slurm.yaml

heat_template_version: 2018-08-31

description: Template that installs a cluster of slurm servers.

parameters:
  base_name:
    type: string
    description: Cluster base name (lowercase)
    constraints:
      - allowed_pattern: "[a-z0-9]*"
    default: debat

  node_count:
    description: Number of slurm nodes
    type: number
    default: 3

  key_name:
    type: string
    description: Name of key-pair to be used
    constraints:
      - custom_constraint: nova.keypair

  flavor_master:
   type: string
   default: 202 #c1.2xlarge
   description: |
     Choose an instance flavor for master
   constraints:
     - custom_constraint: nova.flavor

  flavor_node:
   type: string
   default: 202 #c1.2xlarge
   description: |
     Choose an instance flavor for node
   constraints:
     - custom_constraint: nova.flavor

  image_id:
    type: string
    label: Server image
    description: Choose centOS 7
    default: f85e0272-8e27-43a4-be8c-9d997d067732
    constraints:
      - custom_constraint: glance.image
      #- allowed_pattern: "[-a-zA-Z_0-9]*cent[oO][sS][-a-zA-Z_0-9]*"

  image_default_login:
    type: string
    default: centos
    label: Image default login

  master_data_volume_size:
    type: number
    default: 1024
    label: Master data volume size GB

  master_data_volume_type:
    type: string
    default: "ceph-ssd"
    constraints:
      - allowed_values: ["ceph-ssd", "standard"]

  user1_name:
    type: string
    label: user1 name
    description: User1 account name
    default: ""
    constraints:
      - allowed_pattern: "[a-z]*"

  user1_key:
    type: string
    label: user1 ssh key
    description: User1 ssh public key
    default: ""

  user2_name:
    type: string
    label: user2 name
    description: User2 account name
    default: ""
    constraints:
      - allowed_pattern: "[a-z]*"

  user2_key:
    type: string
    label: user2 ssh key
    description: User2 ssh public key
    default: ""

  user3_name:
    type: string
    label: user3 name
    description: User3 account name
    default: ""
    constraints:
      - allowed_pattern: "[a-z]*"

  user3_key:
    type: string
    label: user3 ssh key
    description: User3 ssh public key
    default: ""

  user4_name:
    type: string
    label: user4 name
    description: User4 account name
    default: ""
    constraints:
      - allowed_pattern: "[a-z]*"

  user4_key:
    type: string
    label: user4 ssh key
    description: User4 ssh public key
    default: ""

  user5_name:
    type: string
    label: user5 name
    description: User5 account name
    default: ""
    constraints:
      - allowed_pattern: "[a-z]*"

  user5_key:
    type: string
    label: user5 ssh key
    description: User5 ssh public key
    default: ""


resources:

  wait_condition:
    type: OS::Heat::WaitCondition
    properties:
      handle: { get_resource: wait_handle }
      count: { get_param: node_count }
      timeout: 300

  wait_handle:
    type: OS::Heat::WaitConditionHandle

  triannot_key:
    properties:
      name:
        str_replace:
          template:
            $name-key-01
          params:
            $name: { get_param: base_name }
      save_private_key: true
    type: OS::Nova::KeyPair

  master_data_volume:
    type: OS::Cinder::Volume
    properties:
      size: { get_param: master_data_volume_size }
      volume_type: { get_param: master_data_volume_type }

  triannot_net:
    type: OS::Neutron::Net
    properties:
      name:
        str_replace:
          template:
            $name-network-public-01
          params:
            $name: { get_param: base_name }

  triannot_subnet:
    type: OS::Neutron::Subnet
    properties:
      name:
        str_replace:
          template:
            $name-subnet-public-01
          params:
            $name: { get_param: base_name }
      network_id: { get_resource: triannot_net }
      cidr: 10.45.8.0/24
      gateway_ip: 10.45.8.1
      allocation_pools:
        - {end: 10.45.8.54,start: 10.45.8.2}
      enable_dhcp: true

  triannot_router:
    type: OS::Neutron::Router
    properties:
      name:
        str_replace:
          template:
            $name-router-public-01
          params:
            $name: { get_param: base_name }
      external_gateway_info:
        network: provider-public-uca1

  triannot_router_interface:
    type: OS::Neutron::RouterInterface
    properties:
      router_id: { get_resource: triannot_router }
      subnet_id: { get_resource: triannot_subnet }

  triannot_master_security_group:
    type: OS::Neutron::SecurityGroup
    properties:
      name:
        str_replace:
          template:
            $name-master-security-group-01
          params:
            $name: { get_param: base_name }
      rules:
        - protocol: tcp
          remote_ip_prefix: 0.0.0.0/0
          port_range_min: 22
          port_range_max: 22
        - protocol: tcp
          remote_ip_prefix: 0.0.0.0/0
          port_range_min: 80
          port_range_max: 80

  triannot_node_security_group:
    type: OS::Neutron::SecurityGroup
    properties:
      name:
        str_replace:
          template:
            $name-node-security-group-01
          params:
            $name: { get_param: base_name }
      rules:
        - protocol: tcp
          remote_ip_prefix: 0.0.0.0/0
          port_range_min: 1
          port_range_max: 65535
        - protocol: udp
          remote_ip_prefix: 0.0.0.0/0
          port_range_min: 1
          port_range_max: 65535

  triannot_master_port:
    type: OS::Neutron::Port
    properties:
      network_id: { get_resource: triannot_net }
      security_groups:
        - default
        - { get_resource: triannot_master_security_group }

  triannot_node_port:
    type: OS::Heat::ResourceGroup
    properties:
      count:  {get_param: node_count}
      resource_def:
        type: OS::Neutron::Port
        properties:
          network_id: { get_resource: triannot_net }
          security_groups:
            - default
            - { get_resource: triannot_node_security_group }
          name:
            str_replace:
                  template:
                    $name$index$port
                  params:
                    $name: triannot_node
                    $index: "%index%"
                    $port: _port

  triannot_cluster:
    type: OS::Heat::ResourceGroup
    depends_on: triannot_node_port
    properties:
      count: { get_param: node_count}
      resource_def:
        type: OS::Nova::Server
        properties:
          image: { get_param: image_id }
          flavor: { get_param: flavor_node }
          key_name: { get_param: key_name }
          networks:
            - port:
                str_replace:
                  template:
                    $name$index$port
                  params:
                    $name: triannot_node
                    $index: "%index%"
                    $port: _port
          user_data_format: RAW
          user_data:
            str_replace:
              params:
                wc_notify: { get_attr: ['wait_handle', 'curl_cli'] }
                __public_key__: { get_attr: [ triannot_key, public_key ]  }
                __private_key__: { get_attr: [ triannot_key, private_key ] }
                __default_login__: { get_param: image_default_login}
                __index__: "%index%"
              template: |
                #!/bin/bash
                setenforce permissive #disable selinux

                DEFAULT_LOGIN_HOME=/home/"__default_login__"
                echo "__public_key__" >> /home/"__default_login__"/.ssh/authorized_keys
                echo "__private_key__" >> /home/"__default_login__"/.ssh/id_rsa
                node_vcpus=$(cat /proc/cpuinfo | grep -i "^processor" | wc -l)
                node_sockets=$(echo "$node_vcpus/2" | bc)
                node_ram_kb=$(cat /proc/meminfo | grep -i "^MemTotal" | awk -F" " '{print $2}')
                node_ram_mb_real=$(echo "$node_ram_kb/1024" | bc)
                node_ram_mb=$(echo "1000*($node_ram_mb_real/1024-1)" | bc)

                printf "node_ram_mb: %s\n" $node_ram_mb >> $DEFAULT_LOGIN_HOME/var.yml
                printf "node_sockets: %s\n" $node_sockets >> $DEFAULT_LOGIN_HOME/var.yml

                TIMEOUT=30    # number of iterations (5 minutes?)
                while :; do
                  status=$(sudo -u "__default_login__" ssh -o BatchMode=yes -o ConnectTimeout=5 -o StrictHostKeyChecking=no 127.0.0.1 -p 22 echo ok 2>&1)
                  RESULT=$?
                  if [ $RESULT -eq 0 ]; then
                      echo "__index__" >> test.txt
                      wc_notify --data-binary '{"status": "SUCCESS", "id": "__index__"}'
                      break
                  fi
                  if [ $RESULT -eq 255 ]; then
                      # connection refused also gets you here
                      if [[ $status == *"Permission"* ]] ; then
                          # permission denied indicates the ssh link is okay
                          break
                      fi
                  fi
                  TIMEOUT=$((TIMEOUT-1))
                  if [ $TIMEOUT -eq 0 ]; then
                      exit 1
                  fi
                  sleep 10s
                done

          name:
            str_replace:
              template:
                $name-$index
              params:
                $name: { get_param: base_name }
                $index: "compute%index%"

  triannot_master:
    type: OS::Nova::Server
    depends_on: triannot_cluster
    depends_on: triannot_master_port
    depends_on: wait_condition
    properties:
      image: { get_param: image_id }
      flavor: { get_param: flavor_master }
      key_name: { get_param: key_name }
      networks:
        - port: { get_resource: triannot_master_port }
      user_data_format: RAW
      user_data:
        str_replace:
          params:
            __private_key__: { get_attr: [ triannot_key, private_key ]  }
            __public_key__: { get_attr: [ triannot_key,  public_key] }
            __prefix__: { get_param: base_name }
            __count__: {get_param: node_count}
            __default_login__: { get_param: image_default_login}
            __user1_name__: { get_param: user1_name }
            __user1_key__: { get_param: user1_key }
            __user2_name__: { get_param: user2_name }
            __user2_key__: { get_param: user2_key }
            __user3_name__: { get_param: user3_name }
            __user3_key__: { get_param: user3_key }
            __user4_name__: { get_param: user4_name }
            __user4_key__: { get_param: user4_key }
            __user5_name__: { get_param: user5_name }
            __user5_key__: { get_param: user5_key }
          template: |
            #!/bin/bash

            setenforce permissive #disable selinux

            DEFAULT_LOGIN_HOME=/home/"__default_login__"
            touch $DEFAULT_LOGIN_HOME/logs.txt
            LOGS=$DEFAULT_LOGIN_HOME/logs.txt

            count="__count__"

            echo "__public_key__" >>$DEFAULT_LOGIN_HOME/.ssh/authorized_keys
            echo "__private_key__" >> $DEFAULT_LOGIN_HOME/.ssh/id_rsa
            chown "__default_login__":"__default_login__" $DEFAULT_LOGIN_HOME/.ssh/id_rsa
            chmod 0600 $DEFAULT_LOGIN_HOME/.ssh/id_rsa
            cd $DEFAULT_LOGIN_HOME

            if [ -b /dev/sdb ]; then
              mkfs.xfs /dev/sdb >> $LOGS
            fi

            echo "__test__" >> $LOGS
            echo "ansible git" >> $LOGS
            #git clone https://github.com/meso-uca/triannot.git #https://github.com/balbes15/ansible-triannot.git
            git clone -b oscar-slurm https://github.com/HeleneRimbert/oscar-triannot.git
            echo "ansible git done" >> $LOGS

            cd oscar-triannot/ansible
            chown -R centos:centos $DEFAULT_LOGIN_HOME/oscar-triannot
            pwd >> $LOGS
            rm ./group_vars/all/slurm.yml
            ls ./group_vars/all/ >> $LOGS

            echo '  - { login: "__user1_name__", key: "__user1_key__", uid: 5001 }' >> ./group_vars/all/users.yml
            echo '  - { login: "__user2_name__", key: "__user2_key__", uid: 5002 }' >> ./group_vars/all/users.yml
            echo '  - { login: "__user3_name__", key: "__user3_key__", uid: 5003 }' >> ./group_vars/all/users.yml
            echo '  - { login: "__user4_name__", key: "__user4_key__", uid: 5004 }' >> ./group_vars/all/users.yml
            echo '  - { login: "__user5_name__", key: "__user5_key__", uid: 5005 }' >> ./group_vars/all/users.yml

            cat <<EOF1 > hosts.ini
            [NODES:vars]
            ansible_ssh_common_args='-o StrictHostKeyChecking=no'

            [MASTER:vars]
            ansible_ssh_common_args='-o StrictHostKeyChecking=no'

            [MASTER]
            __prefix__-master ansible_connection=local

            [NODES]
            EOF1

            node_prefix=$(printf "%s-compute" "__prefix__")
            count_1=$(($count-1))
            for i in `seq 0 $count_1`
            do
              node_name=$(printf "%s%d" $node_prefix $i)
              echo "$node_name" >> hosts.ini
              if [ $i -eq 0 ]; then
                  sudo -u "__default_login__" scp -oStrictHostKeyChecking=no "__default_login__"@$node_name:$DEFAULT_LOGIN_HOME/var.yml ./group_vars/all/slurm.yml >> $LOGS
                  echo node_prefix: "__prefix__"-compute >> ./group_vars/all/slurm.yml
                  echo master_name: "__prefix__"-master >> ./group_vars/all/slurm.yml
                  echo "node_max: $count_1" >> ./group_vars/all/slurm.yml
              fi
            done

            export ANSIBLE_HOST_KEY_CHECKING=False
            sudo -u "__default_login__" ansible-playbook -i hosts.ini deploy.yml >> $LOGS
      name:
        str_replace:
          template:
            $name-master
          params:
            $name: { get_param: base_name }

  volume_att:
    type: OS::Cinder::VolumeAttachment
    properties:
      instance_uuid: { get_resource: triannot_master }
      volume_id: { get_resource: master_data_volume }
      mountpoint: /dev/sdb

  floating_ip:
    type: OS::Neutron::FloatingIP
    properties:
      floating_network: provider-public-uca1

  floating_ip_assoc:
    type: OS::Neutron::FloatingIPAssociation
    depends_on: triannot_master
    properties:
      floatingip_id: { get_resource: floating_ip }
      port_id: { get_resource: triannot_master_port }

outputs:
    public_ip:
      description: The public IP address of this slurm cluster.
      value: { get_attr: [triannot_master, addresses] }