Forum Discussion

Mohanad's avatar
Mohanad
Icon for Cirrostratus rankCirrostratus
Oct 01, 2019
Solved

F5 BIG-IP's in Different Datacenters

Hi everyone

 

if i implemented this setup:

 

 

 

 

What will happen if both ha links are down, each F5 device will think he is the active unit? if yes, how to mitigate this issue?

 

 

is there any solution like "witness" device to signals both devices and control the fail-over process?

 

 

 Thank you

Mohanad

application delivery
BIG-IP
  • Hamish's avatar
    Hamish
    Oct 09, 2019

    If all your links are down, then you go split brain. But if all your links are down, then you have more problems than just the split brain BigIP.

     

    (Note that the way I usually do this is an HA pair in each DC (Or metropolitan DC pair). Then use GTM to provide global resilience).

     

    Regarding fancy devices not requiring your network to be up to prevent split brain. Why bother? You can always put in a path that would encompass the same links as the failover clustering using witness. You're not limited to just two links.

11 Replies

Sort By
  • Greasy_Pretzel's avatar
    Greasy_Pretzel
    Ret. Employee
    Oct 09, 2019

    The link you have shared talks about ASM sync... https://devcentral.f5.com/s/articles/syncing-asm-waf-policies-between-f5-big-ips-in-different-datacenters-or-cloud-regions-32891

     

    f5 recommends; 1) Dedicated VLAN for HA failover; in failover IP address list, you have the self IP

    2) Add Management IP in the failover IP address list

     

    Review K16049 for more information on how to limit failover disruptions.

     

    With Load aware (or HA Order) configurations, if both links, HA and management are down, you will see A-A situation.

     

    However, you could test HA Group.

    • Mohanad's avatar
      Mohanad
      Icon for Cirrostratus rankCirrostratus
      Oct 31, 2019

      Thank Gr3asy so much, i will read K16049

  • Nath's avatar
    Nath
    Icon for Cirrostratus rankCirrostratus
    Oct 03, 2019

    What are you trying to achieve of having both F5 on both DC?

  • Hamish's avatar
    Hamish
    Icon for Cirrocumulus rankCirrocumulus
    Oct 06, 2019

    How to mitigate?

     

    You have multiple links. The links should have no single point of failure. i.e. separate switches, separate links from separate providers and make sure you check the actual paths of each

     

    H

      • Hamish's avatar
        Hamish
        Icon for Cirrocumulus rankCirrocumulus
        Oct 09, 2019

        If all your links are down, then you go split brain. But if all your links are down, then you have more problems than just the split brain BigIP.

         

        (Note that the way I usually do this is an HA pair in each DC (Or metropolitan DC pair). Then use GTM to provide global resilience).

         

        Regarding fancy devices not requiring your network to be up to prevent split brain. Why bother? You can always put in a path that would encompass the same links as the failover clustering using witness. You're not limited to just two links.

  • Azzeddine_S's avatar
    Azzeddine_S
    Icon for Cirrus rankCirrus
    Oct 08, 2019

    Actually connecting two devices directly to each other is not supported by F5

     

    The reason is pretty simple, if the link is down both devices will initate a failover process and you can end up with dual active scenario .

    what F5 is recomanding is passing through switches on both sides, in this case phisical failure will be contained within the segment (switch to BIG-IP) and the high avalability monitoring still work because both devices are monotoring each other through TCP.

     

    Passing through switches can also improve scalability, with direct connection you will be limited to only two BIG-IPs.

     

    This is for a single data center, for more you should carefully take in consideration the F5 recommendations for high availability such as band width for synchronization and mirroring , latency etc...

     

    good luck