This blog is a part of a series on GraphQL where we will dive deep into GraphQL and its ecosystem one piece at a time
What an interesting journey has it been so far! We explored some of the amazing libraries, tools and frameworks which really empowers GraphQL to be what it is today with almost everything being open source created with love from the community. But, I do understand that, this can actually be overwhelming for some of you who are just starting off this journey with GraphQL and may have some trouble putting it all together to work for you.
To address this, we will be talking about the workflow with GraphQL and the tools we have looked at so far and the process of taking it from development to production in this blog.
NOTE: While these steps have been ordered serially, that is just to give you a sense of understanding of the workflow. Some of these can also be deferred for later or parallelly done if working with multiple teams. So, with that in mind, let’s start.
Step 1: Evaluation and Research
As we have discussed before in this blog post, GraphQL may not be the solution to every problem. And even if it does satisfy your use case well, this would be the first thing you would need to do. Understand why you want to use GraphQL, how you would like to use it and the ecosystem of tools you need to solve your problem. This can be done only if you introspect your use case and really get to the basics answering few obvious questions around GraphQL and your use case.
Also, you have to remember that not every organization is operating at the scale of Google, Microsoft or Facebook and what works for them need not work for you. So, while you should definitely be informed on how people do things, do remember that you need to focus on what works for you and what you really need.
Step 2: Get a boilerplate stack ready
GraphQL can get overwhelming if you are re-inventing the wheel every time. From putting together your schema, resolvers, server, to the various tools you would typically use with it like linting, codegen and so on. Now, doing this every time you work on a new service is not a good use of your time.
The best way to avoid this is to put together a boilerplate with all that you would typically want to use and this can become a starting point of all the services you may develop in the future. This would also involve things like setting up your GraphQL Gateway (incase you are using something like Federation or Stitching) since the gateway becomes the single point of contact for all your requests from the clients.
While you can definitely iterate with this as you go along, having the barebones when you start can definitely take a lot of effort away and save a lot of time in the future.
Step 3: Putting together the data graph and documentation
All that you do with GraphQL for your use case mainly revolves around your schema and its types since that becomes the base of everything you would develop on top. Getting your data graph ready would typically be the next important step and the way you do it does not matter. You can either go the SDL-first or code-first route depending on what works well for you.
You might also want to write appropriate documentation parallelly as you work on your schema, especially since GraphQL is self-documenting and it is always good to do it when you have a context of what you are doing rather than as an after-thought.
Now, if you are working on a microservices architecture and you are looking to split the data graph into multiple parts to be composed or stitched from multiple services, using something like Federation or Stitching, you would also need to understand the clear boundaries of the microservices and how all of them relate to each other through the data graph.
These boundaries will also decide which service hosts your resolvers/logic to go along with resolving the various fields in the schema and performing the business logic as needed in isolation.
Step 4: Deploying it all as per the need
Now that you have your boilerplate and data graph ready, the next step you would typically do before working on your resolvers or any of your business logic is to actually deploy it all wherever you want to, and the way you are looking to do it. Be it public cloud, private cloud or on-premise as containers, VMs or bare metal.
Doing this will help you proceed forward as per your architecture be it single/multi-tenant and help you resolve the major questions you might have regarding the end-end flow of data considering all the compliance policies and laws you might want to cater to.
Step 5: Mocking and Testing Client Consumption
Now that you have everything deployed and ready, the next step you would typically do is testing it all together. Now, you might wonder how this will even work without any resolvers, or a backend to serve the data with.
While you can definitely spend your time writing the resolvers, business logic or connecting your backend, you first might want to test out the end-end data flow so that you get a validation on how clients would typically interact with your GraphQL API. To do this, you can either mock your schema or hard code the data initially in your resolvers and then serve the schema and test it all end-end.
This will establish a confidence about your development workflow, give you a clear idea about the data path, how your GraphQL operations (Mutations and Queries would look like), an insight into how you can consume the data and also presents you with opportunities doing things like end-end type checking, code generation and so on with your clients.
Step 6: Getting the resolvers and backend setup
As you might already know, with GraphQL, your clients don’t have to worry about the data source, your backend logic and the various complexities that go with it since they are all abstracted away and this helps you scale the backend and frontend independently of each other.
To do this, try treating your resolvers as just entities which do an operation and respond back with data given a set of inputs (similar to what you would typically do with a REST API). So, try setting up your backend/datasources from which you would want to serve the data (be it a database like Postgres or Mongo with or without an ORM like Prisma, Knex or Sequelize, or even an underlying resource like a REST API maybe with something like GraphQL Mesh or Graph databases like Dgraph) and also your resolvers to process the data as you see fit, adding your business logic on top and return back the fields as needed by the resolvers. This is the point where you replace the mocked data with data from the backend.
Step 7: Optimizing the data path
Now that you have connected your data sources and added your business logic with all the resolvers you need, the next step would typically be to optimize the data path to make sure that you are not doing repeated calls to the database increasing the load and bandwidth usage and reduce the roundtrips and processing needed as much as possible and also provide faster response times as the clients ask for data.
This is where you setup things like batching and also solve N+1 problems with something like a dataloader, setup caching with something like Redis or even an LRU cache to act as a proxy for the frequently accessed data whenever and wherever possible, optimizing the network chatter by using something like persisted queries, optimize your resolvers by retrieving as much data as possible from the parent resolvers, setting up pagination to limit the results returned, setting up things like query complexity to control the level of nesting and computation performed, rate-limiting in the gateway to avoid things like DDOS and so on.
This is really important cause, while GraphQL might provide your clients with a high degree of flexibility, it also comes with its own set of risks if not used right. So, try to keep even the worst case scenarios in mind and design for failures. Do remember that sometimes, it is better to make your application fail and crash rather than having to make it do the wrong thing.
Step 8: Controlling and Securing the data path
GraphQL provides all its clients with access to any data as they request it and while this might sound empowering (and it is), it is not without its own set of risks. You have to make sure that only the authorized clients have access to the data and only the data which they are allowed to have, only when they need it providing a proper context and purpose to the operation.
To do this, all the clients need to be authenticated properly whenever and however needed, authorization rules needs to be setup for all the fields either via directives, resolvers or any other mechanism which works well for you, have an encryption/decryption mechanism for confidential data like PII, ability to blacklist specific clients whenever needed and so on thereby controlling and securing the data as much as possible from your end considering that security must be a first-class citizen and not an after thought.
Step 9: Testing
Testing plays a major role especially when building scalable systems which have to be reliable even with a huge stream of changes which might affect it over time. And this is no exception when you work with GraphQL as well. You can setup automated tests, integration tests and so on as you normally would to improve the confidence people have on the system. And there are lot of libraries which facilitate the same as well like Mocha, Jest, AVA and so on taking a lot of the burden away from you.
You can test your resolvers, your GraphQL endpoint, your schema and so on. Testing can not just improve the reliability of your code, but can also act as a secondary source of documentation for people who are looking to understand what every function is doing and how to use it as part of their workflow. So, doing this as you go along can help.
Step 10: Automating or Scripting the repeatable parts
When you work on GraphQL or anything else for that matter, there is often a set of operations which you repeatedly do again and again. And over time, the cost of doing it would exponentially grow up.
For eg. you might push your schema to a registry if you use something like federation for all the changes you do, validate/lint your schema, do code generation as you change the SDL or anything else which is specific to your use case. This is where automation and scripting plays a major role and I can definitely say that I have saved countless hours of my valuable time by just scripting out things which I do repeatedly as part of my workflow.
Automation, especially with CI/CD becomes even more impactful when you are working with teams. There are a lot of interesting things you can do in your CI/CD pipeline like linting your schema and validating it, getting the list of breaking changes, pushing it to the registry, running automated tests, sending notifications to relevant people in your team as needed and so on saving a lot of time and also providing a high degree of reliability and confidence to what you ship to production.
This summarizes the most important steps you need to perform as part of your workflow with GraphQL and there are somethings which have been purposefully avoided in this list like setting up your infrastructure for file uploads, enabling real-time data exchange with subscriptions/live queries and so on since it all depends on your use case at hand but if you are interested in those, do have a look at our previous blog posts where we discuss about various tools and libraries which can help you with it.
While all of this may seem overwhelming, you need not boil yourself doing it all when you start but rather do it incrementally as you go along.
As all good things come to an end, this blog would be the last of this GraphQL series. But if you are looking for something specific which we have not addressed in this series, do let us know and maybe we can do a follow up blog post or even add it to this series if it makes sense. The reason we conclude here is that we intend to keep this series as a guide rather than a tutorial series since there is a lot of information already out there regarding the various tools, libraries and frameworks we talked about in this series.
But rest assured, we will definitely have a lot of blogs like these in the future as we work with GraphQL more and more and we also intend to provide you with a case study on how we do all of this at Timecampus sometime down the line. Do stick around for that. But in the meantime, there are a lot of other blog posts like these with blogs, videos and books from the community which is really worth checking out.
Also, I intend to keep the blog posts in this series as living documents rather than one-off blog posts. Hence, you might find us updating the information shared if needed over time.
If you are working your way through GraphQL and if this series really did help you in your path, we would love to know your story. GraphQL is where it is today because of people like you, the community, and I am very positive about its present and future especially in a data driven world and the journey towards bringing about a semantic web.
Looking for help or engineering consultancy? Feel free to reach out to me at vignesh[at]timecampus[dot]com and we can take it from there.
And if this helped, do share this across with your friends, do hang around and follow us for more posts this every week. See you all soon.