Refetching queries in Apollo Client

Apollo Client allows you to make local modifications to your GraphQL data by updating the cache, but sometimes it's more straightforward to update your client-side GraphQL data by refetching queries from the server.

In theory, you could refetch every active query after a client-side update, but you can save time and network bandwidth by refetching queries more selectively. The InMemoryCache helps you determine which active queries might have been invalidated by recent cache updates.

Local cache updates and refetching work especially well in combination: your application can display the results of local cache modifications immediately, while also refetching in the background to obtain the very latest data from the server. The UI is then rerendered only if there are differences between local data and refetched data.

Refetching is especially common after a mutation, so mutate functions accept options like refetchQueries and onQueryUpdated to specify which queries should be refetched, and how.

To selectively refetch queries outside of a mutation, you instead use the refetchQueries method of ApolloClient, which is documented here.

client.refetchQueries

This method is new in Apollo Client 3.4.

Refetch options

The client.refetchQueries method take an options object that conforms to the following TypeScript interface:

1interface RefetchQueriesOptions<
2  TCache extends ApolloCache<any>,
3  TResult = Promise<ApolloQueryResult<any>>,
4> {
5  updateCache?: (cache: TCache) => void;
6  include?: Array<string | DocumentNode> | "all" | "active";
7  onQueryUpdated?: (
8    observableQuery: ObservableQuery<any>,
9    diff: Cache.DiffResult<any>,
10    lastDiff: Cache.DiffResult<any> | undefined,
11  ) => boolean | TResult;
12  optimistic?: boolean;
13}

These fields are described below:

Refetch results

The client.refetchQueries method collects the TResult results returned by onQueryUpdated, defaulting to TResult = Promise<ApolloQueryResult<any>> if onQueryUpdated is not provided. It combines those results into a single Promise<TResolved[]> using Promise.all(results).

Thanks to the Promise-unwrapping behavior of Promise.all, this TResolved type is often the same type as TResult, except when TResult is a PromiseLike<TResolved> or a boolean.

The returned Promise object has two other useful properties:

These two arrays parallel each other: they have the same length, and results[i] is the result produced by onQueryUpdated when called with the ObservableQuery found at queries[i], for any index i.

Refetch recipes

Refetching a specific query

To refetch a specific query by name, use the include option by itself:

1await client.refetchQueries({
2  include: ["SomeQueryName"],
3});

The include option can also refetch a specific query using its DocumentNode:

1await client.refetchQueries({
2  include: [SOME_QUERY],
3});

Refetching all queries

To refetch all active queries, pass the "active" shorthand for include:

1await client.refetchQueries({
2  include: "active",
3});

To refetch all queries managed by Apollo Client (even those with no observers, or whose components are currently unmounted), pass "all" for include:

1await client.refetchQueries({
2  include: "all", // Consider using "active" instead!
3});

Refetching queries affected by cache updates

You can refetch queries affected by cache updates performed in the updateCache callback:

1await client.refetchQueries({
2  updateCache(cache) {
3    cache.evict({ fieldName: "someRootField" });
4  },
5});

This refetches any queries that depend on Query.someRootField, without requiring you to know in advance which queries might be included. Any combination of cache operations (writeQuery, writeFragment, modify, evict, etc.) is allowed within updateCache.

Updates performed by updateCache persist in the cache by default. You can perform them in a temporary optimistic layer instead, if you want them to be discarded immediately after client.refetchQueries is done observing them, leaving the cache unchanged:

1await client.refetchQueries({
2  updateCache(cache) {
3    cache.evict({ fieldName: "someRootField" });
4  },
5
6  // Evict Query.someRootField only temporarily, in an optimistic layer.
7  optimistic: true,
8});

Another way to "update" the cache without actually changing cache data is to use cache.modify and its INVALIDATE sentinel object:

1await client.refetchQueries({
2  updateCache(cache) {
3    cache.modify({
4      fields: {
5        someRootField(value, { INVALIDATE }) {
6          // Update queries that involve Query.someRootField, without actually
7          // changing its value in the cache.
8          return INVALIDATE;
9        },
10      },
11    });
12  },
13});

Before client.refetchQueries was introduced, the INVALIDATE sentinel was not very useful, because invalidated queries with fetchPolicy: "cache-first" would typically re-read unchanged results, and therefore decide not to perform a network request. The client.refetchQueries method makes this invalidation system more accessible to application code, so you can control the refetching behavior of invalidated queries.

In all of the examples above, whether we use include or updateCache, client.refetchQueries refetches affected queries from the network and includes the resulting Promise<ApolloQueryResult<any>> results in the Promise<TResolved[]> returned by client.refetchQueries.

If a particular query is included both by include and by updateCache, that query is refetched only once. In other words, the include option is a good way to make sure certain queries are always included, no matter which queries are included by updateCache.

Refetching selectively

In development, you probably want to make sure the appropriate queries are getting refetched, rather than blindly refetching them. To intercept each query before refetching, you can specify an onQueryUpdated callback:

1const results = await client.refetchQueries({
2  updateCache(cache) {
3    cache.evict({ fieldName: "someRootField" });
4  },
5
6  onQueryUpdated(observableQuery) {
7    // Logging and/or debugger breakpoints can be useful in development to
8    // understand what client.refetchQueries is doing.
9    console.log(`Examining ObservableQuery ${observableQuery.queryName}`);
10    debugger;
11
12    // Proceed with the default refetching behavior, as if onQueryUpdated
13    // was not provided.
14    return true;
15  },
16});
17
18results.forEach(result => {
19  // These results will be ApolloQueryResult<any> objects, after all
20  // results have been refetched from the network.
21});

Notice how adding onQueryUpdated in this example did not change the refetching behavior of client.refetchQueries, allowing us to use onQueryUpdated purely for diagnostic or debugging purposes.

If you want to skip certain queries that would otherwise be included, return false from onQueryUpdated:

1await client.refetchQueries({
2  updateCache(cache) {
3    cache.evict({ fieldName: "someRootField" });
4  },
5
6  onQueryUpdated(observableQuery, { complete, result, missing }) {
7    console.log(`Examining ObservableQuery ${
8      observableQuery.queryName
9    } whose latest result is ${JSON.stringify(result)} which is ${
10      complete ? "complete" : "incomplete"
11    }`);
12
13    if (shouldIgnoreQuery(observableQuery)) {
14      return false;
15    }
16
17    // Refetch the query unconditionally from the network.
18    return true;
19  },
20});

In case the ObservableQuery does not provide enough information, you can also examine the latest result for the query, along with information about its completeness and missing fields, using the Cache.DiffResult object passed as the second parameter to onQueryUpdated:

1await client.refetchQueries({
2  updateCache(cache) {
3    cache.evict({ fieldName: "someRootField" });
4  },
5
6  onQueryUpdated(observableQuery, { complete, result, missing }) {
7    if (shouldIgnoreQuery(observableQuery)) {
8      return false;
9    }
10
11    if (complete) {
12      // Update the query according to its chosen FetchPolicy, rather than
13      // refetching it unconditionally from the network.
14      return observableQuery.reobserve();
15    }
16
17    // Refetch the query unconditionally from the network.
18    return true;
19  },
20});

Because onQueryUpdated has the ability to filter queries dynamically, it also pairs well with the bulk include options mentioned above:

1await client.refetchQueries({
2  // Include all active queries by default, which may be ill-advised unless
3  // you also use onQueryUpdated to filter those queries.
4  include: "active";
5
6  // Called once for every active query, allowing dynamic filtering:
7  onQueryUpdated(observableQuery) {
8    return !shouldIgnoreQuery(observableQuery);
9  },
10});

Handling refetch errors

In the examples above, we await client.refetchQueries(...) to find out the final ApolloQueryResult<any> results for all the refetched queries. This combined promise is created with Promise.all, so a single failure rejects the entire Promise<TResolved[]>, potentially hiding other successful results. If this is a problem, you can use the queries and results arrays returned by client.refetchQueries instead of (or in addition to) awaiting the Promise:

1const { queries, results } = client.refetchQueries({
2  // Specific client.refetchQueries options are not relevant to this example.
3});
4
5const finalResults = await Promise.all(
6  results.map((result, i) => {
7    return Promise.resolve(result).catch(error => {
8      console.error(`Error refetching query ${queries[i].queryName}: ${error}`);
9      return null; // Silence this Promise rejection.
10    });
11  })
12});

In the future, just as additional input options may be added to the client.refetchQueries method, additional properties may be added to its result object, supplementing its Promise-related properties and the queries and results arrays.

If you discover that some specific additional client.refetchQueries input options or result properties would be useful, please feel free to open an issue or start a discussion explaining your use case(s).

Corresponding client.mutate options

For refetching after a mutation, client.mutate supports options similar to client.refetchQueries, which you should use instead of client.refetchQueries, because it's important for refetching logic to happen at specific times during the mutation process.

For historical reasons, client.mutate options have slightly different names from the new client.refetchQueries options, but their internal implementation is substantially the same, so you can translate between them using the following table: