Till Rohrmann

A slightly different variant could be that the current leader only proposes the current partition time by periodically writing the time to Bifrost. When applying the partition time message all...

> I investigated this, it needs a bit of internal refactoring of invocations in scheduled state. We need to introduce `InvocationStatus::Scheduled`, which comes before Inboxed as state transition. Thanks for...

> * On release 1.2, we add a migration script that automatically runs on boot to completely migrate all the invocations to the new table. The migration step is probably...

Could be caused by https://github.com/tokio-rs/tokio/issues/4730?

> Hi there, any updates on this? This would be game changing. This feature is still very high on the list of new features that we want to add to...

cc @jackkleeman

The same problem applies to sending responses back to the ingress. If the current leader successfully proposes an `InvokerEffectKind::End` and then loses leadership, then the next leader will think that...

To sum up the underlying problem again: Restate PPs need to send messages to the ingress (responses and `SubmitNotifications`) so that the ingress can respond to external clients. This communication...

Personally, I am leaning towards making the communication between the ingress and PP reliable since it feels a bit odd to me that Restate can "lose" some messages internally which...

Ahmed proposed a new variant: The ingress partition processor communicate via RPCs. Instead of directly writing to Bifrost, the ingress will rpc the leading partition processor (routing based on idempotency...