We've got a 3 node DC cluster with ~ 45 plugins. A few times a month, we'll have issues with a plugin (random vendors, random issues) and the "fix" from the vendor and/or Atlassian is stop the application, delete the plugin cache, start the application. While we're running a > 1 node DC environment means we can do this without user facing impact (rolling restarts), but the time we drain a node off the LB, stop the application, delete the plugin cache, start the application, reconnect to the LB, let traffic re-balance before we move on to node 2, then 3 - this whole process can take ~ 60-90 minutes. Disabling and re-enabling an add on DOES NOT resolve the issues.

For an enterprise marketed product, this is not good enough. Ideally we'd expect plugins to not work on application start and then partially stop working after a random period of time. While the underlying issue is worked on though, we should at least be able to resolve the issue without an application restart. (eg: disable the plugin, have a web UI option to purge that plugins cache, and then re-enable puts the plugin back on disc).

This type of issue costs our support team hours per month in:

• Debugging issues with plugins
• Then having to resolve the plugin cache problem by running through a rolling restart/cache clear process