Why does FCI not have worry of split brain scenario?



  • SQL server AG has lease time-out to ensure there will never be a split brain scenario.

    Where as there is no such thing for FCI.

    Both AG and FCI (HA) require the windows failover cluster, but why does FCI not have worry of split brain scenario?



  • An availability group uses two fully independent systems (compute+storage) and relies on synchronization technology to keep both nodes up to date. It is possible, if they lose communication with each other for them to think they are the primary, the witness you select for your Failover Cluster is supposed to prevent that.

    A Failover Cluster Instance uses shared storage with separate compute. Given the nature of the storage, it is not possible for two threads to share access to the same storage, so you can't have a split brain there. Note caveats about geographically dispersed FCI's using SAN witchcraft to replicate storage.

    A Failover Cluster (OS layer) uses a witness to try and prevent split brain. A node will only automatically come online to primary IF it and the witness agree that it can do so. However, there are scenarios in which the primary node will get isolated (can't talk to secondary OR witness) and it may continue on thinking it's the primary (and any clients that are talking to it). Which is why whatever you use for the witness needs to be HA as well.




Suggested Topics

  • 2
  • 2
  • 2
  • 2
  • 2
  • 2
  • 2
  • 2
  • 2
  • 2
  • 2
  • 2
  • 2
  • 2
  • 2