On 6/5/2013 11:25 PM, TwoFirst TwoLast wrote: > 1) If I change one field's type in my schema, will that cause problems with > the index or searching? My data is pulled in chunks off of a mysql server > so one field in the currently indexed data is simply an "int" type field in > solr. I would like to change this to a string moving forward, but still > expect to search across the int/string field. Will this be ok? > > 2) My motivation for #1 is that I have thousands of records that are > exactly the same in mysql aside from a user_id column. Prior to inserting > into mysql I am thinking that I can concatenate the user_ids together into > a space separated string and let solr just parse the string. So the > database and my data import handler would change a bit. > > 3) If #2 is an appropriate approach, will a solr.TextField with > a solr.WhitespaceTokenizerFactory be an ok way to approach this? This does > produce words where I would expect integers. I tried using a > solr.TrieIntField with the solr.WhitespaceTokenizerFactory, but it throws > an error.
If you change the field type, you have to completely reindex or Solr will not work correctly. This is true of most changes in schema.xml. A field that was indexed as an int will not work correctly if you later tell Solr that the field is a string. The underlying Lucene index is schema-less and has no idea what high-level data type each field is using. Solr reads and writes a particular data representation on each field according to schema.xml. Using an analysis chain (tokenizers and filters) is only supported on a TextField. It will not work on any other type, including StrField and TrieIntField. Thanks, Shawn
