Forum Discussion

BZM's avatar
BZM
Icon for Nimbostratus rankNimbostratus
Feb 28, 2014

Balancing Methods and Priority Groups

Here's a scenario: In a situation with a two members pool for which it'd be desirable to maintain all requests to a primary system, and only fall back to a secondary in the event that the primary goes down. The reason for this is the application does not share state and sending requests to one system is desired. This would mean, that once a switch over happens, the LTM should continue sending to the 'active' member.

 

My initial thought was to use the priority groups and assign those nodes appropriately. My concern with this is when the defined higher priority member recovers, it then is activated because of its priority, which it'd be desired to stick with the current active.

 

I have considered not using priority groups and thought about simply disabling one of the members, as well, but this means manual intervention for bringing the secondary member online which is not desirable. I'm also not seeing a balancing method that would indicate 'stick to one', since that's not really balancing. I'm not opposed to writing an iRule for a solution (i.e. change the priority level of a member once active, if possible), I just was thinking that someone may have came across this scenario before and their would be a system level solution.

 

Thanks for any ideas here.

 

4 Replies

  • You could use PGA combined with the manual resume option on the primary node health monitor, but you'd be in a bad spot if the second node failed hard after going active.

     

    Chris

     

    • BZM's avatar
      BZM
      Icon for Nimbostratus rankNimbostratus
      I appreciate the feedback on that and understand the potential problem there. I'll check out the option for manual resume, as well.
    • Christopher_Boo's avatar
      Christopher_Boo
      Icon for Cirrostratus rankCirrostratus
      If you have a good monitor and alerting, and the app is stable, I'm sure manual resume will probably be sufficient. You could even add a 3rd node in another PG if there is any concern. Chris
  • you can use priority groups and i recommend coupling that with a TCP Manual resume health-check.,... if you don't want to use manual resume then you need to institute Universal persistence which will keep the sessions on the secondary node even if the primary node comes back too service. I still recommend using manual resume if you can