View the original community article here
Last tested: Feb 4, 2019
HyperSQL is an in-memory database. That basically means it needs to be loaded into memory to do its thing (store configuration, users, and other metadata). More specifically, the instructions to create the database (DDL, i.e.: data definition language) are persisted to disk. At startup, the script loads this into main memory and constructs the database in-memory.
On a busy instance this database can grow in size, which leads to performance issues, Java memory pressure, and extremely long startup times.
A concrete example is if more than the recommended 60% of main memory is allocated to the Looker java process, then the internal backend might be pinched to acquire the amount of main memory it needs to do its thing.