12 KiB
title | summary |
---|---|
Upgrading from GraphQL 3 | A high-level view of what you'll need to change when upgrading to GraphQL 4 |
Upgrading from GraphQL 3
[alert] You are viewing docs for a pre-release version of silverstripe/graphql (4.x). Help us improve it by joining #graphql on the Community Slack, and report any issues at github.com/silverstripe/silverstripe-graphql. Docs for the current stable version (3.x) can be found here [/alert]
The 4.0 release of silverstripe-graphql
underwent a massive set of changes representing an
entire rewrite of the module. This was done as part of a year-long plan to improve performance. While
there is no specific upgrade path, there are some key things to look out for and general guidelines on how
to adapt your code from the 3.x release to 4.x.
In this section, we'll cover each of these upgrade issues in order of impact.
GraphQL schemas require a build step
The most critical change moving from 3.x to 4.x affects the developer experience. The key to improving performance in GraphQL requests was eliminating the overhead of generating the schema at runtime. This didn't scale. As the GraphQL schema grew, API response latency increase.
To eliminate this overhead, the GraphQL API relies on generated code for the schema. You need to run a task to build it.
To run the task, use:
$ vendor/bin/sake dev/graphql/build schema=mySchema
You can also run the task in the browser:
http://example.com/dev/graphql/build?schema=mySchema
[info]
Most of the time, the name of your schema is default
. If you're editing DataObjects that are accessed
with GraphQL in the CMS, you may have to build the admin
schema as well.
[/info]
This build process is a larger topic with a few more things to be aware of. Check the building the schema documentation to learn more.
The Manager class, the godfather of GraphQL 3, is gone
silverstripe-graphql
3.x relied heavily on the Manager
class. This became a catch-all that handled
registration of types, execution of scaffolding, running queries and middleware, error handling, and more. This
class has been broken up into separate concerns:
Schema
<- register your stuff hereQueryHandlerInterface
<- Handles GraphQL queries, applies middlewares and context. You'll probably never have to touch it.
Upgrading
before
SilverStripe\GraphQL\Manager:
schemas:
default:
types: {}
queries: {}
mutations: {}
after
SilverStripe\GraphQL\Schema\Schema:
schemas:
default:
src:
- app/_graphql # A directory of your choice
Add the appropriate yaml files to the directory. For more information on this pattern, see the configuring your schema section.
app/_graphql
types.yml
queries.yml
mutations.yml
models.yml
enums.yml
interfaces.yml
unions.yml
TypeCreator, QueryCreator, and MutationCreator are gone
A thorough look at how these classes were being used revealed that they were really just functioning as value objects that basically just created configuration in a static context. That is, they had no real reason to be instance-based. Most of the time, they can easily be ported to configuration.
Upgrading
before
class GroupTypeCreator extends TypeCreator
{
public function attributes()
{
return [
'name' => 'group'
];
}
public function fields()
{
return [
'ID' => ['type' => Type::nonNull(Type::id())],
'Title' => ['type' => Type::string()],
'Description' => ['type' => Type::string()]
];
}
}
after
app/_graphql/types.yml
group:
fields:
ID: ID!
Title: String
Description: String
That's a simple type, and obviously there's a lot more to it than that, but have a look at the working with generic types section of the documentation.
Resolvers must be static callables
You can no longer use instance methods for resolvers. They can't be easily transformed into generated
PHP code in the schema build step. These resolvers should be refactored to use the static
declaration
and moved into a class.
Upgrading
Move your resolvers into one or many classes, and register them.
before
class LatestPostResolver implements OperationResolver
{
public function resolve($object, array $args, $context, ResolveInfo $info)
{
return Post::get()->sort('Date', 'DESC')->first();
}
}
after
app/_graphql/config.yml
resolvers:
- MyProject\Resolvers\MyResolverA
- MyProject\Resolvers\MyResolverB
class MyResolverA
{
public static function resolveLatestPost($object, array $args, $context, ResolveInfo $info)
{
return Post::get()->sort('Date', 'DESC')->first();
}
}
This method relies on resolver discovery, which you can learn more about in the documentation.
Alternatively, you can hardcode the resolver into your config:
app/_graphql/queries.yml
latestPost:
type: Post
resolver: ['MyResolvers', 'latestPost' ]
ScaffoldingProviders are now SchemaUpdaters
If you were updating your schema with procedural code, you'll need to change your ScaffoldingProvider
interface to SchemaUpdater
, and use the updateSchema(Schema $schema): void
function.
Upgrading
Register your schema builder, and change the code.
before
SilverStripe\GraphQL\Manager:
schemas:
default:
scaffolding_providers:
- 'MyProject\MyProvider'
class MyProvider implements ScaffoldingProvider
{
public function provideGraphQLScaffolding(SchemaScaffolder $scaffolder)
{
// updates here...
}
}
after
SilverStripe\GraphQL\Schema\Schema:
schemas:
default:
execute:
- 'MyProject\MyProvider'
class MyProvider implements SchemaUpdater
{
public function updateSchema(Schema $schema): void
{
// updates here...
}
}
The API for procedural code has been completely rewritten. You'll need to rewrite all of the code in these classes. For more information on working with procedural code, read the using procedural code documentation.
Goodbye, scaffolding, hello models
In the 3.x release, a massive footprint of the codebase was dedicated to a DataObject-specific API called "scaffolding" that was used to generate types, queries, fields, and more from the ORM. In 4.x, that approach has been moved to concept called model types.
A model type is just a type that is backed by a class that express awareness of its schema (like a DataObject!). At a high-level, it needs to answer questions like:
- Do you have field X? What type is field Y?
- What are all the fields you offer?
- What operations do you provide?
- Do you require any extra types to be added to the schema?
Upgrading
The 4.x release ships with a model type implementation specifically for DataObjects, which you can use a lot like the old scaffolding API.
before
SilverStripe\GraphQL\Manager:
schemas:
default:
scaffolding:
types:
SilverStripe\Security\Member:
fields: '*'
operations: '*'
SilverStripe\CMS\Model\SiteTree:
fields:
title: true
content: true
operations:
read: true
after
app/_graphql/models.yml
SilverStripe\Security\Member:
fields: '*'
operations: '*'
SilverStripe\CMS\Model\SiteTree:
fields:
title: true
content: true
operations:
read: true
DataObject field names are lowerCamelCase by default
The 3.x release of the module embraced an anti-pattern of using UpperCamelCase field names so that they could map to the conventions of the ORM. This makes frontend code look awkward, and there's no great reason for the Silverstripe CMS graphql server to break convention. In this major release, the lowerCamelCase approach is encouraged.
Upgrading
Change the casing in your queries.
before
query readPages {
edges {
nodes {
Title
ShowInMenus
}
}
}
after
query readPages {
edges {
node {
title
showInMenus
}
}
}
edges
no longer required
We don't have cursor-based pagination in Silverstripe CMS, so
the use of edges
is merely for convention. You can eliminate a layer here and just use nodes
, but edges
still exists for backward compatibility.
query readPages {
nodes {
title
showInMenus
}
}
DataObject type names are simpler
To avoid naming collisions, the 3.x release of the module used a pretty aggressive approach to ensuring
uniqueness when converting a DataObject class name to a GraphQL type name, which was <vendorName><shortName>
.
In the 4.x release, the typename is just the shortName
by default, which is based on the assumption that
most of what you'll be exposing is in your own app code, so collisions aren't that likely.
Upgrading
Change any references to DataObject type names in your queries
before
query SilverStripeSiteTrees {}
after
query SiteTrees {}
If this new pattern is not compatible with your set up (e.g. if you use feature-based namespacing), you have full
control over how types are named. You can use the type_formatter
and type_prefix
on DataObjectModel
to
influence the naming computation. Read more about this in the DataObject model type docs.
The Connection class has been moved to plugins
In the 3.x release, you could wrap a query in the Connection
class to add pagination features.
In 4.x, these features are provided via the new plugin system.
The good news is that all DataObject queries are paginated by default, and you shouldn't have to worry about this, but if you are writing a custom query and want it paginated, check out the section on adding pagination to a custom query.
Additionally, the sorting features that were provided by Connection
have been moved to a plugin dedicated to
SS_List
results. Again, this plugin is applied to all DataObjects by default, and will include all of their
sortable fields by default. This is configurable, however. See the
query plugins section for more information.
Upgrading
There isn't much you have to do here to maintain compatibility. If you prefer to have a lot of control over what your sort fields are, check out the linked documentation above.
Query filtering has been moved to a plugin
The previous QueryFilter
API has been vastly simplified in a new plugin. Filtering is provided to all
read queries by default, and should include all filterable fields, including nested relationships.
This is configurable, however. See the
query plugins section for more information.
Upgrading
There isn't much you have to do here to maintain compatibility. If you prefer to have a lot of control over what your filter fields are, check out the linked documentation above.
Query permissions have been moved to a plugin
This was mostly an internal API, and shouldn't be affected in an upgrade, but if you want more information on how it works, you can read the permissions documentation.
Enums are first-class citizens
In the 3.x release, there was no clear path to creating enum types, but in 4.x, they have a prime spot in the configuration layer.
before
(A type creator that has been hacked to return an Enum
singleton?)
after
app/_graphql/enums.yml
Status:
SHIPPED: Shipped
CANCELLED: Cancelled
PENDING: Pending