[mildbyte]

We use them in production at Splitgraph [0] to power our DDN (like a CDN, but for data). We make a PostgreSQL-compatible endpoint available to the public to query any of the tens of thousands of open datasets by referencing them as virtual tables: they're not hosted by us but we proxy to them using Postgres FDWs. When a query comes in, we intercept it and redirect it to a FDW instance that handles query translation and planning from the PG dialect to that of the backend data source.

We wrote an FDW for Socrata-powered [1] government open data portals to query the public datasets that we index in the Splitgraph catalog as a proof-of-concept. However, there are plenty of other FDWs that we're working on integrating to let people add their own backend data sources (RDS, Snowflake etc).

FDW plugin quality varies (some of them can't push down all predicates or JOINs) but it's definitely an interesting way to think about accessing data. We also added a lot of scaffolding around foreign data wrappers in our open-source tool [2] that makes it easy to add a FDW-managed data source to a PostgreSQL instance.

[0] https://www.splitgraph.com/blog/data-delivery-network-launch

[1] https://www.tylertech.com/products/socrata

[2] https://www.splitgraph.com/blog/foreign-data-wrappers