What you need is to switch the coroutine context in your repository so that it does not get cancelled when viewmodel is destroyed. Your repository should have a scope of its own and if you do some sync operations that should not be cancelled on vm destroy then sync operations should run on repository scope instead of vm scope. These are good articles to read about scopes and their context: https://medium.com/@elizarov/the-reason-to-avoid-globalscope-835337445abc https://medium.com/@elizarov/coroutine-context-and-scope-c8b255d59055

Im confused, youre saying for my sync I should have something like

Repository {
    fun sync() {
       repoScope.launch { .. }
    }
}

One way is to call

join()

on the

Job

returned by

launch {...}

. This means, though, that

sync()

must become

suspend

.

right? but then sync is a normal function, how can I let vm know when sync is done, etc?

Usually these kind of operations are done using channels to retrieve the result. After the sync operation is done it should send the updated data through the channel. This way if you have multiple calls to sync you will always have the most recent data in your UI since you have a single source of truth. If you really need the result then you should use

withContext()

and pass in the repo scope context. This ways the block will not be cancelled by vm destroy.

but if I sideffect to channels, then I lose composability right?