>So I have to ask what the end goal is here. In our case, the purpose of sharding was/is to speed up the process. We've noticed that as the index size was growing, response speed kept going down so we decided to split the index across 5 machines.
>Are your response times really in need of improvement or is this more trying to understand the process? Our response time went from 1second to 5+ seconds, so we thought we could definitely do better with Solr(Cloud). 'start' and 'rows' are generally set to the default values (i.e., 0 and 10 respectively). Any clues how to conduct further investigations? On Sun, Mar 16, 2014 at 7:29 AM, Erick Erickson <erickerick...@gmail.com>wrote: > I wouldn't expect the merge times to be significant > at all, _assuming_ you're not doing something like > setting a very high &start= parameter or returning > a whole of rows. > > Now, it may be that you're sharding with too small > a document set to really notice a difference. > Sharding isn't really about speeding up responses, > as it is being able to handle very large indexes. > > So I have to ask what the end goal is here. Are > your response times really in need of improvement > or is this more trying to understand the process? > > Best, > Erick > > On Thu, Mar 13, 2014 at 1:19 AM, remi tassing <tassingr...@gmail.com> > wrote: > > Hi Erick, > > > > I've used the fl=id parameter to avoid retrieving the actual documents > > (step <4> in your mail) but the problem still exists. > > Any ideas on how to find the merging time(step <3>)? > > > > Remi > > > > > > On Tue, Mar 11, 2014 at 7:29 PM, Erick Erickson <erickerick...@gmail.com > >wrote: > > > >> In SolrCloud there are a couple of round trips > >> that _may_ be what you're seeing. > >> > >> First, though, the QTime is the time spent > >> querying, it does NOT include assembling > >> the documents from disk for return etc., so > >> bear that in mind.... > >> > >> But here's the sequence as I understand it > >> from the receiving node's viewpoint. > >> 1> send the query out to one replica for > >> each shard > >> 2> get the top N doc IDs and scores ( > >> or whatever sorting criteria) from each > >> shard. > >> 3> Merge the lists and select the top N > >> to return > >> 4> request the actual documents for > >> the top N list from each of the shards > >> 5> return the list. > >> > >> So as you can see, there's an extra > >> round trip to each shard to get the > >> full document. Perhaps this is what > >> you're seeing? <4> seems like it > >> might be what you're seeing, I don't > >> think it's counted in QTime. > >> > >> HTH > >> Erick > >> > >> On Tue, Mar 11, 2014 at 3:17 AM, remi tassing <tassingr...@gmail.com> > >> wrote: > >> > Hi, > >> > > >> > I've just setup a SolrCloud with Tomcat. 5 Shards with one replication > >> each > >> > and total 10million docs (evenly distributed). > >> > > >> > I've noticed the query response time is faster than using one single > node > >> > but still not as fast as I expected. > >> > > >> > After turning debugQuery on, I noticed the query time is different to > the > >> > value returned in the debug explanation (see some excerpt below). More > >> > importantly, while making a query to one, and only one, shard then the > >> > result is consistent. It appears the server spends most of its time > doing > >> > result aggregation (merging). > >> > > >> > After searching on Google in vain I didn't find anything concrete > except > >> > that the problem could be in 'SearchComponent'. > >> > > >> > Could you point me in the right direction (e.g. configuration...)? > >> > > >> > Thanks! > >> > > >> > Remi > >> > > >> > Solr Cloud result: > >> > > >> > <lst name="responseHeader"> > >> > > >> > <int name="status">0</int> > >> > > >> > <int name="QTime">3471</int> > >> > > >> > <lst name="params"> > >> > > >> > <str name="debugQuery">on</str> > >> > > >> > <str name="q">project development agile</str> > >> > > >> > </lst> > >> > > >> > </lst> > >> > > >> > <result name="response" numFound="2762803" start="0" > >> > maxScore="0.17022902">...</result> > >> > > >> > ... > >> > > >> > > >> > > >> > <lst name="timing"> > >> > > >> > <double name="time">508.0</double> > >> > > >> > <lst name="prepare"> > >> > > >> > <double name="time">8.0</double> > >> > > >> > <lst name="query"> > >> > > >> > <double name="time">8.0</double> > >> > > >> > </lst> > >> > > >> > <lst name="facet"> > >> > > >> > <double name="time">0.0</double> > >> > > >> > </lst> > >> > > >> > <lst name="mlt"> > >> > > >> > <double name="time">0.0</double> > >> > > >> > </lst> > >> > > >> > <lst name="highlight"> > >> > > >> > <double name="time">0.0</double> > >> > > >> > </lst> > >> > > >> > <lst name="stats"> > >> > > >> > <double name="time">0.0</double> > >> > > >> > </lst> > >> > > >> > <lst name="debug"> > >> > > >> > <double name="time">0.0</double> > >> > > >> > </lst> > >> > > >> > </lst> > >> > > >> > <lst name="process"> > >> > > >> > <double name="time">499.0</double> > >> > > >> > <lst name="query"> > >> > > >> > <double name="time">195.0</double> > >> > > >> > </lst> > >> > > >> > <lst name="facet"> > >> > > >> > <double name="time">0.0</double> > >> > > >> > </lst> > >> > > >> > <lst name="mlt"> > >> > > >> > <double name="time">0.0</double> > >> > > >> > </lst> > >> > > >> > <lst name="highlight"> > >> > > >> > <double name="time">228.0</double> > >> > > >> > </lst> > >> > > >> > <lst name="stats"> > >> > > >> > <double name="time">0.0</double> > >> > > >> > </lst> > >> > > >> > <lst name="debug"> > >> > > >> > <double name="time">76.0</double> > >> > > >> > </lst> > >> > > >> > </lst> > >> > > >> > </lst> > >> >