SQL cache is unusably slow

[edenman]

Version

4.0.0-beta.4

Summary

• New project, interacting with a old and very gnarly graphql schema.
• Response is ~200kb
• We're observing the cache with a CacheOnly policy and watch
• When we make a network call, we're using NetworkOnly and then manually manipulating the response before writing it to the cache
• When we write a response to the cache, it takes 14 seconds for the watch to fire. If I remove the SqlNormalizedCacheFactory that drops down to 3 seconds.

Steps to reproduce the behavior

        ApolloClient.Builder().apply {
            interceptors(apolloInterceptors.toList())
            networkTransport(
                HttpNetworkTransport.Builder()
                    .httpEngine(DefaultHttpEngine(okHttpClient))
                    .httpRequestComposer(RequestComposer)
                    .build(),
            )
            normalizedCache(
                MemoryCacheFactory(10 * 1024 * 1024)
                    .chain(SqlNormalizedCacheFactory("$userId-apollo.db"))
            )
            writeToCacheAsynchronously(true)
            autoPersistedQueries()
        }.build()

Logs

No response

[BoD]

Hi! Thanks for reporting.

Can you tell us a bit more about the response manipulation you're doing, and how it is done? Are you using doNotStore so that the data is not written twice to the cache? Is the writing call itself taking a while, or is it only the watch being slow to be triggered (or both).

Also in general if you could share your (or part of your) schema, query and json payload, that would help a lot to have a look and try to understand what is happening (share privately to benoit.lubek@apollographql.com if that's preferable).

[edenman]

• Correct, we're doing doNotStore(true) on the network query.
• The response is a list of heterogeneous items
• The response includes some special caching instructions in the body, so we're manually processing those and then merging the existing items with the new ones, and then inserting a new FooResponse into the cache
• I'll send you the json payload via email

[BoD]

Thanks a lot for having shared all the details. I could reproduce writing to the cache taking an unusually long time, due to a transaction per insert being used, instead of just one transaction for all of them, when the memory cache is used with chaining. This is fixed here and will be in the next v4 release (planned in the next few days).

[BoD]

Released in 3.8.4

[github-actions]

Do you have any feedback for the maintainers? Please tell us by taking a one-minute survey. Your responses will help us understand Apollo Kotlin usage and allow us to serve you better.