7 Replies Latest reply on Jan 14, 2015 3:09 PM by Cristian Vasile

    Tableau distributed cluster environment

    Jeff Strauss

      Toby Erkson


      I decided to split off to a new discussion rather than add to the tabcmd discussion, as this is definitely worthy of its own topic.  Attached is our new server clustered architecture that we are headed down, it's modeled very similar to ALPO.  It's all already deployed (except for the load balancer which will come within the next month).  It's not yet on production, but will be soon.  There have been many lessons learned and as I think of them I can add onto this post.


      #1 lesson: In a single server environment, all scripts can live on the same server (because the primary and the worker are one in the same).  In a distributed environment, the backup script needs to live on the primary, and if you are leveraging local working files (i.e. via a Tableau export), then we have chosen to designate one of the workers to handle this (insights-direct).


      #2 lesson: Because we have a H.A. architecture, there is a passive read only data engine and a passive read only repository.  Take methodical steps in order to deploy H.A. because if you try to go to fast, then you end up getting burnt and you get errors trying to interact to the PG repository as the workers are not all whitelisted.  And then if you try to correct it on the fly, then you don't know if your deployment is corrupt or not, so layout a detailed plan and do a clean install.