Forum Discussion

adithyodw_18563's avatar
adithyodw_18563
Icon for Nimbostratus rankNimbostratus
Aug 31, 2015

F5 Pool Members flapping issue

Dear All,

 

We are facing issue with monitoring of nodes in one of our virtual server pool, however when we check on the server side we didn’t found any discrepancy on server port side. Due to this F5 keep flipping those servers out of pool and hence we are facing issue . Appreciate please do let us know and what are the fix for this in case any bug or hotfix. Please do let us know in case if you required any further information

 

please check the log : size 128KB download here on google drive

 

thank you

 

application delivery
availability
LTM

26 Replies

Sort By
Show More
  • GaganD's avatar
    GaganD
    Icon for Nimbostratus rankNimbostratus
    Sep 02, 2015

    Hi Buddy, Looking at the tcpdump, your F5 is doing a TCP health check on different server ports. Normal behaviour would be that F5 will only try to connect to the port on which member server is listening on. So for example 10.64.32.16 is listening on port 44070, so health checks should only happen on port 44070. Can you confirm what monitor are you using to monitor the node 44070(may be problem is at node level)? If you are using TCP, then I can understand your problem. Please change it to icmp. You can find it out by typing list /ltm node 10.64.32.16 from tmsh

     

  • adithyodw_18563's avatar
    adithyodw_18563
    Icon for Nimbostratus rankNimbostratus
    Sep 02, 2015

    Hi gagan, based on gui, im using tcp for health monitors for 44070, so i must change it to gateway_icmp? and why? thank you

     

    • Shaun_Simmons1's avatar
      Shaun_Simmons1
      Icon for Altostratus rankAltostratus
      Sep 02, 2015
      ICMP can be used to quick-check the path to the server(s) are good. I use it when I know I have the LTM configured correctly but don't know if the network or server may be the issue.
    • adithyodw_18563's avatar
      adithyodw_18563
      Icon for Nimbostratus rankNimbostratus
      Sep 02, 2015
      hi gagan, based on gui 10.64.32.16 using monitoring icmp. should i change?
    • GaganD_191239's avatar
      GaganD_191239
      Icon for Nimbostratus rankNimbostratus
      Sep 02, 2015
      If node is using icmp as its health check at node level then you don't need to change. But if its using tcp to monitor node at node level then yes please change it to icmp. ICMP is good enough health monitor at node level, all fancy monitors should be used at pool level to monitor health of application. Since by default tcp(or all health monitors in fact) does not specify destination port in the monitor config(its usually set to wildcard), it does not know which port to monitor, But when it is assigned to pool which has member listening to specific port(like 10.64.32.16:44070) in you case, it take the port number (44070) from pool member and use it in its monitoring. So it know what port to monitor. But when you assign the same monitor tcp to a node, which has no port number specified(and you actually cannot specify it), monitor does not know which port to monitor, so it can cause to problems.
  • GaganD's avatar
    GaganD
    Icon for Nimbostratus rankNimbostratus
    Sep 04, 2015

    Hi Buddy,

     

    After checking you tcpdump, I can see that your problematic server us sending TCP zerowindow to F5. upon which after waiting for some time, F5 tears down the session by sending resets, causing health check to fail. There is something which is making your server really busy, to an extent that it is sending message that it cannot receive any more data. What is cpu, memory, interface stats on the server.