Hi, I just noticed, when looking at some performance problems with distributed search using huge stored fields, that Solr by default fetches the Solr (external) ID when executing the first step of the search and then transmit those IDs to the node that executes the search. This requires that the whole stored field block is decompressed.
I just enabled DocValues and disabled stored fields and distributed search (the first part) was much faster, because the IDs were available asap. I also noticed that enabling only docvalues but not disabling the stored flag of the ID field did not help (as the stored field is preferred). Wouldn't it be a good idea to prefer an existent docvalues field for distributed/cloud search when doing the first search step (collecting IDs only)? I'd also suggest to enable docvalues for the ID field by default. At least the 6.5.1 version of Solr does not have this by default. Maybe that’s a good change for Master/7.0. Uwe ----- Uwe Schindler Achterdiek 19, D-28357 Bremen http://www.thetaphi.de eMail: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
