3.6 KiB
| title | summary |
|---|---|
| Building the schema | Turn your schema configuration into executable code |
Getting started
[CHILDREN asList]
[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]
Building the schema
The primary API surface of the silverstripe-graphql module is the configuration YAML, and
some procedural configuration as well. It is important to understand
that none of this configuration gets interpreted at runtime. Loading the schema configuration
at runtime and converting it to executable code has dire effects on performance, making
API requests slower and slower as the schema grows larger.
To mitigate this problem, the schema that gets executed at runtime is generated PHP code. This code generation happens during a build step, and it is critical to run this build step whenever the schema changes.
Running the build
The task that generates the schema code is build-schema. It takes a parameter of schema, whose value should be the name of the schema you want to build.
$ vendor/bin/sake dev/graphql/build schema=default
Keep in mind that many of your changes will be in YAML, which also requires a flush.
`$ vendor/bin/sake dev/graphql/build schema=default flush=1
[info]
If you do not provide a schema parameter, the task will build all schemas.
[/info]`
Building on dev/build
By default, all schemas will be built as a side-effect of dev/build. To disable this, change
the config:
SilverStripe\GraphQL\Extensions\DevBuildExtension:
enabled: false
Caching
Generating code is a pretty expensive process. A large schema with 50 dataobject classes exposing all their operations can take up to 20 seconds to generate. This may be acceptable for initial builds and deployments, but during incremental development this can really slow things down.
To mitigate this, the generated code for each type is cached against a signature. If the type hasn't changed, it doesn't re-render. This reduces build times to under one second for incremental changes.
Clearing the cache
Normally, we'd use flush=1 to clear the cache, but since you almost always need to run flush=1 with the build task, it isn't a good fit. Instead, use clear=1.
$ vendor/bin/sake dev/graphql/build schema=default clear=1
If your schema is producing unexpected results, try using clear=1 to eliminate the possibility
of a caching issue. If the issue is resolved, record exactly what you changed and create an issue.
Build gotchas
Keep in mind that it's not always explicit schema configuration changes that require a build. Anything influencing the output of the schema will require a build. This could include tangential changes such as:
- Updating the
$dbarray (or relationships) of a DataObject that hasfields: '*'. - Adding a new resolver for a type that uses resolver discovery
- Adding an extension to a DataObject
- Adding a new subclass to a DataObject that is already exposed
Viewing the generated code
TODO, once we figure out where it will go
Further reading
[CHILDREN]