On Mon, Mar 12, 2012 at 5:39 PM, Mark Miller <markrmil...@gmail.com> wrote: > > On Mar 12, 2012, at 5:32 PM, Yonik Seeley wrote: > >> On Mon, Mar 12, 2012 at 8:15 AM, Mark Miller <markrmil...@gmail.com> wrote: >>> Currently it doesn't send directly to the leader, but this is planned - >>> it's a little tricky due to lack of access to the Schema for hashing >> >> Hmmm, why is this? Identification of the "uniqueKey" field? Maybe we >> just assume "id", or let the user configure it if it's something >> different. That should really be "best practice" along with sticking >> to normal java identifiers for field names. > > Yeah, for id my plan was just let the user supply the field, perhaps default > to id. The other issue is that we hash on the indexed value though - which we > get though a customizable field type method impl last I recall. I think this > tends to be the same as the raw text for the types we care about. But we have > to make some assumptions - it's not really arbitrary support - though it > should easily cover the current common types of numeric or string. I think > most impls end up using UnicodeUtil.UTF16toUTF8 and hopefully most toInternal > methods simply return what is passed in (ie use the base class impl)...
Non "string" (or compatible) ids are more trouble than they are worth... and since cloud is new, I think it would be fine to say "use a string id". -Yonik lucenerevolution.com - Lucene/Solr Open Source Search Conference. Boston May 7-10 --------------------------------------------------------------------- To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
