Question

So I have an app that if I’m honest doesn’t really need transactional integrity (lots of updates, none of them critical). So I was planning on simply leaving entity groups by the wayside for now. But I’d still like to understand it (coming from a relational background).

The way I see it, all queries for my app will be on a user by user basis. Therefore I do not need to group any higher than a user entity, according to the docs recommendations. But I wasn’t planning on having a specific user entity, instead relying on UserProperty in the entities themselves.

The way I see it, if I want transactions (on a per-user basis), I will need some kind of root user entity as the parent of all entities that are part of the hierarchy of her data, no matter how thin this entity would actually be i.e. basically no properties.

Is this correct?

Apologies for verboseness, I only really pinged what schema-less actually meant in practice tonight…

Answer 1

You are essentially correct. You need to group them if you want transactional capability. However, you can group several entities together without creating an actual root entity, in the sense of an entity in the datastore. You instead create a sort of virtual root entity. One important use case of this feature is the ability to create a child object before you create it parent.

You can create an entity with an
ancestor path without first creating
the parent entity. To do so, you
create a Key for the ancestor using a
kind and key name, then use it as the
parent of the new entity. All entities
with the same root ancestor belong to
the same entity group, whether or not
the root of the path represents an
actual entity.

That quote is from the same doc you linked to.