Forum Discussion

symsonu's avatar
symsonu
Level 6
13 years ago
Solved

VCS behavior

We have one two node SFRAC cluster i.e parallel one. I need to understand the behaviour of VCS for the below situation.

 

We have freezed one node by hasys -freeze node1 and then we pull both the heartbeat cables from the node.

how will cluster behave ?

Note:--> there is no lowpri heartbeat link and there is only two normal HB links.

 

 

business continuity
Cluster Server
  • mikebounds's avatar
    mikebounds
    13 years ago

    If you have SFRAC, then you must have fencing, so if you pull both HB cables at the same time, one of the nodes will panic.  Regardless of which nodes panics, any parallel service groups will remain online (like Oracle service group) , but if you have any failover service groups, then if node 1 panics, they will failover, but if node2 panics, they will not failover as node1 is frozen.  If you don't want failover service group to failover then you should freeze the failover service groups, not freeze the systems.  Freezing a system is generally used if you temporarily want to stop service groups failing to a particular system and you don't want to stop VCS on that system.

    Mike

3 Replies

Sort By
  • mikebounds's avatar
    mikebounds
    Level 6
    13 years ago

    If you have SFRAC, then you must have fencing, so if you pull both HB cables at the same time, one of the nodes will panic.  Regardless of which nodes panics, any parallel service groups will remain online (like Oracle service group) , but if you have any failover service groups, then if node 1 panics, they will failover, but if node2 panics, they will not failover as node1 is frozen.  If you don't want failover service group to failover then you should freeze the failover service groups, not freeze the systems.  Freezing a system is generally used if you temporarily want to stop service groups failing to a particular system and you don't want to stop VCS on that system.

    Mike

  • symsonu's avatar
    symsonu
    Level 6
    13 years ago

    Hello Mike

     

    Thanks for the response. I understood what you explained to me.

    There is a two node sfrac cluster , one node each in different datacenter.In one of the datacenter LAN services will be unavailable i.e both the heartbeats unavailable at the same time ( due to parallel work on upgrade of switch)

     

    From your explanation, I think it would be wise to stop the cluster services on the datacenter where maintenance is planned. As I/o dencing is enabled ..anynode can crash on heartbeat loss.

     

    will it be sufficient to give hastop -local on the affected cluster node.?

     

    Sumit

     

     

  • mikebounds's avatar
    mikebounds
    Level 6
    13 years ago

    I would unload fencing, gab and llt as well - you can use rc scripts to do this.

    Mike