The problem is most likely the classloader bug that affects any CFC-heavy CF application. In other words, the site would have slow CFC initialization even if you dropped ColdSpring completely. I'd recommend not specifying lazy-init (it defaults to false) and just taking the hit once at startup rather than distributing the problem across different user requests. As Dan said, you can also try updating to Java 6 update 10, which corrects the classloader bug (it certainly took them long enough to fix this) and see how that affects things. It will probably make a very big difference.
On Thu, Oct 2, 2008 at 2:46 PM, Roy Martin <[EMAIL PROTECTED]> wrote: > *Problem Overview* > We've been using ColdSpring a lot more in our recent projects, having > roughly 65-100 beans being managed and auto-wired into our Model-Glue > applications. One big side-effect of this however has been the spin-up with > our sites. Since we're auto-wiring and have them all defaulted to > lazy-init=true, visiting a feature on the website for the first time that > injects a a few service layers with several sub-components might take 2-3 > minutes to compile on a busy server. Future page loads are very quick, but > this means we have to run functional test / unit test on the live site with > extremely high timeouts that would touch every aspect of the site to ensure > that all components are compiled. This was amplified this week when one of > our servers was forced to restart after a barage of brute force attacks, > killing the compiled classes and having to recompile them all over again and > we're getting random reports of sites being down, all related to this issue. > > *Resolutions?* > What's the best practice to mitigate this? Here are just some of my > thoughts that I would love some feedback on: > > - Turning all of the root components into lazy-init=false? Doing this > would definately yeild and extremely long spin-up time for our sites, but > should at least fix the hang-up on feature load. > - Changing the JAVA version - I know there was a lot of discussion > about the CF 8.0.1, which is where I'm experiancing this, having the bug > with the class compiles. Is there a stable version someone else is running > in production that has greatly fixed this issue? Overall is anyone running > an alternate recommend Java version in production for CF8.0.1? > - Maybe expanding on the discussion earlier about a read only server. > If we could somehow compile these classes ahead of time then at least we > could do the work up-front on our development machines and not have to > compile on the fly on production that would obviously make the most sense. > I'm not necessarly suggesting an out-of-box approach baked into ColdSpring > but if there was at least a recommended way to accomplish that that would > be > great > - Option D - none of the problems are because of the java version, > large amount of ColdSpring beans at all, it just has to do with setting X > in > my ColdFusion or just me using ColdSpring incorrectly that's causing the > problem. If so I'm all ears. > > > Thanks, > Roy >
