Forum Discussion

cjunior_138458's avatar
cjunior_138458
Icon for Altostratus rankAltostratus
Dec 16, 2014

Health monitor and shared backend server/services

Hello, Since the boxes active / standby, makes the monitoring of servers, so we have two sources to the same destination monitored. If a monitor is making calls by default every 5 seconds, or 12 per minute or 720 per hour. We will have a total of 1440 requests in an hour to the same destination, right? It gets bad when a server or service divided by more than one pool (1440 * N).

 

We can consider a problem if a server is weaker than the other? Because it will receive connections from clients and the BigIP (currently, the monitor HTTP / 1.1 closes the connections). If I increase the time monitoring, for example, the service may be unavailable for someone who will fall on a server that is supposed to fail in the period being not monitored, have some effect by using "reselect" parameter in "action on service down" for this case? To decrease by less the network traffic, I used the HEAD verb, and I captured the HTTP response, that's good? someone do that?

 

What do you usually do to calibrate a health monitor to http to be highly available but fewer requests in monitoring to preserve the backend servers? This is really something I should worry about? You can share your experiences? Thank you so much!

 

application delivery
LTM

8 Replies

Sort By
  • PeteWhite's avatar
    PeteWhite
    Icon for Employee rankEmployee
    Dec 18, 2014

    You are correct - it is a trade-off between amount of traffic and time to detect issues. If you're not sure then go with the default times. If you have issues with servers bouncing then reduce the timeouts. It makes sense to use the HEAD statement although this doesn't send the data obviously so you'd have to just check the response ie "200 OK"

     

    • cjunior's avatar
      cjunior
      Icon for Nacreous rankNacreous
      Dec 19, 2014
      We have some http binds and service ports nitass. I think that node just if I have one service per node, am I correct? Thank you.
    • nitass's avatar
      nitass
      Icon for Employee rankEmployee
      Dec 19, 2014
      i thought you may have multiple websites running on one instance, so checking only one website may be enough (because there is one instance). other things i can think of are (1) to use up interval health monitor setting or (2) to use large interval but mark pool member down (LB::down) and reselect pool member when LB_FAILED is triggered.
    • cjunior's avatar
      cjunior
      Icon for Nacreous rankNacreous
      Dec 19, 2014
      We have some http binds and service ports nitass. I think that node just if I have one service per node, am I correct? Thank you.
    • nitass_89166's avatar
      nitass_89166
      Icon for Noctilucent rankNoctilucent
      Dec 19, 2014
      i thought you may have multiple websites running on one instance, so checking only one website may be enough (because there is one instance). other things i can think of are (1) to use up interval health monitor setting or (2) to use large interval but mark pool member down (LB::down) and reselect pool member when LB_FAILED is triggered.
  • cjunior's avatar
    cjunior
    Icon for Nacreous rankNacreous
    Dec 29, 2014

    Have any of you combined the HTTP monitoring with inband monitor? I plan on using it then I would leave the "up interval" of http monitor in 60 (maybe 90) seconds and leave the inband mark the member as down in this interval when there was some miscommunication. Well, some guests will suffer for this sometimes if service fails until to be marked down by inband monitor. What do you think of that?