Hey guys. I have a strange problem that I need an opinion on. We have an active subscription and a m...

# apollo-kotlin

[Vitaliy Kondratiev]

Hey guys. I have a strange problem that I need an opinion on. We have an active subscription and a mutation that returns the same data. Problem is that the data over the subscription comes back faster than the data over the mutation and is in a more recent state. But we cache the last data that we get from the mutation resulting in having a wrong state. I'm wondering if there is a way around this other than just simply ignoring all data from the mutation. Thanks in advance

[mbonnin]

Hi 👋 Good question! Ignoring the data from the mutation sounds like the way to go

[mbonnin]

You can use ApolloCacheHeaders.DO_NOT_STORE to avoid storing anything in the cache

[Vitaliy Kondratiev]

I saw that. How would I go about specifying it for mutation but not subscriptions?

[mbonnin]

let me check

[mbonnin]

Something like this should do it:

apolloClient
                        .mutate(Mutation(email = Input.fromNullable(email)))
                        .toBuilder()
                        .cacheHeaders(CacheHeaders.builder()
                            .addHeader(ApolloCacheHeaders.DO_NOT_STORE, "true")
                            .build()
                        )
                        .build()
                        .await()

[Vitaliy Kondratiev]

Oh great. I'll give it a try

[mbonnin]

I think you can put whatever instead of "true" and it'll work 🙈

[mbonnin]

Let us know how it goes

[Vitaliy Kondratiev]

So the solution would work but there are still scenarios where the websocker response comes back slower than the mutation. Resulting in longer wait times. Our backend provides a field 'generatedAt' which is a 'Long' representing a timestamp in microseconds. Is there a good way to utilize this field so that I can simply ignore a fragment if it's 'generatedAt' field is less than what is is the cache

[mbonnin]

You might be able to do something with an ApolloInterceptor

[mbonnin]

Mmmm no sorry that won't work, all the application interceptors are registered before the cache ones so I don't think you'll be able to bypass the cache there...

[mbonnin]

https://github.com/apollographql/apollo-android/blob/621c1a192e9434a8981d5ce6825e5c05df79e454/apollo-runtime/src/main/java/com/apollographql/apollo/internal/RealApolloCall.java#L391

[mbonnin]

one thing you could try is using a NO_CACHE policy and saving to the cache manually using apolloClient.apolloStore.write(operation, data) , this way you have full control of what ends up in the cache

[Vitaliy Kondratiev]

Yeah I thought about that too. It's a bit messy in my opinion. I was looking into using the Cache itself. But not sure if it makes sense. It would look something like this if the LruNormalizedCache was an open class

class CustomLruNormalizedCache constructor(evictionPolicy: EvictionPolicy) :
    LruNormalizedCache(evictionPolicy) {

    companion object {
        private const val GENERATED_AT_FIELD = "generatedAt"
    }

    @SuppressWarnings("NestedBlockDepth")
    override fun merge(recordSet: Collection<Record>, cacheHeaders: CacheHeaders): Set<String> {

        recordSet.recordWithGeneratedAt()?.let { record ->
            record.generatedAt()?.let { newGeneratedAt ->

                loadRecord(record.key, CacheHeaders.NONE).generatedAt()?.let { lastGeneratedAt ->

                    Timber.d("Last: $lastGeneratedAt | New: $newGeneratedAt")
                    if (newGeneratedAt < lastGeneratedAt) {
                        return emptySet()
                    }
                }
            }
        }

        return super.merge(recordSet, cacheHeaders)
    }

    private fun Collection<Record>.recordWithGeneratedAt(): Record? = firstOrNull { it.hasField(GENERATED_AT_FIELD) }

    private fun Record?.generatedAt(): Long? = try {
        this?.fields?.get(GENERATED_AT_FIELD).asType<String>()?.toLong()
    } catch (e: Throwable) {
        null
    }
}

[mbonnin]

Yep, that could work

[mbonnin]

Just copy/paste the whole class, it's not that big (<100 lines)

[Vitaliy Kondratiev]

LruNormalizedCache ?

[mbonnin]

Yep

[mbonnin]

I don't see any reason why you shouldn't be able to pass your own CustomLruNormalizedCacheFactory

[Vitaliy Kondratiev]

requires dependency on the nyTimes cache. And I would not want my code to diverge by copy pasting something. I guess I could also write my own

[mbonnin]

Yep, writing a lock free cache like the guava one (which nyTimes is using under the hood) is not an easy thing though so I'd recommend to reuse that at least

[mbonnin]

Or if you ever end up writing your own Kotlin lock-free cache, please contribute it back, we'll need something like this for proper multiplatform support 🙂

[Vitaliy Kondratiev]

I see, did not realize it would not be as simple as I thought