Today I have released the next major version of rspamd - 0.8.0. The main difference from the 0.7 branch is completely reworked fuzzy storage. I have switched the storage itself from own memory-based hashes structure to the sqlite3 database and redesigned protocol for future extensions and new features. At the same time, I have preserved backward compatibility with previous rspamd versions, therefore no specific upgrade movements are needed. Moreover, the conversion of an old database format is performed automatically and does not require special attention as well.
Fuzzy check plugin has been reworked accordingly. First of all, I have changed the algorithm of fuzzy hashes to the probabilistic shingles algorithm. It is blazingly fast and still rather accurate to find close texts in the database. Secondly, I have added the normalization algorithm for the target language using snowball lemmatizer. It allows to remove grammar forms and check merely the first forms of all words in a text improving thus fuzzy matching quality by removing meaningless parts.
Rspamd 0.8 is heavily tested in production environments and I consider it as a production ready release. No manual migration is required for rspamd 0.7 users, however, if you use more old versions of rspamd then you should check the migration guide.
As usually, please feel free to ask any questions in rspamd mailing list or IRC discussion channel (#rspamd at freenode).