When and How to use GraphQL with microservice architecture

I'm trying to understand where GraphQL is most suitable to use within a microservice architecture.

There is some debate about having only 1 GraphQL schema that works as API Gateway proxying the request to the targeted microservices and coercing their response. Microservices still would use REST / Thrift protocol for communication though.

Another approach is instead to have multiple GraphQL schemas one per microservice. Having a smaller API Gateway server that route the request to the targeted microservice with all the information of the request + the GraphQL query.

1st Approach

Having 1 GraphQL Schema as an API Gateway will have a downside where every time you change your microservice contract input/output, we have to change the GraphQL Schema accordingly on the API Gateway Side.

2nd Approach

If using Multiple GraphQL Schema per microservices, make sense in a way because GraphQL enforces a schema definition, and the consumer will need to respect input/output given from the microservice.

Questions

• Where do you find GraphQL the right fit for designing microservice architecture?

• How would you design an API Gateway with a possible GraphQL implementation?

Definitely approach #1.

Having your clients talk to multiple GraphQL services (as in approach #2) entirely defeats the purpose of using GraphQL in the first place, which is to provide a schema over your entire application data to allow fetching it in a single roundtrip.

Having a shared nothing architecture might seem reasonable from the microservices perspective, but for your client-side code it is an absolute nightmare, because every time you change one of your microservices, you have to update all of your clients. You will definitely regret that.

GraphQL and microservices are a perfect fit, because GraphQL hides the fact that you have a microservice architecture from the clients. From a backend perspective, you want to split everything into microservices, but from a frontend perspective, you would like all your data to come from a single API. Using GraphQL is the best way I know of that lets you do both. It lets you split up your backend into microservices, while still providing a single API to all your application, and allowing joins across data from different services.

If you don't want to use REST for your microservices, you can of course have each of them have its own GraphQL API, but you should still have an API gateway. The reason people use API gateways is to make it more manageable to call microservices from client applications, not because it fits well into the microservices pattern.

This article recommends approach #1. See the below image too, taken from the mentioned article:

One of the main benefits of having everything behind a single endpoint is that data can be routed more effectively than if each request had its own service. While this is the often touted value of GraphQL, a reduction in complexity and service creep, the resultant data structure also allows data ownership to be extremely well defined, and clearly delineated.

Another benefit of adopting GraphQL is the fact that you can fundamentally assert greater control over the data loading process. Because the process for data loaders goes into its own endpoint, you can either honor the request partially, fully, or with caveats, and thereby control in an extremely granular way how data is transferred.

The following article explains these two benefits along with others very well: https://nordicapis.com/7-unique-benefits-of-using-graphql-in-microservices/

As of mid 2019 the solution for the 1st Approach has now the name "Schema Federation" coined by the Apollo people (Previously this was often referred to as GraphQL stitching). They also propose the modules @apollo/federation and @apollo/gateway for this.

ADD: Please note that with Schema Federation you can't modify the schema at the gateway level. So for every bit you need in your schema, you need to have a separate service.

For approach #2, in fact that's the way I choose, because it's much easier than maintaining the annoying API gateway manually. With this way you can develop your services independently. Make life much easier :P

There are some great tools to combine schemas into one, e.g. graphql-weaver and apollo's graphql-tools, I'm using graphql-weaver, it's easy to use and works great.

I have been working with GraphQL and microservices

Based on my experience what works for me is a combination of both approaches depending on the functionality/usage, I will never have a single gateway as in approach 1, but neither a GraphQL for each microservice as approach 2.

For example based on the image of the answer from Enayat, what I would do in this case is to have 3 graph gateways (Not 5 as in the image)

• App (Product, Basket, Shipping, Inventory, needed/linked to other services)

• Payment

• User

This way you need to put extra attention to the design of the needed/linked minimal data exposed from the depending services, like an auth token, userid, paymentid, payment status

In my experience for example, I have the "User" gateway, in that GraphQL I have the user queries/mutations, login, sign in, sign out, change password, recover email, confirm email, delete account, edit profile, upload picture, etc... this graph on it own is quite large!, it is separated because at the end the other services/gateways only cares about the resulting info like userid, name or token.

This way is easier to...

• Scale/shutdown the different gateways nodes depending on they usage. (for example people might not always be editing their profile or paying... but searching products might be used more frequently).

• Once a gateways matures, grows, usage is known or you have more expertise on the domain you can identify which are the part of the schema that could have they own gateway (... happened to me with a huge schema that interacts with git repositories, I separated the gateway that interact with a repository and I saw that the only input needed/linked info was... the folder path and expected branch)

• The history of you repositories is more clear and you can have a repository/developer/team dedicated to a gateway and its involved microservices.

UPDATE:

I have a kubernetes cluster online that is using the same approach that I describe here with all the backends using GraphQL, all opensource, here is the main repository: https://github.com/vicjicaman/microservice-realm

This is an update to my answer because I think that it is better if the answer/approach is backed up code that is running and can be consulted/reviewed, I hope that this helps.

To question 1, Intuit acknowledged the power of GraphQL few years back when it announced moving to One Intuit API ecosystem (https://www.slideshare.net/IntuitDeveloper/building-the-next-generation-of-quickbooks-app-integrations-quickbooks-connect-2017). Intuit chose to go with approach 1. The drawback that you mention actually prevents developers from introducing breaking schema changes that could potentially disrupt client applications.

GraphQL has helped improve productivity of developers in a number of ways.

1. When designing a new microservice for a domain, engineers (backend/ frontend/ stakeholders) agree on a schema for the domain entities. Once the schema is approved, it is merged with the master domain schema (universal) and deployed on the Gateway.Front end engineers can start coding client applications with this schema while backend engineers implement the functionality. Having one universal schema means that there are no two microservices with redundant functionality.

2. GraphQL has helped client applications become simpler and faster. Want to retrieve data from /update data to multiple microservices? All client applications have to do is fire ONE GraphQL request and the API Gateway abstraction layer will take care to fetch and collate data from multiple sources (microservices). Open source frameworks like Apollo (https://www.apollographql.com/) have accelerated the pace of GraphQL adoption.

3. With mobile being the first choice for modern applications, it is important to design for lower data bandwidth requirements from ground zero. GraphQL helps by allowing client apps to request for specific fields only.

To question 2: We built a custom abstraction layer at the API Gateway that knows which part of the schema is owned by which service (provider). When a query request arrives, the abstraction layer forwards the request to the appropriate service(s). Once the underlying service returns the response, the abstraction layer is responsible to return the requested fields.

However, today there are several platforms out there (Apollo server, graphql-yoga, etc.) that allow one to build a GraphQL abstraction layer in no time.

As of 2019 the best way is to write microservises that implements apollo gateway specification and then glue together these services using a gateway following approach #1. The fastest way to build the gateway is a docker image like this one Then use docker-compose to start all the services concurrently:

version: '3'

services:
    service1:
        build: service1
    service2:
        build: service2
    gateway:
        ports:
            - 80:80
        image: xmorse/apollo-federation-gateway
        environment: 
            - CACHE_MAX_AGE=5
            - "FORWARD_HEADERS=Authorization, X-Custom-Header" # default is Authorization, pass '' to reset
            - URL_0=http://service1
            - URL_1=http://service2

The way it is being described in this question, I believe that using a custom API gateway as an orchestration service can make a lot of sense for complex enterprise focused applications. GraphQL can be a good technology choice for that orchestration service, at least as far as querying goes. The advantage to your first approach (one schema for all microservices) is the ability to stitch together the data from multiple microservices in a single request. That may, or may not, be very important depending on your situation. If the GUI calls for rendering data from multiple microservices all at once, then this approach can simplify the client code such that a single call can return data that is suitable for data binding with the GUI elements of such frameworks as Angular or React. This advantage doesn't apply for mutations.

The disadvantage is tight coupling between the data APIs and the orchestration service. Releases can no longer be atomic. If you refrain from introducing backwards breaking changes in your data APIs, then this can introduce complexity only when rolling back a release. For example, if you are about to release new versions of two data APIs with the corresponding changes in the orchestration service and you need to roll one of those releases back but not the other, then you will be forced to roll back all three anyway.

In this comparison of GraphQL vs REST you will find that GraphQL is not quite as efficient as RESTful APIs so I would not recommend replacing REST with GraphQL for the data APIs.

We also had similar concerns on the bootstraping a Microservices ecosystem with graphql. And we are able to solve it with Apollo GraphQL.

We built these solutions from scratch for our One Platform. Folders ending with service are microservices others are SPAs aka (Single Page Applications)

This consists of various microservices and spas with an API Gateway which is the primary interface for multiple microservices.

Along with this you can find a OP CLI generator which can help to bootstrap a microservice from scratch.

Project - https://github.com/1-Platform/one-platform

I request to please have a look on this project and feel free to adopt if this looks good to you.

Regards

Rigin Oommen