The first interaction I had with GraphQL the developer who introduced tried to make it TOO smart for its own good. It was building complex SQL dynamically which means the SQL that it was running was borderline non-deterministic.
(Aka. identified the foreign keys and primary keys linked them together and made all other fields options)
Things I don't like about GraphQL (In Java which was my experience)
1. Debugging it is annoying. There's no clear concise way to follow the code there seems to be some magic that happens where it isn't clear when certain parts of the code get invoked.
Things I like about it.
1. Lets you make multiple queries and reduce/extend the size of your payload as needed.
In my view graphQL is much better when you're not constrained by a SQL like backend. It's great at filtering the payloads which is great for mobile and such. It also allows you make multiple calls in one go which also means you can shoot yourself in the foot if you over do it.
I will call out that some of this is trauma based from my last experience. Having a more dynamic language that isn't Java may make the experience better but in general everytime we had to update the GraphQL code it was cringy.
Eventually started to gut the dynamic SQL and replacing it with a simple Query and then used GraphQL and trim the response which worked out much better.
Generally the question to ask is how many iterations of an endpoint do you need and is it worth introducing a new technology vs just having a few query parameters to do some filtering.
That being said, I'm now looking at some Query language to work with Neo4J so I'm back at looking at dynamic APIs. (:
> Lets you make multiple queries and reduce/extend the size of your payload as needed.
While this is cool, in theory, I haven't found it to be in practice. If it's an internal API, you can just provide a way for the client to get exactly what they need in a single query. If it's an external API, you have to deal with putting limits in place to keep users from burdening the system with complicated requests. Limits can become complex very quickly.
I am also considering woriking with Neo4j for a project.
I am wondering why not build on top of their Http Api? You can send multiple Cypher statements over Http and correct me if I am totally wrong, but "stealing" the concept of graphql's persistent queries, you could make use of http caching also.
Just to make the conversation easier the use case I have is a network topology so you have things like routers, switches, ports etc. If you take it all the way up to Layer 7 (Application) you can have say web services etc. So you could in theory say link X was cut what is affected?
Anyways... my basic POC was exposing endpoints so I can do things like.
to get a list of all devices but I'm basically just writing custom Cipher code to do that query and the benefits of Neo4J basically go out the door. It still has some interesting Graph features but if all i'm doing is writing a custom endpoint for every use case it's mainly pointless.
You can do a simple POST statement which takes a Neo4J query and execute it with some caching on top of it for sure.
Either ways in order to make Neo4J worth it I need a way to make the queries more dynamic. So right now I'm thinking of:
Or just having a dumb POST /custom/query that maybe only supports read operations. You can add a layer of auth but I'm not a big fan of just having some endpoint that's basically a Pipe to Neo4j. It feels just as bad as saying, typing any SQL here and we'll execute it on the server. If people know what they're doing that's fine..but at that point just setup phpMyAdmin/ pgAdmin. At that point you're trusting folks to know what they're doing and if folks accidently drop Bobby Tables (https://xkcd.com/327/) then it's an accepted risk.
If you just have a proxy to run any Cipher query, you might as well just provide users on Neo4J web instance and let them play there.
Anyways, still in early stages trying to figure out how to best leverage Neo4J
You could give a POST /custom/query with only read capabilities.
Also you can give some custom "helper" operations along.
And certainly you could give them some option to prettify the response.
Because one thing is to query the correct things, another whole thing is how to structure the response according to your needs.
Yeah this was years ago but we had concept of 'hydrated' objects so you'd pass a flag to get back shallow object or the hydrated version that had all the relationship loaded as well.
Some flags for helpers that fetch additional data would be good.
(Aka. identified the foreign keys and primary keys linked them together and made all other fields options)
Things I don't like about GraphQL (In Java which was my experience)
1. Debugging it is annoying. There's no clear concise way to follow the code there seems to be some magic that happens where it isn't clear when certain parts of the code get invoked.
Things I like about it.
1. Lets you make multiple queries and reduce/extend the size of your payload as needed.
In my view graphQL is much better when you're not constrained by a SQL like backend. It's great at filtering the payloads which is great for mobile and such. It also allows you make multiple calls in one go which also means you can shoot yourself in the foot if you over do it.
I will call out that some of this is trauma based from my last experience. Having a more dynamic language that isn't Java may make the experience better but in general everytime we had to update the GraphQL code it was cringy.
Eventually started to gut the dynamic SQL and replacing it with a simple Query and then used GraphQL and trim the response which worked out much better.
Generally the question to ask is how many iterations of an endpoint do you need and is it worth introducing a new technology vs just having a few query parameters to do some filtering.
That being said, I'm now looking at some Query language to work with Neo4J so I'm back at looking at dynamic APIs. (: