Hello! I am looking to design a technology independent 

#dataCatalog

 for heterogeneous

#datamesh

-es, 

#datalake

-s and `#featureStore`s Found this `GraphQL`/`SHACL` extension as

data schema

language to be very interesting (first tweet below). And would also like to build on top of

DataHub

. In this post https://engineering.linkedin.com/blog/2019/data-hub it says

_... plan to add native support for GraphQL and leverage Pegasus Domain Specific Language (PDL)_

so the

DataHub

project is already considering

GraphQL

🤔 Is there a design channel best suited to discuss how to maybe combine the two ? Please see a series of tweets for better context https://twitter.com/semanticbeeng/status/1338395121344196614 https://twitter.com/semanticbeeng/status/1340659039160680448 https://twitter.com/semanticbeeng/status/1340341801245433858 https://twitter.com/semanticbeeng/status/1340213178735742976 https://twitter.com/semanticbeeng/status/1338165535067017216 /cc @mysterious-waiter-64784 @adorable-processor-445

A contributor, @big-carpet-38439, from @mammoth-bear-12532’s new company, has put together a GraphQL RFC just the other day!

Thanks @microscopic-receptionist-23548 ! That’s right - this RFC assumes that frontend models are still managed independently of the GMA models! However, the ability to auto generate frontend models from Pegasus models is something we want to explore, soon!

Just taking a first look at PDL Schema, but it looks reasonably compatible with GraphQL SDL. E.g. I see that it supports untagged unions, similar to GraphQL and Avro. A mapping between the two probably wouldn't be too lossy.

Yes @dazzling-traffic-87612, we actually have a project at LinkedIn that translates PDL to GraphQL .. @big-carpet-38439 works on that project

Nice... anything I can look at?

@witty-kilobyte-6731 : just created #datahub-modeling

Thanks! Let us continue here please https://datahubspace.slack.com/archives/C01HPV8EKTK/p1608499109001600

@dazzling-traffic-87612 unfortunately nothing formally published around this today — in the case of PDL we do not ‘stash’ divergent features, as you pointed out the expressivity in the 2 type systems are extremely close... we deal with any minor disparities via manually enforced governance :( (ie never use tagged unions in PDL)... I like the Dragon approach of essentially ‘implementing’ the more abstract dragon spec in different ways across type systems - either using native constructs in the case where we will encounter no loss or additional metadata in cases where we will

Sure, that is another way to go, and to a certain extent we do the same thing, e.g. there are structures which Dragon supports but which are not supported in a given target language and for which we have not added a stashing strategy. We do have a way of specifying which combinations of features are supported in which languages, so we can either stash or fail early.

Josh, please let us discuss on the channel 🙂