I am trying to use the SOLR LTR plugin to train a machine learning model for my application. One of the features I require for my model is edismax score of my query over a set of index fields.
The way I have created the feature is as follows: { "store":"sampleFeatureStore", "name" : "edismax_feature", "class" : "org.apache.solr.ltr.feature.SolrFeature", "params" : { "q" : "{!edismax qf=\"field1 field2 field3 field4\" }${user_query}" } } field1 ... field4 are text fields with some analyzers performing tokenization, lowercasing etc. While this feature gives me the edismax scores I desire, however, on load testing my entire setup, presence of these features increased latency conderably and I could see timeouts in the logs for many queries. (The request took too long to iterate over terms) I also saw my CPU usage and JVM memory use increase. Without this feature things were stable and load testing was fine. I have 2 collections in my solr setup with 25M and 2M docs respectively. Load testing involved sending 200 queries per minute. -- View this message in context: http://lucene.472066.n3.nabble.com/solr-LTR-edismax-features-very-slow-and-cause-timeouts-tp4335059.html Sent from the Solr - User mailing list archive at Nabble.com.