IT Process Automation

  • 1.  Work assignment logic in a cluster

    Posted Jan 25, 2018 08:52 AM

    Can someone explain me the exact logic a PAM cluster uses to assign work?

     

    To test the cluster we use a process that calls a subprocess several times.

     

     

    Yesterday, after adding 3 nodes to the cluster, the work seemed to be assigned on a "round robin" basis against the 6 nodes.

     

    However, this morning, the work is assigned on a "round robin" basis against only 4 nodes.

     

     

    Is it always "round robin" no matter how occupied is a node?



  • 2.  Re: Work assignment logic in a cluster

    Posted Jan 25, 2018 08:58 AM

    You still using NetScaler Pier?

     

    And your nodes are all from domain orchestrator, right? (click on the first node administration and exec the java file)

     

    The work definition is defined on the load balancer, as simple as that.

    When I tried to use NetScaler  I saw that it was a simple round robin (just two nodes). But I believe there are somewhere of 13 of types of work distribution, as the network guys told me.



  • 3.  Re: Work assignment logic in a cluster

    Posted Jan 25, 2018 09:09 AM

    Yes we still use NetScaler.

     

    Yeah everything was installed by the first node GUI as a domain orchestrator node

     

    And during the time you was writing your answer, the 3 others nodes started to receive work.

     

     

    So basically the work is assigned using the netscaler? The node running the parent process sends a request to the netscaler to distribute a "run child process" operation and then the NetScaler decides "who" will run it?

     

    I thought it was an intern thing since it's a process calling a sub-process in a loop.

     

    I was thinking that the NetScaler was used to distribute inbound trafic.



  • 4.  Re: Work assignment logic in a cluster

    Posted Jan 25, 2018 09:22 AM

    No, there is no "master node" as there used to be. Netscaler is doing basically everything as it's the loadbalancer.



  • 5.  Re: Work assignment logic in a cluster

    Posted Jan 25, 2018 09:30 AM

    Thank you Marcel.



  • 6.  Re: Work assignment logic in a cluster

    Posted Jan 25, 2018 09:41 AM

    Also, check Pam’s log on some insights on how it controls the cluster and how it gives access to lockable things like datasets 



  • 7.  Re: Work assignment logic in a cluster
    Best Answer

    Broadcom Employee
    Posted Jan 25, 2018 09:42 AM
      |   view attached

    Marcel, The load balancer will control and direct incoming requests, once the request has reached one of the orchestartors and an instance is initiated, how and where parent and child instances run is controlled by internal Orchestrator logic.

     

    Pier, The logic on where an instance runs on an orchestrator cluster is not something simple like round robin, the logic takes into account the work load of the orchestrators and will try to spread instances out so that the overall 'load' on each node is relatively equal.   That can result in all nodes being utilized at one point, and in other cases, one orchestrator may handle all instances in a parent / child relationship.

     

    I have attached a simple process that support uses to test load balancer functionality.    There is a Parent process which kicks off a number of child instances.  Each child will be assigned to a cluster node to run and in the Operations tab will show the hostname of the server each instance ran.  

     

    Single runs may not show all nodes in your Orchestrator, but over the course of several runs you should see all of your cluster nodes.

     

     

    As for the 'MASTER' role, that role still exists and still holds specific functionality such as running the archive and purge component, but starting in 4.2 sp2 the MASTER role is now assumed by the first Orchestartor in the cluster that is started.  

    For example in a 2 node cluster, if Node 2 starts first it will assume the Master role, if Node 2 is shut down, Node 1 will assume the Master role.   

    As far as I am aware the Master role has no impact on instance distribution.

    Attachment(s)

    zip
    loadbalancing_test.zip   11 KB 1 version


  • 8.  Re: Work assignment logic in a cluster

    Posted Jan 25, 2018 01:22 PM

    Thanks Michael!



  • 9.  Re: Work assignment logic in a cluster

    Posted Jan 25, 2018 11:06 AM

    Thanks for the clarification Michael, I might need to open a support case about this.