Tableau Server Warning: “The cluster has inconsistent processor core counts.”

Version 1

    Hi All,

    The warning message I got doesn't seem to be faced very often. But it was not very straight forward for me to get rid of it.

    I hope this document will be helpful to someone.

    I have faced the warning  about a year back while installing Tableau server on  a new hardware and second time when adding new workers to a cluster and it  took a while to recollect what I had done a year back to resolve it.   This time I documented the steps


    Where can it happen?  

    OS- Windows 2012.

    Hardware - HP ProLiant Gen9 Servers - Systems Configured with Intel Xeon E5 2600v3

    When it happens? 

    When you install Tableau Server software on a new server or add new worker nodes to the cluster.

    When you change license type from user based to core based.

    What is the message?

    You start the server after new installation and see the message:

    ===== Starting service...

      *** The cluster is above its license capacity, some workers may be unlicensed. Please run 'tabadmin licenses' for details.

      *** The cluster has inconsistent processor core counts, some workers may be unlicensed. Please run 'tabadmin licenses' for details.

       -- Service was started successfully



    I know that our license is fine.

    You run tabadmin licenses and see:

    d:\Tableau Server\9.1\bin>tabadmin licenses

    Cores used: 16 of 16

    Licenses in use:

    Myserver1: Cores: 8, Cores licensed: 0

    Myserver2: Cores: 8, Cores licensed: 8  Warning: Inconsistent core count.

    Myserver3: Cores: 8, Cores licensed: 8  Warning: Inconsistent core count.

    Inconsistent core counts may reduce the effective capacity of your cluster. This can occur on older versions of Windows, but can be corrected with the appropriate hotfixes below:

    Windows Server 2003 --

    Windows XP --


    My OS is 2012….  It doean't say anything about Windows Server 2012.

    I did some research to find the solution following suggested links for Server 2003 and XP, no luck. I have contacted Tableau Support ans they shared a link for HP document , but I felt document was missing some details. So I wanted to create this document to provide all details, for those who are like me and don't know how to solve it.




    On HP ProLiant Gen9 servers, the System Firmware, by default, reports the maximum number of possible logical processors to reflect the highest core count processor available. The result is that Windows Operating Systems will group the available logical processors into more than one Kgroup in a two-socket populated configuration.

    Applications that rely on older kernel APIs for affnitizing threads inside a process to a specific logical processor will only be able to address one Kgroup. The result is that the process will run on fewer logical processors than are presented to the OS for use. Workloads that utilize multiple processes (in lieu of a single process with multiple threads), or are leveraging the APIs that allow beyond single Kgroup addressing, will have full access to all the logical processors presented to the OS for use.

    And Tableau is presented to see all possible number of processors.


    Change the behavior of reporting processors to the Operating System, using a new ROM-Based Setup Utility, "NUMA Group Size Optimization."

    1. Login to server HP remote server management  processor - ILO  or if you have physical access to the server skip to step 3
    2. Select Remote console and launch it selecting on of the presented options.
    3. Restart the server
    4. During the restart, press F9 on keyboard to select System Utilities

    5. On the next screen select “System Configuration”, using arrow keys on your keyboard

    6. Select “BIOS/Platform Configuration (RBSU)”

    7. “Select Performance Options”

    8. Select  “Advanced Performance Tuning Options”

    9. Select “NUMA Group Size Optimization”

    10. Change the reporting mode to  “Flat”

    11. Save changed and continue server restart.