Forum Discussion

johnbr_45658's avatar
johnbr_45658
Icon for Nimbostratus rankNimbostratus
May 12, 2009

Switching between servers before login completion

Hello,

 

 

We have a Bi-IP load balancer in front of two JBoss instances, which I will call 96 and 97. Failover between the instances occurs flawlessly.

 

 

 

We have a frequent problem with HTTP 408 errors on login, and, based on our analysis, here's the sequence of events:

 

 

1) User requests login page.

 

2) LB sends user to 96 or 97

 

3) User fills in login credentials & submits

 

4) LB sends the POST to the _other_ server (i.e. 97 or 96)

 

5) 408 Timeout

 

 

I'm also trying to chase this problem down from the JBoss side, but I thought I would also check here - can we get the LB to be more rigorous about sending users to the same server?

 

 

 

 

config
design

5 Replies

Sort By
  • The_Bhattman's avatar
    The_Bhattman
    Icon for Nimbostratus rankNimbostratus
    May 12, 2009
    What kind of persistance are you using? Do the client's appear to be behind a proxy or behind a PAT?

     

     

    CB

     

  • johnbr_45658's avatar
    johnbr_45658
    Icon for Nimbostratus rankNimbostratus
    May 12, 2009
    The two JBoss servers have their own IP addresses, and they both listen on port 80, and the domain name(s) we use map to the Load Balancer.

     

     

    I assume that means Proxy instead of PAT.

     

     

    Thanks for the reply!

     

     

  • The_Bhattman's avatar
    The_Bhattman
    Icon for Nimbostratus rankNimbostratus
    May 12, 2009
    Understood. But if you do not use persistance then there is a chance requests would jump between the 2 JBOSS servers. PAT standds for Point Address translations which can when users use a proxy to access JBOSS or through a firewall which translatest the clients address into one address.

     

     

  • johnbr_45658's avatar
    johnbr_45658
    Icon for Nimbostratus rankNimbostratus
    May 12, 2009
    Thanks.

     

    Once the user is logged in, we see them always go to the same server - "Sticky Sessions" I believe they called it. The problem appears to be from when they visit the login page until they log in.

     

     

    I'm presuming the LB doesn't particularly care about whether the user is logged in or not. Is there some volatility during the initial client tunnel setup?

     

     

     

    Oh, and we switch them from HTTP to HTTPS during this process as well. Don't know if that's a factor.
  • hooleylist's avatar
    hooleylist
    Icon for Cirrostratus rankCirrostratus
    May 18, 2009
    As cmbhatt was suggesting, you need to enable some mode of persistence in order to instruct LTM to try to persist a client to the same pool member over multiple TCP connections. If you have an HTTP and HTTPS virtual server set up, you can start with source address persistence on both virtual servers. Take a look at SOL5837 for related info (Click here). If that solves the problem, then you can look at more complex persistence options like cookie insert persistence.

     

     

    Aaron