Re: dynamic schema modeling and performance

We do know where we want to end up. We've had the application running for a
while using a triple store db. We're looking to move away from the triple
store due to performance issues. Our core concept is that each project can
define a set of entities and their relations. Each entity has a set of
attributes. We need to be able to efficiently run ad-hoc queries across
multiple entities in a project, filtering via the attribute values

I think our business needs probably eliminate the possibility of data
integrity at the db level. We currently do application level data
validation.

Regarding EAV, is there a good way to do this? Most everything I read
online says that EAV is a terrible idea and performance is lack luster.
However there are 6NF advocators who say that done right, it is highly
performant. Are there any articles on the correct way to implement EAV?

would jsonb or eav likely provide better query performance?

On Wed, Apr 12, 2017 at 7:43 AM, Merlin Moncure <mmoncure(at)gmail(dot)com> wrote:

> On Tue, Apr 11, 2017 at 12:46 PM, Rj Ewing <ewing(dot)rj(at)gmail(dot)com> wrote:
> > I'm looking for thoughts on the best way to handle dynamic schemas.
> >
> > The application I am developing revolves around user defined entities.
> Each
> > entity is a tabular dataset with user defined columns and data types.
> > Entities can also be related to each other through Parent-Child
> > relationships. Some entities will be 100% user driven, while others
> (such as
> > an entity representing a photo) will be partially user driven (all photo
> > entities will have common fields + custom user additions).
> >
> > I was hoping to get opinions on whether postgresql would be a suitable
> > backend. A couple of options I have thought of are:
> >
> > 1. Each entity is represented as a table in psql. The schema would be
> > dynamically updated (with limits) when an entity mapping is updated. I
> > believe that this would provide the best data constraints and allow the
> best
> > data normalization. A concern I have is that there could be an enormous
> > amount of tables generated and the performance impacts this might have in
> > the future. I could then run elasticsearch as a denormalized cache for
> > efficient querying and full-text-search.
> >
> > 2. Use a nosql database. This provides the "dynamic" schema aspect. A
> > concern here is the lack of relation support, thus leading to a more
> > denormalized data structure and the potential for the data to become
> > corrupted.
> >
> > Any opinions on the use of psql for this case, or other options would be
> > greatly appreciated!
>
> Postgres can function as a nosql database -- you can use jsonb for
> example to archive data in such a way that the data model can be
> changed without making schema adjustments. Another way to do it is
> EAV pattern as noted. These might be good strategies if you're not
> sure where you want to end up.
>
> It really comes down to this: how formal do you want your data model
> to be? Adding formality leads to performance optimizations, exposes
> your data to the fantastic SQL language, and allows rigorous
> assumptions to made made from external dependencies and trusted.
> Formality also brings a degree of inflexibility since your data has to
> be forced into predetermined structures.
>
> merlin
>