I've been trying to consume a subscription in Kotlin Multiplatform and finding that I often have man...

# apollo-kotlin

[Neal Sanche]

I've been trying to consume a subscription in Kotlin Multiplatform and finding that I often have many missing elements in my flow. The code I'm using is in the thread.

[Neal Sanche]

Graph is just a poor-man's dependency graph, where we set up the GraphQL subscription client.

Untitled

[Neal Sanche]

I've set up code like this for the creation of the ApolloClient for the subscription handling. This connects, and does get one or two elements... But if the elements come in too fast, it seems I don't get them all. When the server delays sending, all of the elements do come in.

Untitled

[Neal Sanche]

What am I missing? Am I doing it wrong?

[mbonnin]

That's interesting. Sounds like a CONFLATED channel somewhere... Do you have a rough idea of the delay that makes everything work again? Are we talking milliseconds or seconds?

[Neal Sanche]

More than 100ms and works pretty reliably around 1s. I have a little time to play with this today, but then I have to move on.

[mbonnin]

Do you mind opening an issue with your code in the repo (pseudo code work too)? Make sure we can track this?

[Neal Sanche]

I don't mind at all. Since the GraphQL playground seems to be able to get all of the data, I am going to focus on the KMP side and see where I've gone wrong. The server side is using graphql-kotlin, and the code I used for that is really small.

[mbonnin]

Also quick question, is this JVM, native or both?

[Neal Sanche]

The app will be iOS and Android, so both.

[Neal Sanche]

I have only tested on Android so far.

[Neal Sanche]

Okay, wrote: https://github.com/apollographql/apollo-android/issues/3095 Let me know if you want more information inside that report, @mbonnin

[mbonnin]

Thanks, will look into it! graphql-ws is getting some attention lately, I know @Doug Chappelle has been looking into it too a couple of threads just above this one so looks like there might be something to do there

[mbonnin]

Alright, that'd be the CONFLATED I guess: https://github.com/apollographql/apollo-android/blob/818ff720fe982132e9e0d5bccd20e[…]llographql/apollo/network/ws/ApolloWebSocketNetworkTransport.kt

[Neal Sanche]

Funny I was just looking at exactly that line.

[mbonnin]

🙂

[mbonnin]

Not sure there is a reason for this, I guess we should make it configurable somehow, with the defaultcertainly to not drop items

[Neal Sanche]

I'll see if I can get set up to experiment with the different options on that webSocketConnection.broadcast() function...

[mbonnin]

These days, SharedFlow might be an option too, I think it didn't exist at the time

[Neal Sanche]

That's true.

[mbonnin]

Oh well, there's more to it as the producer gets called from a okhttp non-suspending callback: https://github.com/apollographql/apollo-android/blob/23e78fc09452e17ffe8d5da9a03f7[…]n/kotlin/com/apollographql/apollo/network/ws/ApolloWebSocket.kt

[mbonnin]

Buffering will work to a certain extent but proper backpressure handling will require more work

[Neal Sanche]

I've cloned the repo, but not sure the easiest way to make small code changes, publish to my local maven repo and try it in my code?

[mbonnin]

Yep, publishToMavenLocal works well

[mbonnin]

You can also include the build but tbh that might take longer in gradle sync compared to maven local

[Neal Sanche]

Okay, publishToMavenLocal works. Just doing that and testing a few things out here.

[Neal Sanche]

So, making ApolloWebSocketNetworkTransport.kt:180 read:

private val messageChannel: BroadcastChannel<ByteString> = webSocketConnection.broadcast(Channel.BUFFERED)

works the way I'd expect it to. I get all of the data items delivered.

[mbonnin]

Nice! Want to submit a PR to make this a parameter of ApolloWebSocketNetworkTransport? We can figure out the backpressure stuff later

[Neal Sanche]

Sure, will give that a shot.

[Neal Sanche]

Okay, I've made a PR: https://github.com/apollographql/apollo-android/pull/3096 and do let me know in the PR if you'd prefer a different implementation. I simply went for minimal changes, and defaults to the current value.

[mbonnin]

Looks good to me, thank you so much for the contribution 😊 !

[Neal Sanche]

You're welcome. Looking forward to using it to finish this feature I'm working on. I'll keep my eye out for when it gets into the repo. Thanks for your help with it!