Value Types in Apollo Federation
Share types and fields across multiple subgraphs
In a federated graph, it's common to want to reuse a GraphQL type across multiple subgraphs.
For example, suppose you want to define and reuse a generic Position
type in different subgraphs:
1type Position {
2 x: Int!
3 y: Int!
4}
Types like this are called value types. This article describes how to share value types and their fields in federated graph, enabling multiple subgraphs to define and resolve them.
Sharing object types
By default, in Federation 2 subgraphs, a single object field can't be defined or resolved by more than one subgraph schema.
Consider the following Position
example:
❌
1type Position {
2 x: Int!
3 y: Int!
4}
1type Position {
2 x: Int!
3 y: Int!
4}
Attempting to compose these two subgraph schemas together will break composition. The router doesn't know which subgraph is responsible for resolving Position.x
and Position.y
. To enable multiple subgraphs to resolve these fields, you must first mark that field as @shareable
.
Position
, you can designate the Position
type as an entity.Using @shareable
The @shareable
directive enables multiple subgraphs to resolve a particular object field (or set of object fields).
To use @shareable
in a subgraph schema, you first need to add the following snippet to that schema to opt in to Federation 2:
1extend schema
2 @link(url: "https://specs.apollo.dev/federation/v2.3",
3 import: ["@key", "@shareable"])
Then you can apply the @shareable
directive to an object type, or to individual fields of that type:
✅
1type Position @shareable {
2 x: Int!
3 y: Int!
4}
1type Position {
2 x: Int! @shareable
3 y: Int! @shareable
4}
@shareable
is equivalent to marking all of its fields as @shareable
, so the two subgraph definitions above are equivalent.Both subgraphs A and B can now resolve the x
and y
fields for the Position
type, and our subgraph schemas will successfully compose into a supergraph schema.
⚠️ Important considerations for @shareable
If a type or field is marked
@shareable
in any subgraph, it must be marked either@shareable
or@external
in every subgraph that defines it. Otherwise, composition fails.If multiple subgraphs can resolve a field, make sure each subgraph's resolver for that field behaves identically. Otherwise, queries might return inconsistent results depending on which subgraph resolves the field.
Using @shareable
with extend
If you apply @shareable
to an object type declaration, it only applies to the fields within that exact declaration. It does not apply to other declarations for that same type:
1type Position @shareable {
2 x: Int! # shareable
3 y: Int! # shareable
4}
5
6extend type Position {
7 z: Int! # ⚠️ NOT shareable!
8}
Using the extend
keyword, the schema above includes two different declarations for Position
. Because only the first declaration is marked @shareable
, Position.z
is not considered shareable.
To make Position.z
shareable, you can do one of the following:
Mark the individual field
z
with@shareable
.GraphQL1extend type Position { 2 z: Int! @shareable 3}
Mark the entire
extend
declaration with@shareable
.This strategy requires targeting v2.2 or later of the Apollo Federation specification in your subgraph schema. Earlier versions do not support applying
@shareable
to the same object type multiple times.
GraphQL1extend schema 2 @link(url: "https://specs.apollo.dev/federation/v2.3", import: ["@shareable"]) 3 4extend type Position @shareable { 5 z: Int! 6}
Differing shared fields
Shared fields can only differ in their return types and arguments in specific ways. If fields you want to share between subgraphs differ more than is permitted, use entities instead of shareable value types.
Return types
Let's say two subgraphs both define an Event
object type with a timestamp
field:
❌
1type Event @shareable {
2 timestamp: Int!
3}
1type Event @shareable {
2 timestamp: String!
3}
Subgraph A's timestamp
returns an Int
, and Subgraph B's returns a String
. This is invalid. When composition attempts to generate an Event
type for the supergraph schema, it fails due to an unresolvable conflict between the two timestamp
field definitions.
Next, look at these varying definitions for the Position
object type:
✅
1type Position @shareable {
2 x: Int!
3 y: Int!
4}
1type Position @shareable {
2 x: Int
3 y: Int
4}
The x
and y
fields are non-nullable in Subgraph A, but they're nullable in Subgraph B. This is valid. Composition recognizes that it can use the following definition for Position
in the supergraph schema:
1type Position {
2 x: Int
3 y: Int
4}
This definition works for querying Subgraph A, because Subgraph A's definition is more restrictive than this (a non-nullable value is always valid for a nullable field). In this case, composition coerces Subgraph A's Position
fields to satisfy the reduced restrictiveness of Subgraph B.
x
and y
remain non-nullable.Arguments
Arguments for a shared field can differ between subgraphs in certain ways:
If an argument is required in at least one subgraph, it can be optional in other subgraphs. It cannot be omitted.
If an argument is optional in every subgraph where it's defined, it is technically valid to omit it in other subgraphs. However:
⚠️ If a field argument is omitted from any subgraph, that argument is omitted from the supergraph schema entirely. This means that clients can't provide the argument for that field.
✅
1type Building @shareable {
2 # Argument is required
3 height(units: String!): Int!
4}
1type Building @shareable {
2 # Argument can be optional
3 height(units: String): Int!
4}
❌
1type Building @shareable {
2 # Argument is required
3 height(units: String!): Int!
4}
1type Building @shareable {
2 # ⚠️ Argument can't be omitted! ⚠️
3 height: Int!
4}
⚠️
1type Building @shareable {
2 # Argument is optional
3 height(units: String): Int!
4}
1type Building @shareable {
2 # Argument can be omitted, BUT
3 # it doesn't appear in the
4 # supergraph schema!
5 height: Int!
6}
For more information, see Input types and field arguments.
Omitting fields
Look at these two definitions of a Position
object type:
⚠️
1type Position @shareable {
2 x: Int!
3 y: Int!
4}
1type Position @shareable {
2 x: Int!
3 y: Int!
4 z: Int!
5}
Subgraph B defines a z
field, but Subgraph A doesn't. In this case, when composition generates the Position
type for the supergraph schema, it includes all three fields:
1type Position {
2 x: Int!
3 y: Int!
4 z: Int!
5}
This definition works for Subgraph B, but it presents a problem for Subgraph A. Let's say Subgraph A defines the following Query
type:
1type Query {
2 currentPosition: Position!
3}
According to the hypothetical supergraph schema, the following query is valid against the supergraph:
❌
1query GetCurrentPosition {
2 currentPosition {
3 x
4 y
5 z # ⚠️ Unresolvable! ⚠️
6 }
7}
And here's the problem: if Subgraph B doesn't define Query.currentPosition
, this query must be executed on Subgraph A. But Subgraph A is missing the Position.z
field, so that field is unresolvable!
Composition recognizes this potential problem, and it fails with an error. To learn how to fix it, check out Solutions for unresolvable fields.
Adding new shared fields
Adding a new field to a value type can cause composition issues, because it's challenging to add the field to all defining subgraphs at the same time.
Let's say we're adding a z
field to our Position
value type, and we start with Subgraph A:
⚠️
1type Position @shareable {
2 x: Int!
3 y: Int!
4 z: Int!
5}
1type Position @shareable {
2 x: Int!
3 y: Int!
4}
It's likely that when we attempt to compose these two schemas, composition will fail, because Subgraph B can't resolve Position.z
.
To incrementally add the field to all of our subgraphs without breaking composition, we can use the @inaccessible
directive.
Using @inaccessible
If you apply the @inaccessible
directive to a field, composition omits that field from your router's API schema. This helps you incrementally add a field to multiple subgraphs without breaking composition.
To use @inaccessible
in a subgraph, first make sure you include it in the import
array of your Federation 2 opt-in declaration:
1extend schema
2 @link(url: "https://specs.apollo.dev/federation/v2.3",
3 import: ["@key", "@shareable", "@inaccessible"])
Then, whenever you add a new field to a value type, apply @inaccessible
to that field if it isn't yet present in every subgraph that defines the value type:
1type Position @shareable {
2 x: Int!
3 y: Int!
4 z: Int! @inaccessible
5}
1type Position @shareable {
2 x: Int!
3 y: Int!
4}
Even if Position.z
is defined in multiple subgraphs, you only need to apply @inaccessible
in one subgraph to omit it. In fact, you might want to apply it in only one subgraph to simplify removing it later.
With the syntax above, composition omits Position.z
from the generated API schema, and the resulting Position
type includes only x
and y
fields.
Position.z
does appear in the supergraph schema, but the API schema enforces which fields clients can include in operations. Learn more about federated schemas.Whenever you're ready, you can now add Position.z
to Subgraph B:
1type Position @shareable {
2 x: Int!
3 y: Int!
4 z: Int! @inaccessible
5}
1type Position @shareable {
2 x: Int!
3 y: Int!
4 z: Int!
5}
At this point, Position.z
is still @inaccessible
, so composition continues to ignore it.
Finally, when you've added Position.z
to every subgraph that defines Position
, you can remove @inaccessible
from Subgraph A:
1type Position @shareable {
2 x: Int!
3 y: Int!
4 z: Int!
5}
1type Position @shareable {
2 x: Int!
3 y: Int!
4 z: Int!
5}
Composition now successfully includes Position.z
in the supergraph schema!
Unions and interfaces
In Federation 2, union
and interface
type definitions can be shared between subgraphs by default, and those definitions can differ:
1union Media = Book | Movie
2
3interface User {
4 name: String!
5}
1union Media = Book | Podcast
2
3interface User {
4 name: String!
5 age: Int!
6}
Compositional logic merges these definitions in your supergraph schema:
1union Media = Book | Movie | Podcast
2
3# The object types that implement this interface are
4# responsible for resolving these fields.
5interface User {
6 name: String!
7 age: Int!
8}
This can be useful when different subgraphs are responsible for different subsets of a particular set of related types or values.
enum
type across multiple subgraphs. For details, see Merging types from multiple subgraphs.Challenges with shared interfaces
Sharing an interface type across subgraphs introduces maintenance challenges whenever that interface changes. Consider these subgraphs:
1interface Media {
2 id: ID!
3 title: String!
4}
5
6type Book implements Media {
7 id: ID!
8 title: String!
9}
1interface Media {
2 id: ID!
3 title: String!
4}
5
6type Podcast implements Media {
7 id: ID!
8 title: String!
9}
Now, let's say Subgraph B adds a creator
field to the Media
interface:
❌
1interface Media {
2 id: ID!
3 title: String!
4}
5
6type Book implements Media {
7 id: ID!
8 title: String!
9 # ❌ Doesn't define creator!
10}
1interface Media {
2 id: ID!
3 title: String!
4 creator: String!
5}
6
7type Podcast implements Media {
8 id: ID!
9 title: String!
10 creator: String!
11}
This breaks composition, because Book
also implements Media
but doesn't define the new creator
field.
To prevent this error, all implementing types across all subgraphs need to be updated to include all fields of Media
. This becomes more and more challenging to do as your number of subgraphs and teams grows. Fortunately, there's a solution.
Solution: Entity interfaces
Apollo Federation 2.3 introduces a powerful abstraction mechanism for interfaces, enabling you to add interface fields across subgraphs without needing to update every single implementing type.
Learn more about entity interfaces.
Input types
Subgraphs can share input
type definitions, but composition merges their fields using an intersection strategy. When input
types are composed across multiple subgraphs, only mutual fields are preserved in the supergraph schema:
1input UserInput {
2 name: String!
3 age: Int # Not in Subgraph B
4}
1input UserInput {
2 name: String!
3 email: String # Not in Subgraph A
4}
Compositional logic merges only the fields that all input
types have in common. To learn more, see Merging input types and field arguments.
1input UserInput {
2 name: String!
3}
To learn more about how composition merges different schema types under the hood, see Merging types during composition.