Thanks for the explanation. I will look towards the tomcat mailing list. Per your request about the entire stack trace - there is more above the first call to PDFBox, but the root cause in both cases is indeed "java.lang.NoClassDefFoundError: Could not initialize class". And I have seen that same error with other ClassLoader problems under Tomcat.
On Wed, Aug 1, 2018 at 1:39 PM, Christopher Schultz < [email protected]> wrote: > Joel, > > On 8/1/18 3:54 PM, Joel Hirsh wrote: > > However, your comment that "Tomcat doesn't unload libraries" is not > > entirely true. There is an explanation of such symptions at > > https://haraldk.github.io/TwelveMonkeys/, under the section "Deploying > the > > plugins in a web app". > > Thanks for posting that. Lemmie 'splain. > > In a servlet container, web applications (aka "contexts") each have > their own ClassLoader and if the application is undeployed or > re-deployed, then that ClassLoader will be freed for GC. If a class > loaded by that webapp's ClassLoader is still in use, you have a memory > leak situation knows as a "pinned ClassLoader". > > It happens all the time when you use a system-global cache such as > ImageIO does. This can happen with JDBC DriverManger and a handful of > other common APIs and it can be a serious problem for servers where > there are lots of redeployments. You can also start to see errors like > "ClassCastException: Cannot cast foo.bar.Class to foo.bar.Class" because > instanceof-ness is defined not just by Class but also by ClassLoader. > > The IIOProviderContextListener mentioned in that documentation is one > strategy to handle the unfortunate way ImageIO caches classes. Another > strategy is to put the library into a place where it is accessible to > all web applications instead of just one. > > > Just putting jars in a shared folder did not help. > > Here is where asking on the Tomcat mailing list *will* be a good idea. > Make sure to give your exact Tomcat version and explain where you put > the library. > > Can you post the *entire* stack trace from the failure? It looks like > it's incomplete -- it doesn't seem to have a "true root cause". It just > says "Could not initialize class" because "Could not initialize class". > Somewhere there should be a real root cause like NullPointerException or > something that can actually be debugged. Def post that to the Tomcat > mailing list as well if you post. > > -chris > >
