[SoftTalker]

They are querying postgres information_schema tables (or their equivalent in the other supported databases) to get table names, column names and types, foreign keys, etc.

I doubt the "single query" makes it immensely faster, but reducing query round-trips is usually time-saving. Might be noticable, depending on what you're comparing it to.

[hackernewds]

does it necessitate foreign keys and schemas being defined beforehand? or is that the "AI magic"?

[croes]

It has to be in defined beforehand.

No AI magic can know the foreign keys from the schema alone and even with data the results would be uncertain.

[scotty79]

If you are naming your foreign key fields consistently for example extras.products_id -> products.id or using some fairly unique data type like GUID for your keys it could be intelligently inferred.

[croes]

Still lots of hits and misses.

Column names could sound like foreign keys on pure chance and the data type doesn't show which other table it's referring to.

[parpfish]

If you’re allowed to query the data you could use the column name/data type to come up with lots of good candidates, and then if then verify by checking whether the actual values correspond (assuming that it’s something like a uuid or username column and not a serial/int column)

[lolive]

Newbie here. How do you express foreign key - primary key relationships when you define your schema, in Postgresql and/or SQLServer ? Is it via an integrity constraint, or can it be just an informative metadata ?

[parpfish]

I’m curious how frequently people actually encounter foreign key constraints in production.

I’m 10 years in and have only ever seen it once