This makes no sense at all to me. The history of this plugin is that it was developed a long time ago for Solr 3.x. At that time the flow most certainly did not involve a separate query or query clause for every token in the query. But over time the way this integrates into Solr may very well have changed. But this still seems bizarre to me.
If you can find a more appropriate integration point that would be great. Specifically we need a place that is called ONCE per query, not per token. Ideally this integration point should allow us to be involved in building the query as well. Karl On Fri, Oct 18, 2019 at 10:56 AM Julien Massiera < julien.massi...@francelabs.com> wrote: > Hi Karl, > > I recently had to use the ManifoldCFQParserPlugin to filter the results of > a Solr spellchecker and I noticed a non-neglectable increase of the > response time of the spellchecker at this point. > > After some investigation, I noticed that the query parser is not called > just once, but for each terms suggested by the spellchecker. As a result, > even if the authority servlet called has a cache, the time needed to > retrieve the response is multiplied by the number of suggested terms. This > is what generates the time performance issue that I mentioned. > > So I have two questions : > > - The ManifoldCFQParserPlugin is called for each suggested term. But is it > somehow linked to the boolean type of filter generated by the query parser > ? If yes, what sort of query parser plugin do we need to use for this use > case ? > - Would it sound like a good idea to implement a cache system on the query > parser side to address this performance issue ? > > > Thanks for your help > > Regards > > -- > Julien MASSIERA > >
