As for the question about different weights, down the page of this article there's an explanation of why stats are different on different replicas in the same shard: https://support.lucidworks.com/hc/en-us/articles/115000888308-Getting-different-results-while-issuing-a-query-multiple-times-in-SolrCloud-
A very similar argument holds for two shards with identical documents. Don't have a good idea for your second question. Best, Erick On Tue, Jan 10, 2017 at 4:14 AM, Nicole Bilić <nicole.bil...@gmail.com> wrote: > Hi all, > > We are using Suggester (and Solr 6.3.0) to implement autocomplete. We are > using TSTLookupFactory lookup implementation and > HighFrequencyDictionaryFactory dictionary implementation. If our index > consists of only one shard, everything works perfectly fine. However, when > index is split in 2 shards, we start getting duplicated suggestions. We've > done some testing and it is peculiar that even when we have exactly the > same documents in both shards (just for testing purposes), we get > duplicated terms, but their weights are slightly different. We haven't > found any solution that works (we've tried FieldCollapsing as well). Due to > some limitations, it is not an option to solve this (ie. throw out the > duplicates) client-side. Is there a way to solve this issue solr-side? > > Thanks!