PersonalizedCachePolicy injects context provider and accesses the website workspace in order to account for page variants during caching, typically logging in as anonymous (in production end user scenarios).

Under heavy load this makes a lot concurrent users attempt to log in to JR via its user manager and blocking themselves (user manager uses a single synchronised session to manage log in credentials, see org.apache.jackrabbit.core.security.user.UserManagerImpl#session).

Using system context alleviates the issue somewhat (admin credentials are handled via shortcut), but later during interaction with the repo, we end up locking ourselves on some of its synchronised parts.

Can we optimise the policy so that it at least reduces/scales interactions with the repository?