Dela via


Running HPC SOA Services in Hybrid HPC Pack Cluster

This documentation describes how to deploy and run HPC SOA services in a hybrid HPC Pack cluster which contains both on-premises compute nodes and Azure IaaS compute nodes.

Pre-requisite

  • An Azure Subscription.

    • To create an Azure subscription, go to the Azure site.
    • To access an existing subscription, go to the Azure portal.
  • A Site-to-Site VPN or ExpressRoute connection from your on-premises network to the Azure virtual network.

Deploy a Hybrid HPC Pack Cluster

To deploy a hybrid HPC Pack Cluster, you can leverage IaaS Burst feature from HPC Pack 2016 Update 1. See Burst to Azure IaaS VM from an HPC Pack Cluster for detail information.

Deploy HPC SOA services in a Hybrid HPC Pack Cluster

IaaS compute nodes are domain joined

If the IaaS compute nodes are domain joined, the cluster shall behave as an on-premises HPC Pack Cluster. See our tutorial series for more information on HPC SOA.

IaaS compute nodes are non-domain joined

If the IaaS compute nodes are non-domain joined compute nodes, the way to deploy a new HPC SOA service is slightly different.

Deploy a new HPC SOA service

  1. Deploy service binaries

    When deploying a new HPC SOA service to a hybrid HPC Pack Cluster which contains non-domain joined IaaS compute nodes, the service binaries need to be placed in every compute nodes' local folder. (e.g. C:\Services\Sample.dll)

  2. Change settings in service registration file

    • Change the assembly path to the local file on every compute node

      <microsoft.Hpc.Session.ServiceRegistration>
          <!--Change assembly path below-->
          <service assembly="C:\Services\Sample.dll" />    
      </microsoft.Hpc.Session.ServiceRegistration>
      
    • As domain identity is not available on non-domain joined compute nodes, we need to set security mode between broker node and compute nodes to None. Add following section into service registration file or change existing one:

      <system.serviceModel>
          <bindings>
              <netTcpBinding>
                  <!--binding used by broker's backend-->
                  <binding name="Microsoft.Hpc.BackEndBinding" maxConnections="1000">
                      <!--for non domain joined compute nodes, the security mode should be None-->
                      <security mode="None"/>
                  </binding>
              </netTcpBinding>
          </bindings>
      </system.serviceModel>
      
  3. Deploy service registration file

    There are two ways to deploy the service registration file of new SOA service

    • Deploy the service registration file to every cluster nodes' local folder

      To deploy service registration file using this way, we need to create a local service registration folder on every node (e.g. C:\ServiceRegistration), and add it to HPC Pack cluster's CCP_SERVICEREGISTRATION_PATH environment.

      The cluster environment can be fetched on head node using command cluscfg listenvs and will give output like:

      C:\Windows\system32>cluscfg listenvs
      CCP_SERVICEREGISTRATION_PATH=CCP_REGISTRATION_STORE;\\<headnode>\HpcServiceRegistration
      HPC_RUNTIMESHARE=\\<headnode>\Runtime$
      CCP_MPI_NETMASK=10.0.0.0/255.255.252.0
      CCP_CLUSTER_NAME=<headnode>
      

      To configure the environment, use command cluscfg setenvs like

      cluscfg setenvs CCP_SERVICEREGISTRATION_PATH=CCP_REGISTRATION_STORE;\\<headnode>\HpcServiceRegistration;C:\ServiceRegistration
      

      Then copy the service registration file to newly created local folder (e.g. C:\ServiceRegistration\Sample.config)

      More about cluscfg can be found in the cluscfg documentation.

      More about creating and deploying a new SOA service can be found in HPC Pack SOA Tutorial I – Write your first SOA service and client.

    • Deploy the service registration file using HA configuration file feature

      From HPC Pack Update 1, we added HA configuration file feature to help deploying service registration files in HA clusters and in clusters with non-domain joined compute nodes.

      To use this feature, please ensure CCP_REGISTRATION_STORE is set in CCP_SERVICEREGISTRATION_PATH cluster environment. Check the result of cluscfg listenvs, it should contains

      CCP_SERVICEREGISTRATION_PATH=CCP_REGISTRATION_STORE;\\<headnode>\HpcServiceRegistration
      

      If the value CCP_REGISTRATION_STORE is missing, please set it use command cluscfg setenvs like

      cluscfg setenvs CCP_SERVICEREGISTRATION_PATH=CCP_REGISTRATION_STORE;\\<headnode>\HpcServiceRegistration
      

      To add a HA configuration file, in HPC Pack 2016 Cluster Manager, go to Configuration -> Services panel, click Import High Available Configuration File... in Action panel. Then select the HA configuration file to import.

      Screenshot shows the Configuration page with Services selected and the Import High Available Configuration file highlighted.

      Once imported, a new service entry will show in Services panel with High Available set to True.

      Screenshot shows a service entry with a value of True for High Available.

Using Common Data

From HPC Pack 2016 Update 2, non-domain joined compute nodes can read SOA Common Data from Azure Blob Storage directly without accessing to HPC Runtime share which usually located on the head node. This enables the SOA workloads in a hybrid HPC Pack Cluster with compute resource provisioned in Azure.

  • Configure HPC Pack Cluster

    To enable this feature, an Azure Storage Connection String need to be configured in HPC Cluster. In HPC Pack 2016 Cluster Manager, find Set Microsoft Azure storage connection string in Configuration -> Deployment To-do List, and fill in the correct connection string.

    Screenshot shows the Configuration Deployment To do list with Set Microsoft Azure storage connection string highlighted.

  • Change the SOA service code

    If your service calls ServiceContext.GetDataClient to obtain a reference of DataClient before, no code change is needed.

    Instead if your service explicitly uses DataClient.Open, you'll need to change the call to a new API if your service is running on a non-domain joined node. Here is the sample of it.

    // Get Job ID from environment variable
    string jobIdEnvVar = Environment.GetEnvironmentVariable(Microsoft.Hpc.Scheduler.Session.Internal.Constant.JobIDEnvVar);
    if (!int.TryParse(jobIdEnvVar, out int jobId))
    {
        throw new InvalidOperationException($"jobIdEnvVar is invalid:{jobIdEnvVar}");
    }
    // Get Job Secret from environment variable
    string jobSecretEnvVar = Environment.GetEnvironmentVariable(Microsoft.Hpc.Scheduler.Session.Internal.Constant.JobSecretEnvVar);
    return DataClient.Open(dataClientId, jobId, jobSecretEnvVar);
    

More about cluscfg can be found in the cluscfg documentation.

To read more about Azure Storage Connection Strings, see Configure Azure Storage connection strings.

To read more about HPC SOA Common Data, see HPC Pack SOA Tutorial IV – Common Data.

HPC Pack SOA Tutorials