I'm guessing you have to have your application wait for cache start-up to complete before accessing the cache.
Thanks genman, but my app and the cache are both fully started. In fact, using Node directly instead of TreeCache works. The problem this way is that interceptors dont get called, which may prevent getting evicted nodes.
what I want to know is if this jira is a bug, and why it wont be fixed, or if its a feature, which i dont get.
Regarding direct access to nodes, this has been added in 2.x. Node is a top-level user API now and calls to Node will be passed through the interceptor chain.
Regarding JBCACHE-622, I should have been more explicit on the JIRA: the issue of preferring what is retrieved from the cache loader (a null) over what is in memory is correct behaviour since the cache loader may be shared, etc.
The real bug is probably JBCACHE-131, to persist state when doing an initial state transfer (provided your cache loader is not shared/a singleton and you are not using passivation).
Will look into potentially reviving JBCACHE-131.