Pythian Group

Today I ran into an 11i E-Business Suite instance which is using Apache/JServ to do forms load-balancing. In fact, it’s my first customer to implement and successfully use Apache/JServ for load-balancing. Here is quick sketch of the instance architecture:

                  Users
                    | 11i URL:
                    | http://appsrv1.pythian.com:8000/
                    |
             ----------------
             |   appsrv1    |
             | apache/jserv |
             |  formservlet |
             ----------------
            /                \
           /                  \
----------------             ----------------
|   appsrv2    |             |   appsrv3    |
| apache/jserv |             | apache/jserv |
|  formservlet |             |  formservlet |
----------------             ----------------

            -------------------
            |    dbsrv01      |
            | CM/RepSrv       |
            |  Database       |
            -------------------

Almost all the customers that I have worked on till now, with a requirement for multiple middletier, have implemented load balancing using hardware load-balancers like Cisco Content Switch or F5 Big-IP.

This customer is running forms in servlet mode, a prerequisite for using Jserv to do load-balancing. If somebody wants to do forms load-balancing in socket mode and opts not to use any hardware LBRs, then the only option is a Metrics server.

Here is brief overview on how forms load-balancing happens in above architecture. Read the rest of this entry . . .

mysql-proxy defaults to round-robin load balancing. There are fancy tricks around to get mysql-proxy to balance connections based on how many idle connections there are in a proxy-based connection pool.

But there is no code that I found that would simply load balance based on “always go to one server, go to another server only when the first server is down.”

Well, I spent way too long figuring this out today, again running into the problem where the manual hasn’t been updated. I have indeed made a Forge snippet of this code, but it does not hurt to post it here.

This was in fact taken from http://jan.kneschke.de/projects/mysql/mysql-proxy-learns-r-w-splitting, which I referenced above as having the fancy trick above — but there is one very important difference — the internal structure proxy.servers has been changed to proxy.backends. Sadly, the manual page for the Internal Structures has not been updated to reflect this.

At least now there’s a comment on the documentation page that states that.

function connect_server()
  for i = 1, #proxy.backends do
    local s = proxy.backends[i]

    if s.state ~= proxy.BACKEND_STATE_DOWN then
      proxy.connection.backend_ndx = i
--      print ("connecting to " .. i)
      return
    end
  end
end

function read_query(packet)
  for i = 1, #proxy.backends do
    local s = proxy.backends[i]

    if s.state ~= proxy.BACKEND_STATE_DOWN then
      proxy.connection.backend_ndx = i
--      print ("connecting to " .. i)
      return
    end
  end
end

First of all “the before” time is over — I’m done with my presentation. It’s been the first slot of the day — 8:30 and Cary Milsap was presenting in another hall so what chances do I have to get people in? It turned out that some people actually did show up and quite a few considering the circumstances.

I have mixed feeling on the results. The presentation started very well and I managed to wake people up at the very beginning — thanks to the “equipment” I had at hands (thanks Marco and Riyaj!). You can spot one of them on the photo (thanks for the photo Marco):

Read the rest of this entry . . .