Launch Apollo Studio

Core pagination API

Fetching and caching paginated results


Regardless of which pagination strategy your GraphQL server uses for a particular list field, your Apollo Client app needs to do the following to query that field effectively:

This article describes these core requirements for paginated fields.

The fetchMore function

Pagination always involves sending followup queries to your GraphQL server to obtain additional pages of results. In Apollo Client, the recommended way to send these followup queries is with the fetchMore function. This function is a member of the ObservableQuery object returned by client.watchQuery, and it's also provided by the useQuery hook:

FeedWithData.jsx
const FEED_QUERY = gql`
  query Feed($offset: Int, $limit: Int) {
    feed(offset: $offset, limit: $limit) {
      id
      # ...
    }
  }
`;

const FeedWithData() {
  const { loading, data, fetchMore } = useQuery(FEED_QUERY, {    variables: {
      offset: 0,
      limit: 10
    },
  });
  // ...continues below...
}

You usually call fetchMore in response to a user action, such as clicking a button or scrolling to the current "bottom" of an infinite-scroll feed.

By default, fetchMore executes a query with the exact same shape and variables as your original query. You can pass new values for the query's variables (such as providing a new offset) like so:

FeedWithData.jsx
const FeedWithData() {
// ...continuing from above...

// If you want your component to rerender with loading:true whenever
// fetchMore is called, add `notifyOnNetworkStatusChange:true` to the
// options you pass to useQuery.
if (loading) return 'Loading...';

return (
    <Feed
      entries={data.feed || []}
      onLoadMore={() => fetchMore({        variables: {          offset: data.feed.length        },      })}    />
  );
}

Here, we set the offset variable to feed.length to fetch items after the last item in our cached list. The variables we provide here are merged with the variables provided for the original query, which means that variables omitted here (such as limit) retain their original value (10) in the followup query.

In addition to variables, you can optionally provide an entirely different shape of query to execute. This can be useful when fetchMore needs to fetch only a single paginated field, but the original query contained unrelated fields.

Additional examples of using fetchMore are provided in the detailed documentation for offset-based pagination and cursor-based pagination.

Our fetchMore function is ready, but we're not finished! The cache doesn't know yet that it should merge our followup query's result with the original query's result. Instead, it will store the two results as two completely separate lists. To resolve this, let's move on to Merging paginated results.

Merging paginated results

The examples in this section use offset-based pagination, but this article applies to all pagination strategies.

As mentioned above, a fetchMore followup query doesn't automatically merge its result with the orignal query's cached result. To achieve this behavior, we need to define a field policy for our paginated field.

Why do I need a field policy?

Let's say we have a field in our GraphQL schema that takes an argument:

type Query {
  user(id: ID!): User}

Now, let's say we execute the following query two times and provide different values for the $id variable each time:

query GetUser($id: ID!) {
  user(id: $id) {
    id
    name
  }
}

Our two queries return two entirely different User objects. Helpfully, the Apollo Client cache automatically stores these two objects separately, because it sees that different values were provided for at least one field argument (id). Otherwise, the cache might overwrite the first User object with the second User object, and we want to cache both!

Now, let's say we execute this query two times, with different values for the $offset variable:

query Feed($offset: Int, $limit: Int) {
  feed(offset: $offset, limit: $limit) {
    id
    # ...
  }
}

In this case, we're querying a paginated list field twice to obtain two different pages of results, and we want those two pages to be merged. But the cache doesn't know that! It sees no difference between this scenario and the User scenario above, so it stores the results as two completely separate lists.

With field policies, we can modify the cache's behavior for individual fields that require it. For example, we can tell the cache not to store separate results for the feed field based on the values of offset and limit. Let's look at how.

Defining a field policy

A field policy specifies how a particular field in your InMemoryCache is read and written. You can define a field policy to merge the results of paginated queries into a single list.

Example

Here's the server-side schema for our message feed application that uses offset-based pagination:

type Query {
  feed(offset: Int, limit: Int): [FeedItem!]}

type FeedItem {
  id: String!
  message: String!
}

In our client, we want to define a field policy for Query.feed so that all returned pages of the list are merged into a single list in our cache.

We define our field policy within the typePolicies option we provide the InMemoryCache constructor:

const cache = new InMemoryCache({
  typePolicies: {
    Query: {
      fields: {
        feed: {          // Don't cache separate results based on          // any of this field's arguments.          keyArgs: false,          // Concatenate the incoming list items with          // the existing list items.          merge(existing = [], incoming) {            return [...existing, ...incoming];          },        }      }
    }
  }
})

This field policy specifies the field's keyArgs, along with a merge function. Both of these configurations are necessary for handling pagination:

  • keyArgs specifies which of the field's arguments cause the cache to store a separate value for each unique combination of those arguments.
    • In our case, the cache shouldn't store a separate result based on any argument value (offset or limit). So, we disable this behavior entirely by passing false. An empty array (keyArgs: []) also works, but keyArgs: false is more expressive, and it results in a cleaner field key within the cache (feed in this case).
    • If a particular argument's value could cause items from an entirely different list to be returned in the field, that argument should be included in keyArgs.
    • For more information, see Specifying key arguments and The keyArgs API.
  • A merge function tells the Apollo Client cache how to combine incoming data with existing cached data for a particular field. Without this function, incoming field values overwrite existing field values by default.

With this field policy in place, the cache automatically merges the results of all queries that use the following structure, regardless of argument values:

// Client-side query definition
const FEED_QUERY = gql`
  query Feed($offset: Int, $limit: Int) {
    feed(offset: $offset, limit: $limit) {
      id
      message
    }
  }
`;

Improving the merge function

In the example above, our merge function is a single line:

merge(existing = [], incoming) {
  return [...existing, ...incoming];
}

This function makes risky assumptions about the order in which the client requests pages, because it ignores the values of offset and limit. A more robust merge function can use options.args to decide where to put incoming data relative to existing data, like so:

const cache = new InMemoryCache({
  typePolicies: {
    Query: {
      fields: {
        feed: {
          keyArgs: [],
          merge(existing, incoming, { args: { offset = 0 }}) {
            // Slicing is necessary because the existing data is
            // immutable, and frozen in development.
            const merged = existing ? existing.slice(0) : [];
            for (let i = 0; i < incoming.length; ++i) {
              merged[offset + i] = incoming[i];
            }
            return merged;
          },
        },
      },
    },
  },
});

This logic handles sequential page writes the same way the single-line strategy does, but it can also tolerate repeated, overlapping, or out-of-order writes, without duplicating any list items.

read functions for paginated fields

As shown above, a merge function helps you combine paginated query results from your GraphQL server into a single list in your client cache. But what if you also want to configure how that locally cached list is read? For that, you can define a read function.

You define a read function for a field within its field policy, alongside the merge function and keyArgs. If you define a read function for a field, the cache calls that function whenever you query the field, passing the field's existing cached value (if any) as the first argument. In the query response, the field is populated with the read function's return value, instead of the existing cached value.

If a field policy includes both a merge function and a read function, the default value of keyArgs becomes false (i.e., no arguments are key arguments). If either function isn't defined, all of the field's arguments are considered key arguments by default. In either case, you can define keyArgs yourself to override the default behavior.

A read function for a paginated field typically uses one of the following approaches:

  • Re-pagination, in which the cached list is split back into pages, based on field arguments
  • No pagination, in which the cached list is always returned in full

Although the "right" approach varies from field to field, a non-paginated read function often works best for infinitely scrolling feeds, because it gives your code full control over which elements to display at a given time, without requiring any additional cache reads.

Paginated read functions

The read function for a list field can perform client-side re-pagination for that list. It can even transform a page before returning it, such as by sorting or filtering its elements.

This capability goes beyond returning the same pages that you fetched from your server, because a read function for offset/limit pagination could read from any available offset, with any desired limit:

const cache = new InMemoryCache({
  typePolicies: {
    Query: {
      fields: {
        feed: {
          read(existing, { args: { offset, limit }}) {
            // A read function should always return undefined if existing is
            // undefined. Returning undefined signals that the field is
            // missing from the cache, which instructs Apollo Client to
            // fetch its value from your GraphQL server.
            return existing && existing.slice(offset, offset + limit);
          },

          // The keyArgs list and merge function are the same as above.
          keyArgs: [],
          merge(existing, incoming, { args: { offset = 0 }}) {
            const merged = existing ? existing.slice(0) : [];
            for (let i = 0; i < incoming.length; ++i) {
              merged[offset + i] = incoming[i];
            }
            return merged;
          },
        },
      },
    },
  },
});

Depending on the assumptions you feel comfortable making, you might want to make this code more defensive. For example, you can provide default values for offset and limit, in case someone fetches Query.feed without providing arguments:

const cache = new InMemoryCache({
  typePolicies: {
    Query: {
      fields: {
        feed: {
          read(existing, {
            args: {
              // Default to returning the entire cached list,
              // if offset and limit are not provided.
              offset = 0,
              limit = existing?.length,
            } = {},
          }) {
            return existing && existing.slice(offset, offset + limit);
          },
          // ... keyArgs, merge ...
        },
      },
    },
  },
});

This style of read function takes responsibility for re-paginating your data based on field arguments, essentially inverting the behavior of your merge function. This way, your application can query different pages using different arguments.

Non-paginated read functions

The read function for a paginated field can choose to ignore arguments like offset and limit, and always return the entire list as it exists in the cache. In this case, your application code takes responsibility for slicing the list into pages depending on your needs.

If you adopt this approach, you might not need to define a read function at all, because the cached list can be returned without any processing. That's why the offsetLimitPagination helper function is implemented without a read function.

Regardless of how you configure keyArgs, your read (and merge) functions can always examine any arguments passed to the field using the options.args object. See The keyArgs API for a deeper discussion of how to reason about dividing argument-handling responsibility between keyArgs and your read or merge functions.

Edit on GitHub