There's a recipe how to sample documents and determine their structure:
https://github.com/arangodb/docs/blob/2c842774b457114c571abdaa8391a038715d1458/3.3/cookbook/accessing-shapes-data.md
It is stated, that you can't query internal shapes data. But examining some documents will only approximate what attribute keys are used, or the entire collection must be scanned.
So my question is: does the database store what attributes exist somewhere internally? At least for common attributes?
If yes, why isn't it possible to query that data? It would be far more efficient than a user-defined function that outputs roughly the same information.
It would be great if one could discover schemes "for free":
http://som-research.uoc.edu/tools/jsonDiscoverer/#/
Whenever an attribute is used first in a collection, ArangoDB will store this somewhere internally. That means it does keep track of which attributes were used in a collection. There are a few issues however:
the attribute names are stored globally, but nested attribute names are stored separately (ex: user.name will be stored as user and name). From looking at purely the separate attribute name parts, ArangoDB will not know in which combinations they are used in the data
attribute names are stored whenever an attribute name is first used in a collection. Currently ArangoDB does not keep track of when an attribute is not used anymore. The attribute name will then still be present in the list of attributes
Under these restrictions, the list of attributes could be made available, but I am not sure how useful this will be.