Thanks to David for answering most of these questions.
Sebastian Wrede wrote:
* adding indexes to an already populated container fails when there are
more than about 5000 documents present. all documents are smaller than
1kb. adding the index first and then populating the database works, but
we'd like to be able to add indexes afterwards to optimize query
performance.
The error is:
Dbc::get: Cannot allocate memory.
Again, some code:
DbTxn *txn;
env_->txn_begin(0, &txn, 0);
XmlIndexSpecification(idxSpec=dbxml_.getIndexSpecification(txn);
idxSpec.addIndex (spec.uri_, spec.node_, spec.index_);
dbxml_.setIndexSpecification(txn,idxSpec);
txn->commit(0);
URI is usually empty.
note that we used default database settings throughout and the
environment has been opened using the flags:
DB_INIT_LOG|DB_CREATE|DB_INIT_MPOOL|DB_INIT_TXN|DB_INIT_LOCK
Adding the index causes the indexer to walk over all the documents creating
additional index entries for each document. With many documents this creates
a really huge transaction. I'd suggest you don't transact calls to
addIndex. You
probably don't want to do indexing and update operations at the same time
anyway.
In addition to the above problems, we have two other feature questions:
* is it possible to abort a query in some way, e.g. from a signal
handler? of course, without damaging the database ;-)
Hmm, we could maybe have a progress callback that the query thread calls
between
operations... the callback would get some progress indication and could
return an
error code to terminate the query...
But, why do you want to do this? Because queries take a long time and
the user
gives up and wants to abort them? How about an estimate of the query
cost instead?
* is there a way to use an index for queries like '[@timestamp > 10 and
@timestamp < 20]'? an edge-attribute-equality-number index is used for
equality comparisons but not for interval expressions, apparently
Sebastian Wrede & Ingo Lütkebohle
We do support 'range' operations against numeric indexes... BUT... only
against node
indexes, not edge ones. Would a node index work OK for you instead?
John
| 
