Uploaded image for project: 'Geode'
  1. Geode
  2. GEODE-2290

Provide way to limit scanning of deployed jars




      Currently if you use the GFSH command "deploy jar" the deployed jar will be scanned in such a way that the deployer will load and resolve every .class file in the jar file. The original intention of "deploy jar" was that only Functions would be deployed. Any .class that is found to be a Function is automatically registered with the FunctionService.

      You can also include implementations of other Apache Geode callbacks such as CacheListener. If you then follow up the deploy with "alter region" you can alter the region to add the CacheListener or other callback.

      Some users have reported trying to deploy a jar with much more than just Functions or CacheListeners or domain classes used for Region Entries. If the jar includes a class that the callbacks or domain classes don't directly use but that class requires a missing transitive dependency, then the deployer will generate a NoClassDefFoundError when it tries to load and resolve that class.

      Along with the potential NoClassDefFoundError, forcing every .class to be loaded and resolved can consume a lot of PermGen memory.

      Various ideas have been tossed around to allow someone to deploy such a jar without having the deployer generate NoClassDefFoundError. This requires something like one of the following approaches:

      1) add a new --do-not-deploy option flag
      2) comma-delimited list of classes to load and deploy
      3) regex list of classes to load and deploy
      4) have I missed something?

      This would give the User better control over which classes to actually load and resolve.

      [NOTE: I'll update this ticket based on feedback and discussion on dev@geode.apache.org]




            jstewart Jared Stewart
            klund Kirk Lund
            0 Vote for this issue
            3 Start watching this issue