Comment by marcus_holmes

I ran a team a few years ago. The FE folks really wanted to use GraphQL, and the BE folks agreed, because someone had found an interesting library that made it easy. No-one had any experience of GraphQL before.

After a month's development I found out that there was one GraphQL call at the root of each React page, and it fetched all the data for that userID in a big JSON blob, that was then parsed into a JS object and used for the rest of the life of that page. Any updates sent the entire, modified, blob back to the server and the BE updated all the tables with the changed data. This didn't cause problems because users didn't share data or depend on shared data.

Everyone was happy because they got to put GraphQL on their resume. The application worked. We hit the required deadline. The company didn't get any traction with the application and we pivoted to something else very quickly, and was sold to private equity within two years. None of the code we wrote is running now, which is probably a good thing.

I get the feeling, from conversations with other people using GraphQL, that this is the sort of thing that actually happens in practice. The author's arguments make sense, as do the folks defending GraphQL. But I'd suggest that 80-90% of the GraphQL actually written and running out there is the kind of crap my team turned out.