On 6/6/06, Darren Vengroff <[EMAIL PROTECTED]> wrote:
I appreciate your performance point.  One of the things I considered, if
performance becomes an issue, is putting a cache in that would map from a
field name to the full set of fields it should copy to.  We would only go
through the longer dynamic matching exercise once, and then put the results
in the cache.  But I wanted to hold off on until such time as there is
evidence that the dynamic approach is too costly.

Very understandable.  I'm a self-admitted premature optimizer, because
I often find it a very fun part of development :-)

Has Solr been profiled to see where the bottlenecks really are?

Only for the query side, where the bottlenecks understandably turned
out to be the Lucene queries themselves.

For the indexing side, there are one or two lucene patches I want to
apply that should have a bigger impact on indexing speed.

Anyway, I was comfortable enough that any possible bottlenecks could
be eliminated in the future that I was already working on committing
your patch.  Unfortunately incompatible changes were made to
IndexSchema in the meantime, so I'm applying it by hand.

-Yonik

Reply via email to