Btree corruption handling question

[Jim Wissner <ji...@jbrix.org>]

Hello,

I've been working with Xindice on and off since last year on a couple 
different projects, both server-side and client side (embedded in 
application).  As I've mentioned before, I think it's an impressive piece 
of work, and have been trying to evangelize as fit.

Anyway, I'm now working more closely with some of the internals of Xindice, 
specifically BTreeFiler and ValueIndexer, as the core for some more finely 
tuned app-specific db needs (independent from the primary Xindice database 
files).

It all seems to work quite fantastically, except that it doesn't appear to 
be too fault tolerant.  The code must be solid enough to work in a fairly 
volatile environment where crashes/full disks/etc may be anticipated.

What I'm experiencing is that when these things happen, quite often the 
BTree (BTreeFiler, ValueIndexer) files leave themselves in a corrupted 
state.  It's generally apparent when just trying to read back from the 
file, throwing an exception at the end of 
BTree.query(IndexQuery,BTreeCallback), under "No Query - Just Walk The 
Tree", where the Invalid Page Type exception is thrown.

The question is - is there something specific this is indicating as to the 
cause of corruption, and more generally, how does the larger Xindice system 
account for such corruption?  Could anyone point me towards the right place 
in the xindice source tree, or even give some pseudo-code, as to what a 
good strategy for handling and recovering from such corruption would be?

Many, many thanks in advance..

Jim




--
jim@jbrix.org

Visit www.jbrix.org for:
   + SpeedJAVA jEdit Code Completion Plugin
   + Xybrix XML Application Framework
   + other great Open Source Software