Customizing the behavior of cached fields

You can customize how a particular field in your Apollo Client cache is read and written. To do so, you define a field policy for the field. A field policy can include:

• A read function that specifies what happens when the field's cached value is read
• A merge function that specifies what happens when field's cached value is written
• An array of key arguments that help the cache avoid storing unnecessary duplicate data.

You provide field policies to the constructor of InMemoryCache. Each field policy is defined inside whatever TypePolicy object corresponds to the type that contains the field. The following example defines a field policy for the name field of a Person type:

const cache = new InMemoryCache({
  typePolicies: {
    Person: {
      fields: {
        name: {          read(name) {            // Return the cached name, transformed to upper case            return name.toUpperCase();          }        }      },
    },
  },
});

The field policy above defines a read function that specifies what the cache returns whenever Person.name is queried.

The read function

If you define a read function for a field, the cache calls that function whenever your client queries for the field. In the query response, the field is populated with the read function's return value, instead of the field's cached value.

The first parameter of a read function provides the field's currently cached value, if one exists. You can use this to help determine the function's return value.

The second parameter is an object that provides access to several properties and helper functions, which are explained in the FieldPolicy API reference.

The following read function assigns a default value of UNKNOWN NAME to the name field of a Person type, if the actual value is not available in the cache. In all other cases, the cached value is returned.

const cache = new InMemoryCache({
  typePolicies: {
    Person: {
      fields: {
        name: {
          read(name = "UNKNOWN NAME") {
            return name;
          }
        },
      },
    },
  },
});

If a field accepts arguments, the second parameter includes the values of those arguments. The following read function checks to see if the maxLength argument is provided when the name field is queried. If it is, the function returns only the first maxLength characters of the person's name. Otherwise, the person's full name is returned.

const cache = new InMemoryCache({
  typePolicies: {
    Person: {
      fields: {
        // If a field's TypePolicy would only include a read function,
        // you can optionally define the function like so, instead of
        // nesting it inside an object as shown in the example above.
        name(name: string, { args }) {
          if (args && typeof args.maxLength === "number") {
            return name.substring(0, args.maxLength);
          }
          return name;
        },
      },
    },
  },
});

You can define a read function for a field that isn't even defined in your schema. For example, the following read function enables you to query a userId field that is always populated with locally stored data:

const cache = new InMemoryCache({
  typePolicies: {
    Person: {
      fields: {
        userId() {
          return localStorage.loggedInUserId;
        },
      },
    },
  },
});

Note that to query for a field that is only defined locally, your query should include the @client directive on that field so that Apollo Client doesn't include it in requests to your GraphQL server.

Other use cases for a read function include:

• Transforming cached data to suit your client's needs, such as rounding floating-point values to the nearest integer
• Deriving local-only fields from one or more schema fields on the same object (such as deriving an age field from a birthDate field)
• Deriving local-only fields from one or more schema fields across multiple objects

For a full list of the options provided to the read function, see the API reference. You will almost never need to use all of these options, but each one has an important role when reading fields from the cache.

The merge function

If you define a merge function for a field, the cache calls that function whenever the field is about to be written with an incoming value (such as from your GraphQL server). When the write occurs, the field's new value is set to the merge function's return value, instead of the original incoming value.

Merging arrays

A common use case for a merge function is to define how to write to a field that holds an array. By default, the field's existing array is completely replaced by the incoming array. Often, it's preferable to concatenate the two arrays instead, like so:

const cache = new InMemoryCache({
  typePolicies: {
    Agenda: {
      fields: {
        tasks: {
          merge(existing = [], incoming: any[]) {
            return [...existing, ...incoming];
          },
        },
      },
    },
  },
});

Note that existing is undefined the very first time this function is called for a given instance of the field, because the cache does not yet contain any data for the field. Providing the existing = [] default parameter is a convenient way to handle this case.

Your merge function cannot push the incoming array directly onto the existing array. It must instead return a new array to prevent potential errors. In development mode, Apollo Client prevents unintended modification of the existing data with Object.freeze.

Merging non-normalized objects

Another common use case for custom field merge functions is to combine nested objects that do not have IDs, but are known (by you, the application developer) to represent the same logical object, assuming the parent object is the same.

Suppose that a Book type has an author field, which is an object containing information like the author's name, language, and dateOfBirth. The Book object has __typename: "Book" and a unique isbn field, so the cache can tell when two Book result objects represent the same logical entity. However, for whatever reason, the query that retrieved this Book did not ask for enough information about the book.author object. Perhaps no keyFields were specified for the Author type, and there is no default id field.

This lack of identifying information poses a problem for the cache, because it cannot determine automatically whether two Author result objects are the same. If multiple queries ask for different information about the author of this Book, the order of the queries matters, because the favoriteBook.author object from the second query cannot be safely merged with the favoriteBook.author object from the first query, and vice-versa:

query BookWithAuthorName {
  favoriteBook {
    isbn
    title
    author {
      name
    }
  }
}

query BookWithAuthorLanguage {
  favoriteBook {
    isbn
    title
    author {
      language
    }
  }
}

In such situations, the cache defaults to replacing the existing favoriteBook.author data with the incoming data, without merging the name and language fields together, because the risk of merging inconsistent name and language fields from different authors is unacceptable.

Note: Apollo Client 2.x would sometimes merge unidentified objects. While this behavior might accidentally have aligned with the intentions of the developer, it led to subtle inconsistencies within the cache. Apollo Client 3.0 refuses to perform unsafe merges, and instead warns about potential loss of unidentified data.

You could fix this problem by modifying your queries to request an id field for the favoriteBook.author objects, or by specifying custom keyFields in the Author type policy, such as ["name", "dateOfBirth"]. Providing the cache with this information allows it to know when two Author objects represent the same logical entity, so it can safely merge their fields. This solution is recommended, when feasible.

However, you may encounter situations where your data graph does not provide any uniquely identifying fields for Author objects. In these rare scenarios, it might be safe to assume that a given Book has one and only one primary Author, and the author never changes. In other words, the identity of the author is implied by the identity of the book. This common-sense knowledge is something you have at your disposal, as a human, but it must be communicated to the cache, which is neither human nor capable of telepathy.

In such situations, you can define a custom merge function for the author field within the type policy for Book:

const cache = new InMemoryCache({
  typePolicies: {
    Book: {
      fields: {
        author: {
          merge(existing, incoming) {
            // Better, but not quite correct.
            return { ...existing, ...incoming };
          },
        },
      },
    },
  },
});

Alternatively, if you prefer to keep the default behavior of completely replacing the existing data with the incoming data, while also silencing the warnings, the following merge function will explicitly permit replacement:

const cache = new InMemoryCache({
  typePolicies: {
    Book: {
      fields: {
        author: {
          merge(existing, incoming) {
            // Equivalent to what happens if there is no custom merge function.
            return incoming;
          },
        },
      },
    },
  },
});

Since writing this kind of merge function can become repetitive, the following shorthand will provide the same behavior:

const cache = new InMemoryCache({
  typePolicies: {
    Book: {
      fields: {
        author: {
          // Short for always preferring incoming over existing data.
          merge: false,
        },
      },
    },
  },
});

When you use { ...existing, ...incoming }, Author objects with differing fields (name, dateOfBirth) can be combined without losing fields, which is definitely an improvement over blind replacement.

But what if the Author type defines its own custom merge functions for fields of the incoming object? Since we're using object spread syntax, such fields will immediately overwrite fields in existing, without triggering any nested merge functions. The { ...existing, ...incoming } syntax may be an improvement, but it is not fully correct.

Fortunately, you can find a helper function called options.mergeObjects in the options passed to the merge function, which generally behaves the same as { ...existing, ...incoming }, except when the incoming fields have custom merge functions. When options.mergeObjects encounters custom merge functions for any of the fields in its second argument (incoming), those nested merge functions will be called before combining the fields of existing and incoming, as desired:

const cache = new InMemoryCache({
  typePolicies: {
    Book: {
      fields: {
        author: {
          merge(existing, incoming, { mergeObjects }) {
            // Correct, thanks to invoking nested merge functions.
            return mergeObjects(existing, incoming);
          },
        },
      },
    },
  },
});

Because this Book.author field policy has no Book- or Author-specific logic in it, you can reuse this merge function for any field that needs this kind of handling.

Since writing this kind of merge function can become repetitive, the following shorthand will provide the same behavior:

const cache = new InMemoryCache({
  typePolicies: {
    Book: {
      fields: {
        author: {
          // Short for options.mergeObjects(existing, incoming).
          merge: true,
        },
      },
    },
  },
});

In summary, the Book.author policy above allows the cache to safely merge the author objects of any two Book objects that have the same identity.

Merging arrays of non-normalized objects

Once you're comfortable with the ideas and recommendations from the previous section, consider what happens when a Book can have multiple authors:

query BookWithAuthorNames {
  favoriteBook {
    isbn
    title
    authors {
      name
    }
  }
}

query BookWithAuthorLanguages {
  favoriteBook {
    isbn
    title
    authors {
      language
    }
  }
}

In this case, the favoriteBook.authors field is no longer just a single object, but an array of authors, so it's even more imporant to define a custom merge function to prevent loss of data by replacement:

const cache = new InMemoryCache({
  typePolicies: {
    Book: {
      fields: {
        authors: {
          merge(existing: any[], incoming: any[], { readField, mergeObjects }) {
            const merged: any[] = existing ? existing.slice(0) : [];
            const authorNameToIndex: Record<string, number> = Object.create(null);
            if (existing) {
              existing.forEach((author, index) => {
                authorNameToIndex[readField<string>("name", author)] = index;
              });
            }
            incoming.forEach(author => {
              const name = readField<string>("name", author);
              const index = authorNameToIndex[name];
              if (typeof index === "number") {
                // Merge the new author data with the existing author data.
                merged[index] = mergeObjects(merged[index], author);
              } else {
                // First time we've seen this author in this array.
                authorNameToIndex[name] = merged.length;
                merged.push(author);
              }
            });
            return merged;
          },
        },
      },
    },
  },
});

Instead of blindly replacing the existing authors array with the incoming array, this code concatenates the arrays together, while also checking for duplicate author names, merging the fields of any repeated author objects.

The readField helper function is more robust than using author.name, because it also tolerates the possibility that the author is a Reference object referring to data elsewhere in the cache, which could happen if you (or someone else on your team) eventually gets around to specifying keyFields for the Author type.

As this example suggests, merge functions can become quite sophisticated. When this happens, you can often extract the generic logic into a reusable helper function:

const cache = new InMemoryCache({
  typePolicies: {
    Book: {
      fields: {
        authors: {
          merge: mergeArrayByField<AuthorType>("name"),
        },
      },
    },
  },
});

Now that you've hidden the details behind a reusable abstraction, it no longer matters how complicated the implementation gets. This is liberating, because it allows you to improve your client-side business logic over time, while keeping related logic consistent across your entire application.

Handling pagination

When a field holds an array, it's often useful to paginate that array's results, because the total result set can be arbitrarily large.

Typically, a query includes pagination arguments that specify:

• Where to start in the array, using either a numeric offset or a starting ID
• The maximum number of elements to return in a single "page"

If you implement pagination for a field, it's important to keep pagination arguments in mind if you then implement read and merge functions for the field:

const cache = new InMemoryCache({
  typePolicies: {
    Agenda: {
      fields: {
        tasks: {
          merge(existing: any[], incoming: any[], { args }) {
            const merged = existing ? existing.slice(0) : [];
            // Insert the incoming elements in the right places, according to args.
            const end = args.offset + Math.min(args.limit, incoming.length);
            for (let i = args.offset; i < end; ++i) {
              merged[i] = incoming[i - args.offset];
            }
            return merged;
          },

          read(existing: any[], { args }) {
            // If we read the field before any data has been written to the
            // cache, this function will return undefined, which correctly
            // indicates that the field is missing.
            const page = existing && existing.slice(
              args.offset,
              args.offset + args.limit,
            );
            // If we ask for a page outside the bounds of the existing array,
            // page.length will be 0, and we should return undefined instead of
            // the empty array.
            if (page && page.length > 0) {
              return page;
            }
          },
        },
      },
    },
  },
});

As this example shows, your read function often needs to cooperate with your merge function, by handling the same arguments in the inverse direction.

If you want a given "page" to start after a specific entity ID instead of starting from args.offset, you can implement your merge and read functions as follows, using the readField helper function to examine existing task IDs:

const cache = new InMemoryCache({
  typePolicies: {
    Agenda: {
      fields: {
        tasks: {
          merge(existing: any[], incoming: any[], { args, readField }) {
            const merged = existing ? existing.slice(0) : [];
            // Obtain a Set of all existing task IDs.
            const existingIdSet = new Set(
              merged.map(task => readField("id", task)));
            // Remove incoming tasks already present in the existing data.
            incoming = incoming.filter(
              task => !existingIdSet.has(readField("id", task)));
            // Find the index of the task just before the incoming page of tasks.
            const afterIndex = merged.findIndex(
              task => args.afterId === readField("id", task));
            if (afterIndex >= 0) {
              // If we found afterIndex, insert incoming after that index.
              merged.splice(afterIndex + 1, 0, ...incoming);
            } else {
              // Otherwise insert incoming at the end of the existing data.
              merged.push(...incoming);
            }
            return merged;
          },

          read(existing: any[], { args, readField }) {
            if (existing) {
              const afterIndex = existing.findIndex(
                task => args.afterId === readField("id", task));
              if (afterIndex >= 0) {
                const page = existing.slice(
                  afterIndex + 1,
                  afterIndex + 1 + args.limit,
                );
                if (page && page.length > 0) {
                  return page;
                }
              }
            }
          },
        },
      },
    },
  },
});

Note that if you call readField(fieldName), it returns the value of the specified field from the current object. If you pass an object as a second argument to readField, (e.g., readField("id", task)), readField instead reads the specified field from the specified object. In the above example, reading the id field from existing Task objects allows us to deduplicate the incoming task data.

The pagination code above is complicated, but after you implement your preferred pagination strategy, you can reuse it for every field that uses that strategy, regardless of the field's type. For example:

function afterIdLimitPaginatedFieldPolicy<T>() {
  return {
    merge(existing: T[], incoming: T[], { args, readField }): T[] {
      ...
    },
    read(existing: T[], { args, readField }): T[] {
      ...
    },
  };
}

const cache = new InMemoryCache({
  typePolicies: {
    Agenda: {
      fields: {
        tasks: afterIdLimitPaginatedFieldPolicy<Reference>(),
      },
    },
  },
});

Specifying key arguments

If a field accepts arguments, you can specify an array of keyArgs in the field's FieldPolicy. This array indicates which arguments are key arguments that are used to calculate the field's value. Specifying this array can help reduce the amount of duplicate data in your cache.

Example

Let's say your schema's Query type includes a monthForNumber field. This field returns the details of particular month, given a provided number argument (January for 1 and so on). The number argument is a key argument for this field, because it is used when calculating the field's result:

const cache = new InMemoryCache({
  typePolicies: {
    Query: {
      fields: {
        monthForNumber: {
          keyArgs: ["number"],
        },
      },
    },
  },
});

An example of a non-key argument is an access token, which is used to authorize a query but not to calculate its result. If monthForNumber also accepts an accessToken argument, the value of that argument does not affect which month's details are returned.

By default, the cache stores a separate value for every unique combination of argument values you provide when querying a particular field. When you specify a field's key arguments, the cache understands that any non-key arguments don't affect that field's value. Consequently, if you execute two different queries with the monthForNumber field, passing the same number argument but different accessToken arguments, the second query response will overwrite the first, because both invocations use the exact same value for all key arguments.

If you need more control over the behavior of keyArgs, you can pass a function instead of an array. This keyArgs function will receive the arguments object as its first parameter, and a context object providing other relevant details as its second parameter. See KeyArgsFunction in the types below for further information.

FieldPolicy API reference

Here are the definitions for the FieldPolicy type and its related types:

// These generic type parameters will be inferred from the provided policy in
// most cases, though you can use this type to constrain them more precisely.
type FieldPolicy<
  TExisting,
  TIncoming = TExisting,
  TReadResult = TExisting,
> = {
  keyArgs?: KeySpecifier | KeyArgsFunction | false;
  read?: FieldReadFunction<TExisting, TReadResult>;
  merge?: FieldMergeFunction<TExisting, TIncoming> | boolean;
};

type KeySpecifier = (string | KeySpecifier)[];

type KeyArgsFunction = (
  args: Record<string, any> | null,
  context: {
    typename: string;
    fieldName: string;
    field: FieldNode | null;
    variables?: Record<string, any>;
  },
) => string | KeySpecifier | null | void;

type FieldReadFunction<TExisting, TReadResult = TExisting> = (
  existing: Readonly<TExisting> | undefined,
  options: FieldFunctionOptions,
) => TReadResult;

type FieldMergeFunction<TExisting, TIncoming = TExisting> = (
  existing: Readonly<TExisting> | undefined,
  incoming: Readonly<TIncoming>,
  options: FieldFunctionOptions,
) => TExisting;

// These options are common to both read and merge functions:
interface FieldFunctionOptions {
  cache: InMemoryCache;

  // The final argument values passed to the field, after applying variables.
  // If no arguments were provided, this property will be null.
  args: Record<string, any> | null;

  // The name of the field, equal to options.field.name.value when
  // options.field is available. Useful if you reuse the same function for
  // multiple fields, and you need to know which field you're currently
  // processing. Always a string, even when options.field is null.
  fieldName: string;

  // The FieldNode object used to read this field. Useful if you need to
  // know about other attributes of the field, such as its directives. This
  // option will be null when a string was passed to options.readField.
  field: FieldNode | null;

  // The variables that were provided when reading the query that contained
  // this field. Possibly undefined, if no variables were provided.
  variables?: Record<string, any>;

  // Easily detect { __ref: string } reference objects.
  isReference(obj: any): obj is Reference;

  // Returns a Reference object if obj can be identified, which requires,
  // at minimum, a __typename and any necessary key fields. If true is
  // passed for the optional mergeIntoStore argument, the object's fields
  // will also be persisted into the cache, which can be useful to ensure
  // the Reference actually refers to data stored in the cache. If you
  // pass an ID string, toReference will make a Reference out of it. If
  // you pass a Reference, toReference will return it as-is.
  toReference(
    objOrIdOrRef: StoreObject | string | Reference,
    mergeIntoStore?: boolean,
  ): Reference | undefined;

  // Helper function for reading other fields within the current object.
  // If a foreign object or reference is provided, the field will be read
  // from that object instead of the current object, so this function can
  // be used (together with isReference) to examine the cache outside the
  // current object. If a FieldNode is passed instead of a string, and
  // that FieldNode has arguments, the same options.variables will be used
  // to compute the argument values. Note that this function will invoke
  // custom read functions for other fields, if defined. Always returns
  // immutable data (enforced with Object.freeze in development).
  readField<T = StoreValue>(
    nameOrField: string | FieldNode,
    foreignObjOrRef?: StoreObject | Reference,
  ): T;

  // Returns true for non-normalized StoreObjects and non-dangling
  // References, indicating that readField(name, objOrRef) has a chance of
  // working. Useful for filtering out dangling references from lists.
  canRead(value: StoreValue): boolean;

  // A handy place to put field-specific data that you want to survive
  // across multiple read function calls. Useful for field-level caching,
  // if your read function does any expensive work.
  storage: Record<string, any>;

  // Instead of just merging objects with { ...existing, ...incoming }, this
  // helper function can be used to merge objects in a way that respects any
  // custom merge functions defined for their fields.
  mergeObjects<T extends StoreObject | Reference>(
    existing: T,
    incoming: T,
  ): T | undefined;
}