Comware

 View Only
  • 1.  Bond failover ping loss

    Posted Aug 18, 2023 12:59 PM

    Hi guys,

    i neede some help.

    Per todays session we have discovered some interesting things


    OS runs Bonding in active \ passive

    switch has no bonding but STP edged enabled

    Whenever you do a failover of the bond master you will drop 2-20 packets

    Whenever you do a if down of the bond master you will drop 2-20 packets

    However...

    If we down the interface on the switch side we have no packet loss (shutdown interface xyz)

    This was kinda what I was discussing testing with someone onsite pulling the cable, so I'm guessing the result will be the same with a pulled cable, i.e no packet loss

    I'm curious that the interfaces switch side stays UP when you down them in the OS, I'm not 100% sure this is correct behavior.. we can't see anything special in the switch logs at least.

    I would say the failover behaves as expected during an actual error (death of switch, card, cable etc) so that should be good for most stuff, however it should also behave properly on a manual failover, we need to investigate why it does not?



  • 2.  RE: Bond failover ping loss

    Posted Aug 21, 2023 11:08 AM
    Hi, can you explain better these two conflicting sentences?

    "OS runs Bonding in active \ passive

    switch has no bonding but STP edged enabled"

    Personally, if I read "bond" [*] on one side (your host side) then I expect to find a matching "bond" on the other side (your switch side). Isn't it?

    [*] to me a "bond" represents a logical interface made of two (or more) physical interfaces aggregated together using a particular aggregation mode.






  • 3.  RE: Bond failover ping loss

    Posted Aug 21, 2023 02:08 PM

    Hi Parnassus,

    thanks for the quick response. We have on one side one storage node and on the other side 2 comware 5945, because the bond from the storage node has 2 connections to 2 different 5945 for redundancy. So when we play arrond and do some tests we saw some packets were lost.

    If i from the switch side shutdown the ports bond failover works as expected, no packet lost, but when shutting down the port from server/storage side we have packes getting lost, when the we pull out the cable we see again some packets getting lost.

    Bond mode 1 active/backup

    Switch port - stp edge port

    Thanks in advance!




  • 4.  RE: Bond failover ping loss

    Posted Aug 21, 2023 02:34 PM
    Aren't the two HPE 5945 part of an IRF? thus are both switches running standalone (with respect to connected peers)?





  • 5.  RE: Bond failover ping loss

    Posted Aug 21, 2023 02:46 PM

    Hi,

    no, it is a DRNI config between switches.




  • 6.  RE: Bond failover ping loss

    Posted Aug 21, 2023 04:02 PM
    Edited by parnassus Aug 22, 2023 06:53 AM
    But from the peer standpoint both - the DRNI and the IRF - are the same with regard to bond interface on the peer (network side: Multi-Chassiss LAGs in case of DRNI or BAGGs in case of IRF)...so you have DRNI without a MC-LAG to your storage server (where a bond is configured)?






  • 7.  RE: Bond failover ping loss

    Posted Aug 21, 2023 04:26 PM
    To me the DRNI approach on Comware looks very similar to the VSX approach on AOS-CX.

    Given that, an host configured with a Bond interface connected to a DRNI should be connected to a MC-LAG (called DR Group) of that DRNI, exactly as that very same host would be connected (from its very point of view) to a VSX or to an IRF, the former uses VSX LAG (equivalent to MC-LAG or DR Group) due its separate control plane and the latter uses simple LAG due to its unified control plane architecture.

    In any case Bond definition on peer - to me at least - should mean using DR Group=MC-LAG on DRNI or VSX LAG on VSX or, again, simple LAG on IRF. Maybe I'm wrong with that but I would double check that "host-to-DRNI architectural approach" first.





  • 8.  RE: Bond failover ping loss

    Posted Sep 13, 2023 06:12 AM

    "OS runs Bonding in active \ passive"

    can you give more config details ?

    (When I see "active/passive" I read "not LACP")

    On the switch, check for stp status of interfaces (maybe the bonding sends or "passes" BPDUs ?

    display stp

    display stp abnormal



    ------------------------------
    Frederic MEUNIER
    ------------------------------



  • 9.  RE: Bond failover ping loss

    Posted Sep 13, 2023 06:23 AM

    Hi,

    thanks for the replay, but i think we find out the issue, that is on the other side in the Linux drivers 99%, we are waiting for some verification from the Linux administrators, but the issue is not on switches.