Experience GraphQL Summit 2024

Programmatic cache IDs in Apollo Kotlin

In situations where declarative cache IDs don't fit your use case, you can programmatically generate cache IDs for object types in your normalized cache.

You can generate a given object type's cache ID from one of two sources:

SourceClassDescription
From a response object's fields (e.g., Book.id)CacheKeyGeneratorThis happens after a network request and is essential to merging a query result with existing cached data. This is the most common case.
From a GraphQL operation's arguments (e.g., author(id: "au456"))CacheKeyResolverThis happens before a network request and enables you to avoid a network round trip if all requested data is in the cache already. This is an optional optimization that can avoid some cache misses.

Apollo Kotlin provides a class for generating cache keys from each of these sources.

CacheKeyGenerator

The CacheKeyGenerator class enables you to generate custom cache IDs from an object's field values. This basic example generates every object type's cache ID from its id field:

Kotlin
1val cacheKeyGenerator = object : CacheKeyGenerator {
2  override fun cacheKeyForObject(obj: Map<String, Any?>, context: CacheKeyGeneratorContext): CacheKey? {
3    // Generate the cache ID based on the object's id field
4    return CacheKey(obj["id"] as String)
5  }
6}

To use your custom CacheKeyGenerator, include it in your cache initialization code like so:

Kotlin
1val apolloClient = ApolloClient.Builder()
2  .serverUrl("https://...")
3  .normalizedCache(
4    normalizedCacheFactory = cacheFactory,
5    cacheKeyGenerator = cacheKeyGenerator,
6  )
7  .build()

You can get the current object's typename from the context object and include it in the generated ID, like so:

Kotlin
1val cacheKeyGenerator = object : CacheKeyGenerator {
2  override fun cacheKeyForObject(obj: Map<String, Any?>, context: CacheKeyGeneratorContext): CacheKey? {
3    val typename = obj["__typename"] as String
4    val id = obj["id"] as String
5
6    return CacheKey(typename, id)
7  }
8}

You can also use the current object's typename to use different cache ID generation logic for different object types.

Note that for cache ID generation to work, your GraphQL operations must return whatever fields your custom code relies on (such as id above). If a query does not return a required field, the cache ID will be inconsistent, resulting in data duplication. Also, using context.field.type.leafType().name yields the typename of the Union as opposed to the expected runtime value of Union received in the response. Instead querying for the __typename is safer. To make sure __typename is included in all operations set the adTypename gradle config:

Text
1apollo {
2  addTypename.set("always")
3  //
4}

CacheKeyResolver

The CacheKeyResolver class enables you to generate custom cache IDs from an field's arguments. This basic example generates every object type's cache ID from the id argument if it's present:

Kotlin
1val cacheKeyResolver = object: CacheKeyResolver() {
2  override fun cacheKeyForField(field: CompiledField, variables: Executable.Variables): CacheKey? {
3    // [field] contains compile-time information about what type of object is being resolved.
4    // Even though we call leafType() here, we're guaranteed that the type is a composite type and not a list
5    val typename = field.type.leafType().name
6
7    // resolveArgument returns the runtime value of the "id" argument
8    // from either the variables or as a literal value
9    val id = field.resolveArgument("id", variables)
10
11    if (id is String) {
12      // This field has an id argument, so we can use it to compute a cache ID
13      return CacheKey(typename, id)
14    }
15
16    // Return null to use the default handling
17    return null
18  }
19}

To use your custom CacheKeyResolver, include it in your cache initialization code like so:

Kotlin
1val apolloClient = ApolloClient.Builder()
2  .serverUrl("https://...")
3  .normalizedCache(
4    normalizedCacheFactory = cacheFactory,
5    cacheKeyGenerator = cacheKeyGenerator,
6    cacheResolver = cacheKeyResolver
7  )
8  .build()

Note the following about using a custom CacheKeyResolver:

  • The cacheKeyForField function is called for every field in your operation that returns a composite type, so it's important to return null if you don't want to handle a particular field.

  • The function is not called for fields that return a list of composite types. See below.

Handling lists

Let's say we have this query:

GraphQL
1query GetBooks($ids: [String!]!) {
2  books(ids: $ids) {
3    id
4    title
5  }
6}

To have the cache look up all books in the ids list, we need to override listOfCacheKeysForField in CacheKeyResolver:

Kotlin
1override fun listOfCacheKeysForField(field: CompiledField, variables: Executable.Variables): List<CacheKey?>? {
2  // Note that the field *can* be a list type here
3  val typename = field.type.leafType().name
4
5  // resolveArgument returns the runtime value of the "id" argument
6  // from either the variables or as a literal value
7  val ids = field.resolveArgument("ids", variables)
8
9  if (ids is List<*>) {
10    // This field has an id argument, so we can use it to compute a cache key
11    return ids.map { CacheKey(typename, it as String) }
12  }
13
14  // Return null to use the default handling
15  return null
16}

For the sake of simplicity, only one level of list is supported. To support more nested lists, you can implement CacheResolver. CacheResolver is a generalization of CacheKeyResolver that can return any value from the cache, even scalar values:

Kotlin
1val cacheResolver = object: CacheResolver {
2  override fun resolveField(
3      field: CompiledField,
4      variables: Executable.Variables,
5      parent: Map<String, @JvmSuppressWildcards Any?>,
6      parentId: String,
7  ): Any? {
8
9    var type = field.type
10    var listDepth = 0
11
12    while (true) {
13      when (type) {
14        is CompiledNotNullType -> type = type.ofType
15        is CompiledListType -> {
16          listDepth++
17          type = type.ofType
18        }
19        else -> break
20      }
21    }
22
23    // Now type points to the leaf type and lestDepth is the nesting of lists required
24
25    // Return a kotlin value for this field
26    // No type checking is done here so it must match the expected GraphQL type
27
28    if (listDepth == 2) {
29      return listOf(listOf("0", "1"))
30    }
31
32    // CacheResolver must always call DefaultCacheResolver last or all fields will be null else
33    return DefaultCacheResolver.resolveField(field, variables, parent, parentId)
34  }
35}
Feedback

Edit on GitHub

Forums