dai_gpu_template.yaml

asc_template_version: 2016-05-19

description: |
    H2O Driverless AI

parameter_groups:
    - label: Conductor information
      description: Parameters that define Conductor REST configuration
      parameters:
        - conductor_ssl_key_file
        - conductor_ssl_enabled
        - ascd_port
    - label: Port information
      description: Parameters that define external system login information
      parameters:
        - dai_port
        - h2o_port
        - procsy_port
        - vis_server_port
    - label: SSL information
      description: Paramers for configuring SSL for Driverless AI
      parameters:
        - ssl_enabled
        - ssl_key_file
        - ssl_crt_file

parameters:
    conductor_ssl_key_file:
        type: string
        label: Location of Conductor SSL key file
        description: Location of Conductor SSL key file
        default: ${EGO_TOP}/wlp/usr/shared/resources/security/cacert.pem

    conductor_ssl_enabled:
        type: string
        label: Conductor SSL enabled
        description: Whether ssl for Conductor is enabled
        default: true

    ascd_port:
        type: string
        label: Port used by the ascd REST service
        description: Port used by the ascd REST service
        default: 8643

    dai_port:
        type: string
        label: Port used by the Driverless AI process
        description: Port used by the Driverless AI process
        default: 12345

    h2o_port:
        type: string
        label: Port used by the h2o process
        description: Port used by the h2o process
        default: 54321

    procsy_port:
        type: string
        label: Port used by the procsy process
        description: Port used by the procsy process
        default: 12347

    vis_server_port:
        type: string
        label: Port used by the vis server process
        description: Port used by the vis server process
        default: 12346

    dai_installdir:
        type: string
        label: Install Directory
        description: "Installation Directory for H2O Driverless AI, example: /var/dai/dai-1.3.1-linux-x86_64"

    data_dir:
        type: string
        label: Data Directory
        description: Directory that will store information regarding expermiments. Should be on a shared filesystem to support failover.

    log_dir:
        type: string
        label: Log Directory
        description: Directory that will store per-instance logs.

    dai_rg:
        type: string
        label: Resource Group for Driverless AI
        description: Resource Group for Driverless AI
        constraints:
            # this custom_constraint will tell the GUI to allow a selectable list of RG.
            - custom_constraint: asc.getresourcegroup

    dai_execuser:
        type: string
        label: Execution user
        description: Execution user for H2O Driverless AI

    ssl_enabled:
        type: string
        label: Enable SSL
        description: Whether to enable ssl for Driverless AI
        default: false

    ssl_key_file:
        type: string
        label: Location of SSL key file
        description: Location of SSL key file
        default: /etc/dai/private_key.pem

    ssl_crt_file:
        type: string
        label: Location of SSL crt file
        description: Location of SSL crt file
        default: /etc/dai/cert.pem

    additional_config:
        type: string
        label: Additional Driverless AI Configuration
        description: Specify additional Driverless AI Configuration options in the environment variable form. If specifying multiple semi colon separate them.
        default: none

resources:
    serviceDAI_activity:
        type: IBM::ASC::Activity
        properties:
            commands:
                start: #required
                    command: { list_join: [ "", [ { get_param: dai_installdir }, "/spectrum_conductor/start.sh" ] ] }
                stop:
                    command: { list_join: [ "", [ { get_param: dai_installdir }, "/kill-dai.sh" ] ] }
                monitor:
                    command: { list_join: [ "", [ { get_param: dai_installdir }, "/spectrum_conductor/jobMonitor.sh" ] ] }
            workingdirectory: { list_join: [ "", [ { get_param: dai_installdir }, "/spectrum_conductor" ] ] }
            log_path: { list_join: [ "", [ { get_param: log_dir }, "/*"] ] }
            environmentvariables: #optional
                INSTALL_DIR: { get_param: dai_installdir }
                EXEC_USER: { get_param: dai_execuser }
                DRIVERLESS_AI_DATA_DIRECTORY: { get_param: data_dir }
                DRIVERLESS_AI_PORT: { get_param: dai_port }
                DRIVERLESS_AI_H2O_PORT: { get_param: h2o_port }
                DRIVERLESS_AI_PROCSY_PORT: { get_param: procsy_port }
                DRIVERLESS_AI_VIS_SERVER_PORT: { get_param: vis_server_port }
                DRIVERLESS_AI_ENABLE_HTTPS: { get_param: ssl_enabled }
                DRIVERLESS_AI_SSL_KEY_FILE: { get_param: ssl_key_file }
                DRIVERLESS_AI_SSL_CRT_FILE: { get_param: ssl_crt_file }
                DRIVERLESS_AI_AUTHENTICATION_METHOD: "ibm_spectrum_conductor"
                RUN_DAI_LOG_DIR: { get_param: log_dir }
                ADDITIONAL_CONFIG: { get_param: additional_config }
                APP_UUID: { get_attr: [ _appinstance_, uuid ] }
                ASCD_HTTPS_ENABLED: { get_param: conductor_ssl_enabled }
                ASCD_REST_CACERT_PATH: { get_param: conductor_ssl_key_file }
                ASCD_REST_LOCATION: ${SERVICE_ascd_LOCATION}
                ASCD_REST_PORT: { get_param: ascd_port }

    serviceDAI: #AppName will be prepended for uniqueness
        type: IBM::ASC::Service
        properties:
            description: DAI Service
            numinstances:
                min: 1 # required, can also be a %
                max: 1  # required, can also be a %
                instancetoslotratio: "1:3"
            usegpu: true
            exclusivegpu: false
            resourcegroup:  { get_param: dai_rg } # required
            resourcerequirement: select('LINUXPPC64'||'LINUXPPC64LE'||'LINUX86'||'X86_64'||'IA64'||'NTX86'||'NTX64'||'SOLX8664'||'SOL64') # optional
            activity: # at least one must be specified. Specify more than one to support multiple OS types
                - { get_resource: serviceDAI_activity }
            consumer:
                name: DAItest
                executionuser: { get_param: dai_execuser }
            controlpolicy: # starttype will be manual for all ASC services
                maxrestarts: 10
                hostfailoverinterval_sec: 1
                blockedhostexitvalues: [-1, 1]
            dependency:
              - target: ascd
                satisfy:
                  - STARTED
                keep:
                  - DEFINED
                  - INIT
                  - ALLOCATING
                  - STARTED
                  - DEALLOCATING
                  - ERROR
                  - HANG
                  - FROZEN
                  - TENTATIVE
                autostart: false

outputs:
    dai_url:
        description: H2O Driverless AI URL from job Monitor
        value: { get_attr: [serviceDAI, userinfo, 1] } # the last number is sequence number