Andrei Tokar

I've seen that before, but was not able to reproduce it locally, neither from TestPowerOff, nor from TestMultiThread running 100 times each.

> I don't think that it's proper to leak memory without the length specification. Technically, it is not a memory leak at all, and "proper" name would be "increased memory...

@wburzyns I am not trying to convince you, it was mostly a joke. But the truth is that while you are thinking about "one LOB of 10 KB", in reality...

>2. LOBs larger than Java array length limit need a specialized implementation for in-memory databases. @katzyn Do you really think it worth it to spend time (and bloat codebase) on...

@forchid Are you sure that the size you are talking about is all strongly reachable? CacheLongKeyLIRS may hold a lot of data using WeakReference, so 1M limit is only related...

@forchid Yes, you are right, I forgot that in absence of the size it counts plain number of entries (and 1 million is too much, of course) Huge space occupied...

And if we really going to support well much bigger data volumes, single-file storage idea (very convenient for small data sets) has to be abandoned, because on-disk garbage collection would...

I guess, what @wburzyns mean is this: **if** MVStore would use a rolling sequence of append-only files, you can copy all but the last file in any way you want,...

I was not talking about file per version. Of course that would be an overkill. But instead of doing garbage collection within a single file, which forces us to shuffle...

Unfortunately AtomicReferenceArray is not resizable and 1 million looks like yet another arbitrary parameter, but actually we do not have to have any special structure to hold chunks at all,...