TrailDB

• Made for Ad Roll/Adroll
• Stores data in partitions (user sessions)
• Exploits similarity of events within the partition to compress them

Its primarily for storing clickstreams from a user session or something which looks close enough to that.

Queries work by writing a script that walks over records in a particular stream, calculates what it needs, and returns it.

I tried this with sales records from Second Life. Took sales notifications from e-mail, indexed with Notmuch, queried, ETL'd to JSON documents, converted to a traildb file.