Re: [xwiki-devs] Fw: Re: [Issue] Commented: (XWIKI-2006) allows to configurate name of database schema [patch[

From reading source code: situation is more interesting, that I was think before. Why: We need call setDatabase() not once per xwiki, but once per new hibernate session. But sessions are controlled by hibernate, not by us. Typical solution in such situation (near any middleware has something simular) is to keep own weak map of sessions, and if database was not set for this session -- set one. I see exists commented out 'ConcurrentHashMap' of connections (for monitoring) which can server this purpose - but non-weak HashMap without nurse process can cause memory leaks. (it's why one commented) So, my original plan now looks like incorrect, instead we must choose one from three possible solutions: 1) [I guess best in short-term perspective] Force setDatabase() in beginTransaction() only. This will solve all security problems and partially - overhead. But we will have non-empty overhead for non-virtual mode and non-empty xwiki.db, becouse once-per-transaction is worse then once-per-session (but better than onec-per-operation). Since this overhead will take place only when we set xwiki.db, I think that this is acceptable and will submit appropriative patch. (in case of absence of abjections). 2) Restore map of connections, but not as ConcurrentHashMap, but as WeakHashMap with lock. This will restore functionality for monitoring, but add overhead (I think small, but who know about large-scale installations) for serializing sessions access. 3) In reality we need data structure, something like ConcurrentWeakHashMap. [JBoss Remoting hase one]. We can get one from JBoss, but it's possible to write one from scratch (and may be safer for license issues). I think this is the best way in long-term perspective, but will require some additional work. So, I want now do [1] and may be think about [3] in future, but without any estimations of time. -- Ruslan Shevchenko GradSoft. http://www.gradsoft.ua