Forum Discussion

Martin_Vlasko's avatar
Martin_Vlasko
Icon for Altocumulus rankAltocumulus
Nov 16, 2015

Monitor which does not bring the service down for 20 seconds?

Hi,

 

The default setting for monitor is 5 sec interval and 16 sec timeout. This means the total time during which a user might not be able to reach the service is at least 5+16 seconds, which is quite a long time.

 

I could decrease these values. Let's say, 2 sec interval and 7 seconds timeout. This would give me total possible downtime of at least 9 seconds. This is still quite a long time from user perspective, and on the other side, such low values can cause more issues because of the response speed of the server (it could drop legitimate connections which only need more time to load).

 

I want to be able to switch to other pool member quicker, for example immediately when I get a reset from the server.

 

Any ideas how to deal with monitors, so that they do not introduce 20 seconds of service downtime in case one pool member goes down?

 

Thanks.

 

devops
health monitor
LTM
monitor

9 Replies

Sort By
    • Martin_Vlasko's avatar
      Martin_Vlasko
      Icon for Altocumulus rankAltocumulus
      Nov 17, 2015
      Hi Nitass, Very good tip, thanks a lot, I will first try it with the inband monitor (in combination with active monitor). If I don't like the result I will go for the irule example.
    • Martin_Vlasko's avatar
      Martin_Vlasko
      Icon for Altocumulus rankAltocumulus
      Nov 17, 2015
      Hi Nitass, Very good tip, thanks a lot, I will first try it with the inband monitor (in combination with active monitor). If I don't like the result I will go for the irule example.
  • Brad_Parker_139's avatar
    Brad_Parker_139
    Icon for Nacreous rankNacreous
    Nov 16, 2015

    "The default setting for monitor is 5 sec interval and 16 sec timeout." That is not 5+16 until marked down. If a successful response isn't received with 16s it is marked down. The timeout value is the value that determines if a member is marked down. The interval is just how often a probe is sent. You could set your interval to 30s and have a timeout of 2s if you really wanted to. That would mean a probe is only sent every 30s and a member must respond within 2s or it will be marked down. Bets practice though is typically to set the timeout to the (interval*3)+1. That gives you the possibility of 3 failures before a member is marked down. I hope this explanation helps.

     

    • Martin_Vlasko's avatar
      Martin_Vlasko
      Icon for Altocumulus rankAltocumulus
      Nov 17, 2015
      Hi Brad, that's what I meant, just didn't explain correctly. Anyway, the inband monitor suggested by Amine and Nitass looks promising.
  • Brad_Parker's avatar
    Brad_Parker
    Icon for Cirrus rankCirrus
    Nov 16, 2015

    "The default setting for monitor is 5 sec interval and 16 sec timeout." That is not 5+16 until marked down. If a successful response isn't received with 16s it is marked down. The timeout value is the value that determines if a member is marked down. The interval is just how often a probe is sent. You could set your interval to 30s and have a timeout of 2s if you really wanted to. That would mean a probe is only sent every 30s and a member must respond within 2s or it will be marked down. Bets practice though is typically to set the timeout to the (interval*3)+1. That gives you the possibility of 3 failures before a member is marked down. I hope this explanation helps.

     

    • Martin_Vlasko's avatar
      Martin_Vlasko
      Icon for Altocumulus rankAltocumulus
      Nov 17, 2015
      Hi Brad, that's what I meant, just didn't explain correctly. Anyway, the inband monitor suggested by Amine and Nitass looks promising.
  • Amine_Kadimi's avatar
    Amine_Kadimi
    Icon for MVP rankMVP
    Nov 16, 2015

    You may consider passive monitoring with an inband monitor which is the quickest way to mark a member down. Don't forget to also associate an active monitor to bring the member up when it comes back again.

     

    " Passive monitoring occurs as part of a client request. This kind of monitoring checks the health of a pool member based on a specified number of connection attempts or data request attempts that occur within a specified time period. If, after the specified number of attempts within the defined interval, the system cannot connect to the server or receive a response, or if the system receives a bad response, the system marks the pool member as down. There is only one passive monitor, called an Inband monitor.

     

    A passive monitor creates no additional network traffic beyond the client request and server response. It can mark a pool member as down quickly, as long as there is some amount of network traffic. "

     

    Source: https://support.f5.com/kb/en-us/products/big-ip_ltm/manuals/product/ltm-monitors-reference-11-6-0/1.html