Forum Discussion

kldev_132889's avatar
kldev_132889
Icon for Nimbostratus rankNimbostratus
Sep 03, 2013

F5 connection limit and queuing, unexpected behavior

I setup the F5 to limit the number of connections to each member server in the pool associated with the VIP and configured them each with a connection limit of 500. Then I turned on in the pool Enable Request Queuing. However, I ran into a issue where each member server in the pool showed that the server had hit their connection limit and were taken out of service. But, when the server team looked at the servers they didn't see any connections except a handful, but it did appear that the F5 was queuing the traffic. Is there something else that needs to be modified to prevent that issue? The servers normally on average have only about 300 connection each. I am trying to figure out why the F5 didn't see that the servers had no connections and forward the connections?

 

  • Are you sure the server people are looking at the right metrics? Is OneConnect in use here?

     

    Perhaps try and use the [tmsh] show sys connection command (with appropriate filters) to get an idea of the number of connections in the connection table on the F5 when it reports limits have been reached (is it anywhere near 500 etc.)

     

  • are you using tcp queuing with oneconnect? if yes, can you try to turn off tcp queuing? there is a bug which connection is miscounted when tcp queuing and oneconnect are used together. you may open a case to verify if you are experiencing this bug. this bug has not been fixed.

     

    Bug 402510 - TCP connection queuing and OneConnect miscounts at conn limit

     

    • kldev_132889's avatar
      kldev_132889
      Icon for Nimbostratus rankNimbostratus
      There isn't a oneconnect profile attached to the VIPs. And I have disabled the re-queuing at this time as it was impact traffic.
  • are you using tcp queuing with oneconnect? if yes, can you try to turn off tcp queuing? there is a bug which connection is miscounted when tcp queuing and oneconnect are used together. you may open a case to verify if you are experiencing this bug. this bug has not been fixed.

     

    Bug 402510 - TCP connection queuing and OneConnect miscounts at conn limit

     

    • kldev_132889's avatar
      kldev_132889
      Icon for Nimbostratus rankNimbostratus
      There isn't a oneconnect profile attached to the VIPs. And I have disabled the re-queuing at this time as it was impact traffic.
  • I can re-enable the queuing and connection limit and see what is going on.

     

    can you also try to fine-tune queue setting (i.e. queue-depth-limit, queue-time-limit) to see if there is any difference?