Should insert(...).get() provide some feedback information?

[lkairies]

From Oliver.H...@googlemail.com on November 01, 2010 14:51:07

It's important in many cases to distinguish an insertion of a new key from a modification of data of an existing key. According to this it could be useful if insert(...).get() returns a appropriate Boolean.

As for group inserts I could imagine a collection of Booleans or at least statistical information (#new, #changes).

I have similar ideas for (group) deletions: True means "An existing key has been deleted", False stands for "There isn't such a key".

Or does my proposal contradict the Key-Value-Store philosophy?

Original issue: http://code.google.com/p/babudb/issues/detail?id=16

[lkairies]

From sten...@gmail.com on November 02, 2010 03:14:06

Such a behavior would be feasible in theory. However, I'm against this, as it does not come for free performance-wise. Indices internally consist of a stack of in-memory trees and a sorted list of records on disk. To determine if a record is effectively replaced or deleted, an additional lookup would have to be performed before inserting or deleting the key, which might hit all of the internal data structures in the worst case. As against this, an 'unchecked' insertion or deletion only requires an insertion on the topmost in-memory tree, which is cheap by comparison.

If you need to find out if a record is effectively replaced or deleted, I suggest to perform this prior lookup manually.

[lkairies]

From david.yu...@gmail.com on December 23, 2010 08:46:00

This is where the bloom filter comes in. (The missing piece from Bigtable design).
It could be added a separate storage layer (to keep the current impl).
This would be efficient for atomic operations like compare-and-set, increment, etc.

P.S Stenjan, you seem to be the only one left committing to the java babudb ... nevertheless, good job for an awesome project!
Might be a good idea to have a dedicated babudb googlegroup.