesos Standalone on VirtualBox
  • Mesos Cluster on VirtualBox
• Using Amazon EC2
  • Mesos Standalone on EC2
  • Mesos Cluster on EC2 (VPC)

The mesos installation is powered by mesos chef cookbook. Please see everpeace/cookbook-mesos.

Prerequisites

• vagrant 1.2+ and <1.5: http://www.vagrantup.com/
  • vagrant-berkshelf plugin is not compatible to vagrant 1.5. see the issue
• VirtualBox: https://www.virtualbox.org/ (not required if you use ec2.)
• vagrant plugins
  • vagrant-omnibus $ vagrant plugin install vagrant-omnibus
  • vagrant-berkshelf $ vagrant plugin install vagrant-berkshelf
  • vagrant-hosts $ vagrant plugin install vagrant-hosts
  • vagrant-cachier $ vagrant plugin install vagrant-cachier
  • vagrant-aws (only if you use ec2.) $ vagrant plugin install vagrant-aws

Mesos Standalone on VirtualBox

It's so simple!

$ cd standalone
$ vagrant up

After box is up, you can see

• mesos web UI on: http://192.168.33.10:5050
• marathon web UI on: http://192.168.33.10:8080

if everything went well.

Mesos Standalone on EC2

1. set ec2 credentials and some configurations defined in standalone/aws_config.yml. You have to fill up EDIT_HERE parts. Security group you'll set must accept at least tcp port 22(SSH) and 5050(mesos-master web ui) from outside of ec2.

    # Please set AWS credentials
    access_key_id:  EDIT_HERE
    secret_access_key: EDIT_HERE
   
    # please choose one from
    # ["ap-northeast-1", "ap-southeast-1", "eu-west-1", "sa-east-1", "us-east-1",
    #  "us-west-1", "ap-southeast-2", "us-west-2"]
    region: us-east-1
   
    # array of security groups. e.g. ['sg*** ']
    security_groups: EDIT_HERE
   
    # see http://aws.amazon.com/ec2/instance-types/#selecting-instance-types
    # for other instance types and its specs.
    instance_type: m1.small
   
    keypair_name: EDIT_HERE
   
    ssh_private_key_path: EDIT_HERE
   

2. you can spin up mesos box on ec2 by the same way with the case of virtual box

    cd standalone
    vagrant up --provider=aws
   

   After box is up, you can see

   • mesos web UI on: http://#_public_dns_of_the_VM_#:5050
   • marathon web UI on: http://#_public_dns_of_the_VM_#:8080

   if everything went well.

   Tips: you can get public dns of the vm by:

   $ vagrant ssh -- 'echo http://`curl --silent http://169.254.169.254/latest/meta-data/public-hostname`:5050'
   http://ec2-54-193-24-154.us-west-1.compute.amazonaws.com:5050
   

Mesos Cluster on VirtualBox

Cluster Configuration

Cluster configuration is defined at multinodes/cluster.yml. You can edit the file to congigure cluster settings.

# Mesos cluster configurations
mesos_version: 0.15.0

# The numbers of servers
##############################
zk_n: 1          # hostname will be zk1, zk2, …
master_n: 1      # hostname will be master1,master2,…
slave_n : 1      # hostname will be slave1,slave2,…

# Memory and Cpus setting(only for virtualbox)
##########################################
zk_mem     : 256
zk_cpus    : 1
master_mem : 256
master_cpus: 1
slave_mem  : 512
slave_cpus : 2

# private ip bases
# When ec2, this should be matched with
# private addresses defined by subnet_id below.
################################################
zk_ipbase    : "172.31.0."
master_ipbase: "172.31.1."
slave_ipbase : "172.31.2."

Launch Cluster

This takes several minutes(10 to 20 min.). It's time to go grabbing some coffee.

$ cd multinodes
$ ./vagrant up

At default setting, after all the boxes are up, you can see

• mesos web UI on: http://172.31.1.11:5050
• marathon web UI on: http://172.31.3.11:8080

if everything went well.

Destroy Cluster

this operations all VM instances forming the cluster.

$ cd multinodes
$ ./vagrant destroy

Mesos Cluster on EC2 (VPC)

Because we assign private IP addreses to VM instances, this Vagrantfile requires Amazon VPC (you'll have to set subnet_id and security grooups both of which associates to the same VPC instance).

Note: Using default VPC is highly recommended. If you used non-default vpc, you should make sure that limit of the elastic ips is no less than (zk_n + master_n + slave_n). The limit default is 5.

Cluster Configuration

You have to configure some additional stuffs in multinodes/cluster.yml which are related to EC2. Please note that

• subnet_id should be a VPC subnet
• security_groups should be ones associated to the VPC instance.
  • security_groups should allow accesses to ports 22(SSH), 2181(zookeeper) and 5050--(mesos).

(cont.)
# EC2 Configurations
# please choose one region from
# ["ap-northeast-1", "ap-southeast-1", "eu-west-1", "sa-east-1",
#  "us-east-1", "us-west-1", "ap-southeast-2", "us-west-2"]
# NOTE: if you used non-default vpc, you should make sure that
#       limit of the elastic ips is no less than (zk_n + master_n + slave_n).
#       In EC2, the limit default is 5.
########################
access_key_id:  EDIT_HERE
secret_access_key: EDIT_HERE
default_vpc: true                  # default vpc or not.
subnet_id: EDIT_HERE               # VPC subnet id
security_groups: ["EDIT_HERE"]     # array of VPN security groups. e.g. ['sg*** ']
keypair_name: EDIT_HERE
ssh_private_key_path: EDIT_HERE
region: EDIT_HERE

# see http://aws.amazon.com/ec2/instance-types/#selecting-instance-types
zk_instance_type: m1.small
master_instance_type: m1.small
slave_instance_type: m1.small

Launch Cluster

After editting configuration is done, you can just hit regular command.

$ cd multinode
$ vagrant up --provider=aws --no-parallel

NOTE: --no-parallel is highly recommended because vagrant-berkshelf plugin is prone to failure in parallel provisioning.

After instances are all up, you can see

• mesos web UI on: http://#_public_dns_of_the_master_N_#:5050
• marathon web UI on: http://#_public_dns_of_marathon_#:8080

if everything went well.

Tips: you can get public dns of the vms by:

$ vagrant ssh master1 -- 'echo http://`curl --silent http://169.254.169.254/latest/meta-data/public-hostname`:5050'
http://ec2-54-193-24-154.us-west-1.compute.amazonaws.com:5050

CAUTION: due to MESOS-672, mesos web ui fails redirection to the present leader of mesos master. So, you would need to access each master web ui manually to find the leader. The issue will be fixed on 0.18.0 release.

If you wanted to make sure that the specific mastar(e.g. master1) could be an initial leader, you can cotrol the order of spinning up VMs like below.

$ cd multinode
# spin up an zookeeper ensemble
$ vagrant up --provider=aws /zk/

# spin up master1. master1 will be an initial leader
$ vagrant up --provider=aws master1

# spin up remained masters
$ vagrant up --provider=aws /master[2-9]/

# spin up slaves
$ vagrant up --provider=aws /slave/

# spin up marathon
$ vagrant up --provider=aws marathon

Stop your Cluster

$ cd multinodes
$ vagrant halt

Resume your Cluster

$ cd multinodes
$ vagrant reload --provision

Destroy your Cluster

this operations terminates all VMs instances forming the cluster.

$ cd multinodes
$ vagrant destroy