GraphQL - when REST API is to less - lessons learned

GraphQL - when REST API is
not enough - lessons learned
Marcin Stachniuk
5 March 2019

Marcin Stachniuk
mstachniuk.github.io
/mstachniuk/graphql-java-example
@MarcinStachniuk
wroclaw.jug.pl
shipkit.org

http://segfault.events/sites/gdansk2019/abstracts/shipkit/

Who is using REST?
@MarcinStachniukGraphQL - when REST API is not enough - lessons learned

REST - REpresentational State Transfer
https://api.example.com/customers/123
DELETE
PUT
POST
GET
PATCH

REST fixed response
GET /customers/111
{
"customer": {
"id": "111",
"name": "John Doe",
"email": "john@doe.com",
"company": {
"id": "222"
},
"orders": [
{
"id": "333"
},
{
"id": "444"
}
]
}
}
{
"customer": {
"id": "111",
"name": "John Doe",
"company": {
"href": "https://api.example.com/companies/222"
},
"orders": [
{
"href": "https://api.example.com/orders/333"
},
{
"href": "https://api.example.com/orders/444"
}
]
}
}
RE
ST

REST consequences: several roundtrips
U
nderFetching

REST response with nested data
GET /customers/111
{
"customer": {
"id": "111",
"name": "John Doe",
"company": {
"id": "222",
"name": "My Awesome Corporation",
"website": "MyAwesomeCorporation.com"
},
"orders": [
{
"id": "333",
"status": "delivered",
"items": [
{
"id": "555",
"name": "Silver Bullet",
"amount": "42",
"price": "10000000",
"currency": "USD",
"producer": {
"id": "777",
"name": "Lorem Ipsum",
"website": "LoremIpsum.com"
}
}
]
},
{
"id": "444",
"name": "Golden Hammer",
"amount": "5",
"price": "10000",
"currency": "USD",
"producer": {
...
}
}
] } }
RE
ST
O
verFetching

REST response with nested data and limit fields
GET /customers/111?fields=name,company/*,orders.status,orders.items(name,producer/name)
{
"customer": {
"id": "111",
"name": "John Doe",
"company": {
"id": "222",
},
"orders": [
{
"id": "333",
"items": [
{
"id": "555",
"amount": "42",
"price": "10000000",
"currency": "USD",
"producer": {
"id": "777",
}
}
]
},
{
"id": "444",
"amount": "5",
"price": "10000",
"currency": "USD",
"producer": {
...
}
}
] } }
RE
ST

REST response with projection
GET /customers/111?projection=customerPreview
{
"customer": {
"id": "111",
"name": "John Doe",
"company": {
"id": "222",
},
"orders": [
{
"id": "333",
"items": [
{
"id": "555",
"amount": "42",
"price": "10000000",
"currency": "USD",
"producer": {
"id": "777",
}
}
]
},
{
"id": "444",
"amount": "5",
"price": "10000",
"currency": "USD",
"producer": {
...
}
}
] } }
RE
ST

Different clients - different needs
/web /iphone /android /tv
Application
Web iPhone Android TV
RE
ST

Different clients - different needs
/web /iphone /android /tv
Application
Web iPhone Android TV
Content-Type: application/vnd.myawesomecorporation.com+v1+web+json
iphone
android
tv
RE
ST

REST contract is rigid

Platform Architecture
Platform
App 1 App 2
Customer
App X...

REST API Versioning
GET /v2/customers/111RE
ST
GET /customers/111?v=v2RE
ST
GET /customers/111 Accept header: application/vnd.myapp.2+jsonRE
ST
GET /customers/111 Custom header: x-ms-version:2RE
ST
● How to migrate?
● How to document?
● Is versionless API possible?
● What about continuous evolution of API?

Modularisation at UI

New Frontend Framework
ReactJS
Relay GraphQL
TypeScript

GraphQL
● Graph Query Language
● Published by Facebook in 2015
● Growth from Facebook Graph API
● Reference implementation in JavaScript

GraphQL main concepts
● One endpoint for all operations
● Always define in request what you need
● Queries, Mutations and Subscriptions
● Defined by schema

Data is a graph

GraphQL time for demo
● Fragments
● Aliases
● Directives
● Interfaces
● Unions
● Query
● Syntax Error
● Mutation
● Operation name
● Variables

GraphQL Simple API
GET /customers/2?fields=id,name,email
type Customer {
#fields with ! are not null
id: ID!
name: String!
email: String!
}
type Query {
customer(id: String!): Customer!
}
{
"data": {
"customer": {
"id": "2",
"name": "name",
"email": "a@b.com"
}
}
}
{
customer(id: "2") {
id
name
email
}
}
RE
ST

GraphQL Bad Request
GET /custo!@#$ -> 404
{
"data": null,
"errors": [
{
"message": "Invalid Syntax",
"locations": [
{
"line": 2,
"column": 8
}
],
"errorType": "InvalidSyntax",
"path": null,
"extensions": null
} ] }
{
custo!@#$
}
RE
ST
http.cat/200

Go back to the roots

GraphQL Simple API
GET /customers/2?fields=id,name,email,company(id,name)
type Customer {
id: ID!
name: String!
email: String!
company: Company
}
type Company {
id: ID!
name: String!
website: String!
}
type Query {
}
{
"data": {
"customer": {
"id": "2",
"name": "name",
"email": "a@b.com",
"company": {
"id": "211",
"name": "Company Corp."
}
}
}
}
{
customer(id: "2") {
id
name
email
company {
id
name
}
}
}
RE
ST

GraphQL Simple API
GET /customers/2?fields=id,name,email,orders(id,status)
type Customer {
id: ID!
name: String!
email: String!
company: Company
orders: [Order]
}
type Order {
id: ID!
status: Status
}
enum Status {
NEW, CANCELED, DONE
}
{
"data": {
"customer": {
"id": "2",
"name": "name",
"orders": [
{
"id": "55",
"status": "NEW"
},
{
"id": "66",
"status": "DONE"
}
] } } }
{
customer(id: "2") {
id
name
orders {
id
status
}
}
}
RE
ST

GraphQL mutations
input CreateCustomerInput {
name: String
email: String
clientMutationId: String!
}
type CreateCustomerPayload {
customer: Customer
clientMutationId: String!
}
type Mutation {
createCustomer(input: CreateCustomerInput):
CreateCustomerPayload!
}
{
"data": {
"createCustomer": {
"customer": {
"id": "40",
},
"clientMutationId":
"123"
}
}
}
POST /customers PUT /customers/123 DELETE /customers/123 PATCH /customers/123
mutation {
createCustomer(input: {
name: "MyName"
email: "me@me.com"
clientMutationId: "123"
}) {
customer {
id
}
clientMutationId
}
}
RE
ST

GraphQL mutations + operation name
mutation crCust {
name: "MyName"
email: "me@me.com"
}) {
customer {
id
}
clientMutationId
}
}
{
"data": {
"createCustomer": {
"customer": {
"id": "40",
},
"clientMutationId":
"123"
}
}
}
POST /customers PUT /customers/123 DELETE /customers/123 PATCH /customers/123RE
ST
mutation {
name: "MyName"
email: "me@me.com"
}) {
customer {
id
}
clientMutationId
}
}

GraphQL Variables
{
"data": {
"createCustomer": {
"customer": {
"id": "40",
},
"clientMutationId":
"123"
}
}
}
mutation crCust {
name: "MyName"
email: "me@me.com"
}) {
customer {
id
}
clientMutationId
}
}
RE
ST
mutation crCust ($input: CreateCustInput) {
createCustomer(input: $input) {
customer {
id
}
clientMutationId
}
}
{
"input": {
"name": "MyName 2",
"email": "me2@me.com",
"clientMutationId": "123"
}
}

GraphQL Aliases
GET /customers/2?fields=id,name,email +
{
"data": {
"cust1": {
"id": "2",
"name": "name",
"email": "a@b.com"
},
"cust2": {
"id": "3",
"name": "John Doe",
"email": "john@doe.com"
}
}
}
query get2Cust {
cust1: customer(id: "2") {
id
name
email
}
id
name
email
}
}
RE
ST
GET /customers/3?fields=id,name,emailRE
ST

GraphQL Fragment
GET /customers/2?fields=id,name,email +
{
"data": {
"cust1": {
"id": "2",
"name": "name",
"email": "a@b.com"
},
"cust2": {
"id": "3",
"name": "John Doe",
"email": "john@doe.com"
}
}
}
query get2Cust {
... frag1
}
... frag1
}
}
fragment frag1 on Customer {
id
name
email
}
RE
ST
GET /customers/3?fields=id,name,emailRE
ST

GraphQL Directive
GET /customers/2?fields=id,name
{
"data": {
"customer": {
"id": "2",
"name": "name"
}
}
}
query getCust ($showEmail: Boolean!) {
customer(id: "2") {
id
name
email @include(if: $showEmail)
}
}
{
"showEmail": false
}
RE
ST

GraphQL Interface
GET /users?fields=id,name,superAdmin,permissions
type Query {
users: [User]
}
interface User {
id: ID!
name: String!
email: String!
}
type Admin implements User {
superAdmin: Boolean! // + id...
}
type Moderator implements User {
permissions: [String] // + id...
}
{
"data": {
"users": [
{
"id": "777",
"name": "Admin a",
"superAdmin": true
},
{
"id": "888",
"name": "Moderator",
"permissions": [
"Delete Customer",
"Delete comment"
]}]}}
query getUsers {
users {
... on Admin {
id
name
superAdmin
}
... on Moderator {
id
name
permissions
}
}
}
RE
ST

GraphQL Union
GET /search?input=a?fields=typename,name
type Query {
search(input: String):
[SearchResult]
}
union SearchResult =
Customer |
Admin |
Moderator
{
"data": {
"search": [
{
"__typename": "Customer",
"name": "name"
},
{
"__typename": "Admin",
"name": "Admin a"
},
{
"__typename": "Moderator",
"name": "Moderator"
}]}}
query searchSmth {
search(input: "a") {
__typename
... on Customer {
name
}
... on Admin {
name
}
... on Moderator {
name
}
}
}
RE
ST

GraphQL
● Graph Query Language
● Published by Facebook in 2015
● Growth from Facebook Graph API
● Reference implementation in JavaScript
● First version of Java Library: 18 Jul 2015
https://github.com/graphql-java/graphql-java
● First usage: 21 Sep 2015

Lessons Learned #1
Never add a library to your project
few days after init release
● No community
● A lot of bugs
● Bad documentation
● Strict following reference
implementation and specification
DO NOT TRY
THIS AT WORK

How to implement DataFetcher for queries
@Component
public class CustomerFetcher extends PropertyDataFetcher<Customer> {
@Autowired
private CustomerService customerService;
@Override
public Customer get(DataFetchingEnvironment environment) {
String id = environment.getArgument("id");
return customerService.getCustomerById(id);
}
}
RE
ST
{
customer(id: "2") {
id
name
orders {
id
status
}
}
}

How to implement DataFetcher for queries
public class Customer {
private String id;
private String name;
private String email; // getters are not required
}
RE
ST
{
customer(id: "2") {
id
name
orders {
id
status
}
}
}
public class OrderDataFetcher extends PropertyDataFetcher<List<Order>> {
@Override
public List<Order> get(DataFetchingEnvironment environment) {
Customer source = environment.getSource();
String customerId = source.getId();
return orderService.getOrdersByCustomerId(customerId);
}
}

How to implement DataFetcher for mutations
@Component
public class CreateCustomersFetcher extends
PropertyDataFetcher<CreateCustomersPayload> {
@Override
public CreateCustomerPayload get(DataFetchingEnvironment env) {
Map<String, Object> input = env.getArgument("input");
String name = (String) input.get("name");
String email = (String) input.get("email");
String clientMutationId = (String) input.get("clientMutationId");
Customer customer = customerService.create(name, email);
return new CreateCustomerPayload(customer, clientMutationId);
}
RE
ST
mutation {
name: "MyName"
email: "me@me.com"
}) {
customer {
id
}
clientMutationId
}
}

In case Interfaces and Unions - Type Resolver is needed
public class UserTypeResolver implements TypeResolver {
@Override
public GraphQLObjectType getType(TypeResolutionEnvironment env) {
Object javaObject = env.getObject();
if (javaObject instanceof Admin) {
return env.getSchema().getObjectType("Admin");
} else if (javaObject instanceof Moderator) {
return env.getSchema().getObjectType("Moderator");
} else {
throw new RuntimeException("Unknown type " + javaObject.getClass().getName());
}
}
}
public interface User {...}
public class Admin implements User {...}
public class Moderator implements User {...}

Glue everything together
RuntimeWiring runtimeWiring = RuntimeWiring.newRuntimeWiring()
.type("Query", builder ->
builder.dataFetcher("customer", customerFetcher))
.type("Mutation", builder ->
builder.dataFetcher("createCustomer", createCustomerFetcher))
.type(newTypeWiring("User")
.typeResolver(new UserTypeResolver())
.build())
...
.build();

Abstraction over GraphQL Java
Our abstraction
Data Fetcher 2
Inputs mapping to objects
Schema definition
Pagination
...
Data Fetcher 1 Data Fetcher N...

Lessons Learned #2
Abstraction is not good if you don’t understand
how it works under the hood
● Copy paste errors
● Wrong usage
● Hard to update to new version

GraphQL type system
How to define your schema?

Code First approach
private GraphQLFieldDefinition customerDefinition() {
return GraphQLFieldDefinition.newFieldDefinition()
.name("customer")
.argument(GraphQLArgument.newArgument()
.name("id")
.type(new GraphQLNonNull(GraphQLString)))
.type(new GraphQLNonNull(GraphQLObjectType.newObject()
.name("Customer")
.field(GraphQLFieldDefinition.newFieldDefinition()
.name("id")
.description("fields with ! are not null")
.type(new GraphQLNonNull(GraphQLID))
.build())
….
.build()))
.dataFetcher(customerFetcher)
.build();
}
Schema First approach
type Query {
}
type Customer {
#fields with ! are not null
id: ID!
name: String!
email: String!
company: Company
orders: [Order]
}

Code First approach - How to build
Introspection
query
Introspection
response
Replace Relay
definitions
Typescript relay
plugin

Schema First approach
type Customer {
# fields with ! are required
id: ID!
name: String!
email: String!
company: Company
orders: [Order]
}
*.graphqls
SchemaParser schemaParser = new SchemaParser();
File file = // ...
TypeDefinitionRegistry registry = schemaParser.parse(file);
SchemaGenerator schemaGenerator = new SchemaGenerator();
RuntimeWiring runtimeWiring = RuntimeWiring.newRuntimeWiring()
.type("Query", builder ->
builder.dataFetcher("customer", customerFetcher))
// ...
.build();
return schemaGenerator.makeExecutableSchema(registry, runtimeWiring);

Schema First approach - project building diagram
model.graphqls

GraphQL SPQR
https://github.com/leangen/graphql-spqr
SPQR - Senatus Populusque Romanus

SPQR approach
public class UserService {
@GraphQLQuery(name = "user")
public User getById(@GraphQLArgument(name = "id") Integer id) {
//...
}
}
public class Customer {
//...
@GraphQLQuery(name = "name", description = "A person's name")
public String getName() {
return name;
}
}

Lessons Learned #3
Code First Approach is the worst
Schema First Approach:
● Easy to maintain and
understand
● Helps organise work
● Demo schema is 3x smaller
GraphQL SPQR - ?
Code First approach:
● Hard to maintain
● It was the only way at the
beginning to define a schema
● No possibility to mix both
● No easy way to migrate to
Schema First

From Code First to Schema First migration
https://github.com/mstachniuk/graphql-schema-from-introspection-generator

GraphQL - How to define pagination, filtering, sorting?
Pagination:
● before, after
● offset, limit
Filtering:
● filter(name: “Bob” email: “%@gmail.com”)
● filter: {
OR: [{
email: “%@gmail.com”
}]
}, name: “Bob”
}
Sorting:
● orderBy: ASC, DESC
● sort: NEWEST, IMPORTANCE

Lessons Learned #4
GraphQL is not full query language
● Flexibility
● Less common conventions
● Dgraph.io created GraphQL+-

GraphQL downsides

GraphQL downsides: N+1 problem
{
customers { 1 call
id
name
orders { n calls
id
status
}
}
}
java-dataloader
● Add async BatchLoader
● Add caching
graphQL = GraphQL.newGraphQL(schema)
.queryExecutionStrategy(
new BatchedExecutionStrategy())
.build();
+ @Batched in DataFetcher#get()

Lessons Learned #5
If you have N + 1 problem
use java-dataloader

Complex Queries
fragment TypeRef on __Type {
kind
name
ofType {
kind
name
ofType {
kind
name
...
}
}
}
graphQL = GraphQL.newGraphQL(schema)
.instrumentation(
new ChainedInstrumentation(asList(
new MaxQueryComplexityInstrumentation(200),
new MaxQueryDepthInstrumentation(20)
)))
.build();

Lessons Learned #6
Performance killer?
Define instrumentation!

GraphQL downsides: Bad GraphQL API definition - examples
{
customer(id: "2") { … }
customerFull(id: "2") { … }
customerFull2(id: "2") { … }
customerWithDetails(id: "2") { … }
...
}
{
customer(id: "2") {
id
name
orders {
id
status
}
}
}

query getUser {
user(id: "123") {
... on Admin {
id
superAdmin
}
... on Moderator {
id
permissions
}
}
}
{
admin(id: "123") {
id
superAdmin
}
moderator(id: "123") {
id
permissions
}
}

{
orders (input: {
status: "NEW"
first: "2"
offset: "3"
}, first: "1", offset: "3") {
Items { … }
}

Lessons Learned #7
Thinking shift is a key
● Let’s think in graphs and NOT in
endpoints / resources / entities / DTOs
● Bad design of our API

GraphQL Testing

Testing GraphQL
@SpringBootTest
@ContextConfiguration(classes = Main)
class CustomerFetcherSpec extends Specification {
@Autowired
GraphQLSchema graphQLSchema
GraphQL graphQL
def setup() {
graphQL = GraphQL.newGraphQL(graphQLSchema).build()
}

Testing GraphQL
def "should get customer by id"() {
given:
def query = """{ customer(id: "2") { … } }"""
def expected = [ "customer": [ … ] ]
when:
def result = graphQL.execute(query)
then:
result.data == expected
}

Lessons Learned #8
Testing is easy
Trap Adventure 2 - "The Hardest Retro Game"

Versioning
GraphQL takes a strong opinion on avoiding
versioning by providing the tools for the
continuous evolution of a GraphQL schema.
Src: https://graphql.org/learn/best-practices/#versioning

Versioning
type Query {
customer(id: String!): Customer! @deprecated(reason: "not used ...")

Lessons Learned #9
Versioning is not a problem
GET /v2/customers/111RE
ST
GET /customers/111?v=v2RE
ST
GET /customers/111 Accept header: application/vnd.myapp.2+jsonRE
ST
GET /customers/111 Custom header: x-ms-version:2RE
ST
Versioning problem is different

Tools

Tools GraphiQL

Tools Relay
@MarcinStachniukGraphQL - when REST API is not enough - lessons learned @MarcinStachniukGraphQL - when REST API is not enough - lessons learned
user(...) {
photo(width: "120", height: "120")
}
user(...) {
name
}
user(...) {
email
}
user(...) {
name
email
photo(width: "120", height: "120")
}

Distelli/graphql-apigen
https://github.com/Distelli/graphql-apigen
Generate Java APIs with GraphQL Schemas in order to
facilitate "schema first" development.

GraphQL Java Tools
https://github.com/graphql-java/graphql-java-tools
Map GraphQL schema to POJOs

More libraries and projects related to graphql-java
https://github.com/graphql-java/awesome-graphql-java
● Generates a GraphQL schema from a JDBC data source
● Annotations-based syntax for GraphQL schema definition
● JPA Implementation of GraphQL (builds on graphql-java)
● And more examples

Apollo
Build a universal GraphQL API on top of your existing REST
APIs, so you can ship new application features fast without
waiting on backend changes.

Lessons Learned #10
Tooling is nice
now

Summary
GraphQL Pros:
● Nice alternative to REST
● It can be used together with REST
● Good integration with Relay / ReactJS
● You get exactly what you want to get
● Good for API with different clients
● Good to use on top of existing API
● Self documented
● Easy testing
● Nice tooling
● Thin layer
GraphQL Cons:
● High entry barrier
● Hard to return simple Map
● Not well know (yet)
● Performance overhead
● A lot of similar code to write
● No Operation Idempotency
● No Cache & Security
● No Authorisation
● Always HTTP 200 -
unsupported tools

Nothing is a silver bullet

Q&A

GraphQL - when REST API is
not enough - lessons learned
Marcin Stachniuk
5 March 2019
Thank
you!

GraphQL - when REST API is to less - lessons learned

Recommended

Recommended

More Related Content

What's hot

What's hot (8)

Similar to GraphQL - when REST API is to less - lessons learned

Similar to GraphQL - when REST API is to less - lessons learned (20)

More from MarcinStachniuk

More from MarcinStachniuk (19)

Recently uploaded

Recently uploaded (20)

GraphQL - when REST API is to less - lessons learned