Interesting combination, using Solr as a key/value store.
The article mentions that it is for “smaller” data sets and later says that approximately 200M “records” with reasonable response times.
That is something that gets overlooked in the rush to scale.
There are a lot of interesting data sets that are < 200M "records." The Library of Congress for example has 143 million items in its catalogs.
Perhaps your data set is < the Library of Congress?