Forum Discussion

patrick_hayes's avatar
Icon for Nimbostratus rankNimbostratus
Feb 01, 2011

Monitoring in a cluster

Hello all,



One information I have been unable to find in the documentation either online or off. Do monitors run on only on the active member of a cluster or on both the active & passive members?



Yes, a simple test once the cluster is installed would give me the answer but this is my first cluster install, I'd like to know beforhand & I hope the answer is easy for one of you to answer...





8 Replies

  • nathe's avatar
    Icon for Cirrocumulus rankCirrocumulus



    Monitors run from both boxes and use the non-floating address of both boxes as the source of the traffic.





  • Thank you Nathan,



    Does anyone know what the status of nodes on a cluster will be when the the monitors only work from the currently active device?



  • Posted By patrick.hayes on 02/03/2011 07:19 AM


    Thank you Nathan,



    Does anyone know what the status of nodes on a cluster will be when the the monitors only work from the currently active device?





    They'd be down. Any reason they'd only work from the active device? If that's the case and there's a fail-over, it'll take the new active unit a bit of time to realize the nodes are now up.
  • Thank you Chris,

    There are two redundantly connected HP superdomes hosting a number of ESXi virtual machines where the F5 active/passive cluster is directly connected to a port on each superdome.

    +----------+ redundant links +----------+
    +----------+                 +----------+
          |                     |
          |1.1                   1.1 |
    +----------+ 1.2         1.2 +----------+
    | activeF5 |-----------------|passiveF5| 
    +----------+                 +----------+ 
          |1.3                   1.3 |

    1.1 is inside (using multiple tagged VLANs), network failover+session sync os on 1.2 & outside is on 1.3

    The problem is that the virtual switch inside the Superdome performs a kind of spanning tree where only one exit port is functional as a kind of spanning tree is used to shut down the connection to the passiveF5 which normally only becomes active when the port on the activeF5 fails. Following Nathan's reply I placed the inside VLANS on 1.2 as well as 1.1 to allow the passiveF5's monitors to function.

    No, this is not the architecture I advised. IMO, both the F5s & the Superdome assume that there is a redundant switch layer between them but the network guy thinks that because the F5 can do the bridging by itself that we should use the F5's to do so. I'm currently performing tests to see whether or not this will work or if we must use the usual architecture with redundant switches.

    Anyone who has been in a similar situation & has any advice they can share would be appreciated...
  • Hamish's avatar
    Icon for Cirrocumulus rankCirrocumulus
    Put a switch between the superdome(s) and the F5's. If you want redundancy, use two and bridge them.



  • Thanks Hamish, that's my position as well.



    Unfortunately a pre-sales from F5 told the net admin that a redundant switch layer between the F5s & the superdomes isn't really necessary so every time I push for a redundant switch layer he uses that to bludgeon me with...



    I need more than just "it's not a best practice" or "use of this configuration is highly unusual and puts you at risk of encountering bugs that would never show up otherwise" to convince him.



    Better arguments welcome...
  • As a final update on the subject, The F5 works very well all by itself without needing a intermediate switch.



    Placing the VLANS that can come from either of the SuperDomes on both the 1.1 & 1.2 interfaces will let the F5s treat data coming from either Superdome. For example, if the active link to the Superdome is on the left & the active F5 is on the right, the left F5 propagates traffic to the righthandsided F5.



    Thanks everyone for your help!