I tried arangodb, every breach is a collection, then extract the info to edges

e.g. make a breach collection

- b_2021_xyz
  - username
  - password

Then make tow general collection

username
password

Make tow edge

b_2021_xyz - breached -> username
b_2021_xyz - breached -> password

By this way, I can find by a username in multi breach, followed with other infos.

But this redundant too much info, arangodb is based on rocksdb, storage is worse than column based.

Any suggestions on how to model the data ?