Monday, February 4, 2013

Stuck-In-Processing Solved!

Our stuck-in-processing problem continued and then during our annual enrollment period, it worsened dramatically.  We attributed it to the increase traffic during enrollment, but after enrollment it didn't get much better.  We investigated problem further and found "How To Identify Session Stickiness Problems On Load Balancer?" [ID 1307344.1] on Oracle support.  We also found this article useful, "How to Setup PeopleSoft and WebLogic when using a Load Balancer [ID 653998.1].

After doing some testing with going directly to a web server, rather than through the content switch, all clues pointed to the content switch as the source of the problem.  Working with our network group was slow.  They explained that the content switch only passed on what it was given and couldn't be the source of a problem.  We used a sniffer to trace the network connection on a user that experienced the stuck-in-processing problem a lot.  They saw that the user was fine until a reset was sent to the user from a different server than the one that they were connected to.  


The fix for the problem did turn out to be a setting on the content switch.  There was a session timeout that was set to a low setting -- less than a minute.  Here's what our network guy said "...the solution is to increase the flow timeout setting.  I decided to change the set the setting to what {server} is using (560 seconds or 9.33 minutes) to fix this..."

My understanding of this issue is not complete.  Here is what I have so far.  Please let me know if you have more information.  Starting with IE 8, the browser opens six connections to the server as soon as you login.  Depending on the what you are doing, PeopleSoft may use just a few of these connections.  Then if you go to a page that is more complex, you may use more of the connections.  With a short timeout, these previously unused connections may have timed out and could then connect to a different web server.  That web server, unfamiliar with your connection, would send a reset.  Then you were stuck-in-processing.

This fix has solved our problem.  I hope it helps you.