lucene-solr-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Markus Jelsma <markus.jel...@openindex.io>
Subject RE: Results not ordered by score and debug info is incorrect, crazy
Date Tue, 27 Sep 2016 14:15:00 GMT
Ok, thanks. I've added SOLR-9573

https://issues.apache.org/jira/browse/SOLR-9573

Regards,
Markus
 
 
-----Original message-----
> From:Shalin Shekhar Mangar <shalinmangar@gmail.com>
> Sent: Tuesday 27th September 2016 16:04
> To: solr-user@lucene.apache.org
> Subject: Re: Results not ordered by score and debug info is incorrect, crazy
> 
> Wow, it took me some time to realize what you were referring to :-)
> 
> The manual (or the reference guide page) is under construction and I didn't
> finish that page. The sentence about faceting is just a copy/paste relic :)
> I'll go fix that asap.
> 
> I remember a bunch of Jira issues w.r.t NPE and distrib.singlePass which
> have since been fixed. If you find something else, then please open a
> issue. The distrib.singlePass approach reads all the stored fields from all
> shards for all documents instead of just the top N docs. This can be
> advantageous if you are returning a small number of fields but may cause
> performance degradation if you are requesting all or a lot of fields.
> 
> But coming back to this bug, it seems like debugQuery=true is behaving
> incorrectly for the normal two phase search? Please open a jira issue. This
> warrants more investigation.
> 
> On Tue, Sep 27, 2016 at 6:26 PM, Markus Jelsma <markus.jelsma@openindex.io>
> wrote:
> 
> > Shalin, that does the trick indeed!
> >
> > Also noted that the manual is incorrect, setting it to a blank or false
> > value, did not disable my facets:
> > If set to "true," this parameter enables facet counts in the query
> > response. If set to "false" to a blank or missing value, this parameter
> > disables faceting. None of the other parameters listed below will have any
> > effect unless this parameter is set to "true." The default value is blank.
> >
> > In the past, i do remember some discussions about this parameter, and
> > vaguely recall there were some drawbacks for this parameters, so i didn't
> > enable it.
> >
> > The manual is not clear about this though, any recommendations?
> >
> > Thanks,
> > Markus
> >
> > -----Original message-----
> > > From:Shalin Shekhar Mangar <shalinmangar@gmail.com>
> > > Sent: Tuesday 27th September 2016 14:41
> > > To: solr-user@lucene.apache.org
> > > Subject: Re: Results not ordered by score and debug info is incorrect,
> > crazy
> > >
> > > This may be relevant or not, I am not sure but one difference between
> > > fl=title_nl,score,id and fl=score,id is that the former executes a two
> > pass
> > > distributed search i.e. get ids, merge, get fields for top N docs but the
> > > latter skips the "get fields" phase because it already has all the right
> > > information.
> > >
> > > You can force the single pass for the first request as well by adding
> > > distrib.singlePass=true as a request parameter. It might be interesting
> > to
> > > get that output as well and compare it with the others.
> > >
> > > On Tue, Sep 27, 2016 at 5:39 PM, Markus Jelsma <
> > markus.jelsma@openindex.io>
> > > wrote:
> > >
> > > > Hi,
> > > >
> > > > I just spotted something weird, again. A regular search popped up a
> > weird
> > > > candidate for first result, so i've reproduced it on our production
> > system.
> > > >
> > > > Digging deeper, it appears that the fl parameter has something to do
> > with
> > > > it. Not the order of results but the scores / explain in the debug
> > section
> > > > are different. This means i am not sure in what case the debugging
> > > > information is really correcty.
> > > >
> > > > Check this out. These are the top four results with
> > > > fl=title_nl,score,id&hl=false&facet=false&debugQuery=true&
> > > > wt=xml&indent=true&q=misofonie&rows=4
> > > >
> > > > <result name="response" numFound="26" start="0" maxScore="153.99518">
> > > >   <doc>
> > > >     <str name="id">http://www.gezondheidenco.nl/179901/
> > > > website-over-misofonie-online/</str>
> > > >     <str name="title_nl">Website over misofonie online</str>
> > > >     <float name="score">153.99518</float></doc>
> > > >   <doc>
> > > >     <str name="id">http://www.gezondheidenco.nl/tag/misofonie/</str>
> > > >     <str name="title_nl">misofonie Archives - Gezondheid &amp;
Co</str>
> > > >     <float name="score">149.88002</float></doc>
> > > >   <doc>
> > > >     <str name="id">http://www.gezondheidenco.nl/263802/
> > > > misofonie-psychische-aandoening-is-behandelen/</str>
> > > >     <str name="title_nl">Misofonie: een psychische aandoening. Is
het
> > te
> > > > behandelen? | Gezondheid &amp; Co</str>
> > > >     <float name="score">147.82176</float></doc>
> > > >   <doc>
> > > >     <str name="id">http://www.gezondheidenco.nl/tag/
> > > > ereniging-misofonie-nl/</str>
> > > >     <str name="title_nl">ereniging Misofonie NL Archives - Gezondheid
> > > > &amp; Co</str>
> > > >     <float name="score">144.10309</float></doc>
> > > > </result>
> > > >
> > > > And here's the truncated explain section for above request, notice
> > that it
> > > > is not ordered correctly and the scores are different than above:
> > > >
> > > >     <str name="http://www.gezondheidenco.nl/179901/
> > > > website-over-misofonie-online/">
> > > > 148.65146 = sum of:
> > > >   148.65146 = max plus 0.65 times others of:
> > > >     <str name="http://www.gezondheidenco.nl/tag/misofonie/">
> > > > 144.63885 = sum of:
> > > >   144.63885 = max plus 0.65 times others of:
> > > >     <str name="http://www.gezondheidenco.nl/263802/
> > misofonie-psychische-
> > > > aandoening-is-behandelen/">
> > > > 152.7493 = sum of:
> > > >   152.7493 = max plus 0.65 times others of:
> > > >     <str name="http://www.gezondheidenco.nl/tag/
> > ereniging-misofonie-nl/">
> > > > 139.05501 = sum of:
> > > >   139.05501 = max plus 0.65 times others of:
> > > >
> > > > Here's the other request with only the fl-parameter having a different
> > > > value: fl=score,id&hl=false&facet=false&debugQuery=true&wt=xml&
> > > > indent=true&q=misofonie&rows=4
> > > >
> > > > <result name="response" numFound="26" start="0" maxScore="153.99518">
> > > >   <doc>
> > > >     <str name="id">http://www.gezondheidenco.nl/179901/
> > > > website-over-misofonie-online/</str>
> > > >     <float name="score">153.99518</float></doc>
> > > >   <doc>
> > > >     <str name="id">http://www.gezondheidenco.nl/tag/misofonie/</str>
> > > >     <float name="score">149.88002</float></doc>
> > > >   <doc>
> > > >     <str name="id">http://www.gezondheidenco.nl/263802/
> > > > misofonie-psychische-aandoening-is-behandelen/</str>
> > > >     <float name="score">147.82176</float></doc>
> > > >   <doc>
> > > >     <str name="id">http://www.gezondheidenco.nl/tag/
> > > > ereniging-misofonie-nl/</str>
> > > >     <float name="score">144.10309</float></doc>
> > > > </result>
> > > >
> > > > And here are the scores for this query, notice these are ordered
> > > > descending, the ones above are not, also notice the scores are
> > identical to
> > > > that of the result set itself.
> > > >
> > > >     <str name="http://www.gezondheidenco.nl/179901/
> > > > website-over-misofonie-online/">
> > > > 153.9952 = sum of:
> > > >   153.9952 = max plus 0.65 times others of:
> > > >     <str name="http://www.gezondheidenco.nl/tag/misofonie/">
> > > > 149.88002 = sum of:
> > > >   149.88002 = max plus 0.65 times others of:
> > > >     <str name="http://www.gezondheidenco.nl/263802/
> > misofonie-psychische-
> > > > aandoening-is-behandelen/">
> > > > 147.82175 = sum of:
> > > >   147.82175 = max plus 0.65 times others of:
> > > >     <str name="http://www.gezondheidenco.nl/tag/
> > ereniging-misofonie-nl/">
> > > > 144.10309 = sum of:
> > > >   144.10309 = max plus 0.65 times others of:
> > > >
> > > > I am completely surprised to see this crazy behaviour. But i vaguely
> > > > remeber something about this before, but cannot find references.
> > > >
> > > > Any idea?
> > > >
> > > > Thanks,
> > > > Markus
> > > >
> > >
> > >
> > >
> > > --
> > > Regards,
> > > Shalin Shekhar Mangar.
> > >
> >
> 
> 
> 
> -- 
> Regards,
> Shalin Shekhar Mangar.
> 

Mime
View raw message