Here's your chance to speak at GraphQL Summit in New York City, October 8 - 10, 2024! 🏙️ Submit your proposal by May 31.
Launch GraphOS Studio
You're viewing documentation for a previous version of this software. Switch to the latest stable version.

Client-side caching

As mentioned in the introduction, does more than simply run your queries against a . It normalizes results to construct a client-side cache of your data, which is kept up to date as further queries and are run.

This means your UI is always internally consistent, and can be kept fully up-to-date with the state on the server with the minimum number of queries required.

Types of caches

All caches used by the ApolloClient must conform to the NormalizedCache protocol. There are two types of cache provided automatically by Apollo:

  • InMemoryNormalizedCache: This is included with the main Apollo library, and is the default caching strategy for the . This stores normalized results in-memory, so results are not persisted across sessions of the application.
  • SQLiteCache: This is included via the ApolloSQLite library. This writes out cache results to a SQLite file rather than holding the results in memory. Note that this in turn causes cache hits to go to disk, which may result in somewhat slower responses. However, this also reduces the chances of unbounded memory growth, since everything gets dumped to disk.

All caches can be cleared in their entirety by calling clear(callbackQueue:completion:). If you need to work more directly with the cache, please see the Direct Cache Access section.

Cache Setup

In-Memory Cache

For InMemoryNormalizedCache, no sub-libraries are needed.

This type of cache is used by default when setting up an ApolloClient. If you want to use an in-memory cache without modifications, all you have to do is instantiate an ApolloClient instance and not pass anything into the store parameter.

If for some reason you find you need to instantiate the in-memory cache yourself, you can do so with one line:

Cache Setup
import Apollo
let cache = InMemoryNormalizedCache()

SQLite Cache

To use the SQLiteNormalizedCache, you need to add the ApolloSQLite sub-library to your project using your preferred package manager:

Once added, you can do the following:

  1. Set up a file URL for your SQLite file.
  2. Use that file URL to instantiate a SQLite cache.
  3. Use that SQLite cache to instantiate an ApolloStore.
  4. Pass that ApolloStore into the initializer of ApolloClient:
Client Setup
import Apollo
// NOTE: You need this import line if you are **NOT** using CocoaPods. In CocoaPods,
// ApolloSQLite files are collapsed into the Apollo framework. For other dependency managers,
// ApolloSQLite is a separate framework.
import ApolloSQLite
// 1. You'll have to figure out where to store your SQLite file.
// A reasonable place is the user's Documents directory in your sandbox.
// In any case, create a file URL for your file:
let documentsPath = NSSearchPathForDirectoriesInDomains(
let documentsURL = URL(fileURLWithPath: documentsPath)
let sqliteFileURL = documentsURL.appendingPathComponent("test_apollo_db.sqlite")
// 2. Use that file URL to instantiate the SQLite cache:
let sqliteCache = try SQLiteNormalizedCache(fileURL: sqliteFileURL)
// 3. And then instantiate an instance of `ApolloStore` with the cache you've just created:
let store = ApolloStore(cache: sqliteCache)
// 4. Assuming you've set up your `networkTransport` instance elsewhere,
// pass the store you just created into your `ApolloClient` initializer,
// and you're now set up to use the SQLite cache for persistent storage
let apolloClient = ApolloClient(networkTransport: networkTransport, store: store)

Controlling normalization

While Apollo can do basic caching based on the shape of queries and their results, Apollo won't be able to associate objects fetched by different queries without additional information about the identities of the objects returned from the server.

This is referred to as cache normalization. You can read about our caching model in detail in our blog post, "GraphQL Concepts Visualized".

By default, Apollo does not use object IDs at all, doing caching based only on the path to the object from the root query. However, if you specify a function to generate IDs from each object, and supply it as cacheKeyForObject to an ApolloClient instance, you can decide how Apollo will identify and de-duplicate the objects returned from the server:

apollo.cacheKeyForObject = { $0["id"] }

NOTE: In some cases, just using cacheKeyForObject is not enough for your application UI to update correctly. For example, if you want to add something to a list of objects without refetching the entire list, or if there are some objects that to which you can't assign an object identifier, Apollo cannot automatically update existing queries for you.

Cache normalization concepts

There are 2 primary ways you will want to manually update the cache. Either you'll want to update the cache for a query, or you will want to update a cached object directly.

Manual Scenario A

  1. You use the id of the object (after setting up the afore mentioned apollo.cacheKeyForObject = { $0["id"] }) to fetch and change the object. This will update any query where this object is referenced. This works well for updating queries which reference this object, but in the case of a create , your queries won't contain this object to update. Which leads us into Scenario B.

Manual Scenario B

  1. You fire off a mutation which creates a new object.
  2. You may then want to update the cache for a List that should contain this new object. This is a bit fiddly at the moment, as Droid for CreateDroidsMutation is strongly typed: CreateDroidsMutation.Droid. When inserting this object into the cache for ListDroidsQuery you need to init a ListDroidsQuery.Droid object from a CreateDroidsMutation.Droid or the types won't match. Your alternative to this is to manually refetch queries on a mutation which will trigger any watchers to update.

Where you may not need to manually update the cache:

If you use which contain ID's then a query which returns or mutates this fragment and returns a new state for this object will automatically be merged into your cache and any query which references that object will be updated. It may therefore be advantageous to plan your schemas so Fragments are reused in List / Detail queries and then the same Fragment is returned as the result of a mutation.

Specifying a cache policy

ApolloClient's fetch(query:) method takes an optional cachePolicy that allows you to specify when results should be fetched from the server, and when data should be loaded from the local cache.

The default cache policy is .returnCacheDataElseFetch, which means data will be loaded from the cache when available, and fetched from the server otherwise.

Other cache polices which you can specify are:

  • .fetchIgnoringCacheData to always fetch from the server, but still store results to the cache.
  • .fetchIgnoringCacheCompletely to always fetch from the server and not store results from the cache. If you're not using the cache at all, this method is preferred to fetchIgnoringCacheData for performance reasons.
  • .returnCacheDataDontFetch to return data from the cache and never fetch from the server. This policy will return an error if cached data is not available.
  • .returnCacheDataAndFetch to return cached data immediately, then perform a fetch to see if there are any updates. This is mostly useful if you're watching queries, since those will be updated when the call to the server returns.

If you're interested in returning cached data after a failed fetch, the current recommended approach is to use an additionalErrorInterceptor on your interceptor chain to examine if the error is one it makes sense to show old data for rather than something that needs to be passed on to the user, and then retrying with a returnCacheDataDontFetch retry policy. An example of this setup can be found in the Cache-dependent interceptor tests.

Watching queries

Watching a query is very similar to fetching a query. The main difference is that you don't just receive an initial result, but your result handler will be invoked whenever relevant data in the cache changes:

let watcher = HeroNameQuery(episode: .empire)) { result in
guard let data = try? result.get().data else { return }
print(data.hero?.name) // Luke Skywalker

NOTE: Remember to call cancel() on a watcher when its parent object is deallocated, or you will get a memory leak! This is not (presently) done automatically.

Direct cache access

Similarly to the Apollo React API, you can directly read and update the cache as needed using Swift's inout parameters.

This functionality is useful when performing mutations or receiving data, as you should always update the local cache to ensure consistency with the that was just performed. The ability to write to the cache directly also prevents you from needing to re-fetch data over the network after a mutation is performed.


The read function is similar to React Apollo's readQuery and React Apollo's readFragment methods and will return the cached data for a given GraphQL query or a GraphQL :

// Assuming we have defined an ApolloClient instance `client`:
// Read from a GraphQL query{ transaction in
let data = try
query: HeroNameQuery(episode: .jedi)
// Prints "R2-D2"
// Read from a GraphQL fragment{ transaction -> HeroDetails in
let data = try transaction.readObject(
ofType: HeroDetails.self,
withKey: id
// Prints "R2-D2"


The update function is similar to React Apollo's writeQuery method and will update the Apollo cache and propagate changes to all listeners (queries using the watch method):

// Assuming we have defined an ApolloClient instance `client`:
store.withinReadWriteTransaction({ transaction in
let query = HeroNameQuery(episode: .jedi)
try transaction.update(query: query) { (data: inout HeroNameQuery.Data) in
data.hero?.name = "Artoo"
let graphQLResult = try? store.load(query: query).result?.get()
// Prints "Artoo"


Delete functionality is limited at this time. There are presently three deletion methods available on ApolloStore, which are supported by both the in memory and the SQLite caches:

  1. clear - Removes everything from the cache immediately. This is basically the "Nuke it from orbit, it's the only way to be sure" option.
  2. removeObject(for:) on ReadWriteTransaction. Removes a single object for the given CacheKey.
  3. removeObjects(matching:) on ReadWriteTransaction. Removes all objects with a CacheKey that matches the given pattern. Pattern matching is not case sensitive. For an in memory cache it is the equivalent of checking whether the cache key contains the pattern and SQLite caches will perform a LIKE query to remove the objects. This method can be very slow depending on the number of records in the cache, it is recommended that this method be called in a background queue.

removeObject(for:) and removeObjects(matching:) will only remove things at the level of an object - that is, they cannot remove individual properties from an object, and cannot remove outside references to an object.

You will need to have an understanding of how you are generating cache keys to be able to use these methods effectively. If you're taking advantage of our cacheKeyForObject function, that will help significantly: You'll have a clear understanding of how cache keys are generated, so you can easily figure out how to construct a cache key or pattern to delete.

For instance, let's say your cacheKeyForObject function looks like this:

apollo.cacheKeyForObject = { $0["id"] }

This would indicate that for every object which has an id property, it will be cached with that id as the key. This would be ideal for an API which uses globally unique identifiers, as our Star Wars API does.

This means that if you have previously fetched an object with the ID "2001", you can:

  1. Call transaction.removeObject(for: "2001") on a ReadWriteTransaction to remove any object cached with that key, along with all of its associated properties and the references to other objects it stores.
  2. Call transaction.removeObjects(matching: "200") on a ReadWriteTransaction and all objects with 200 somewhere in their cache key would be removed, such as 2001, 2002, and 3200.

removeObject(for:) and removeObjects(matching:) do not cascade removals - if you remove an object which has reference to another object, the reference will be removed, but that other object will not be removed and will remain in the cache. Likewise, if you delete an object, references to that object will not be deleted, they will simply fail, causing a cache miss, when you attempt to load the object again.

This means that if you are planning to remove something, be sure that you either a) Know for sure you no longer need it, or b) Are fine with your cache policy potentially triggering an additional fetch if the missing value causes a read to fail.

Note: As of right now, there is not a way to delete a single property's value. For instance, calling try transaction.removeRecord(for: "") will result in no action, as there would not be a record with the cache key "", since name is a scalar on the "2001" record.

Using fragments
Edit on GitHubEditForumsDiscord

© 2024 Apollo Graph Inc.

Privacy Policy