Closed jeremydmiller closed 6 months ago
One active node per queue could be a start, but wouldn’t wolverine need to route to a certain node based on the actor id? (either via a partitioning scheme, or via a persisted registry of which key is running where?)
@lcorneliussen Cruder than that. If wolverine is publishing to an endpoint where only the one sticky listener is handling it, you're de facto a poor man's actor.
Oh, and Wolverine does have the ability to let you define your own distribution of "sticky" agents across a cluster of nodes. Not documented right now, but used under the covers for some of the inbox/outbox stuff and the "Critter Stack Pro" commercial tooling that distributes asynchronous projections over a cluster
ListenerConfiguration.StrongOrderedGuarantees()
, sets all the necessary defaultsListenerAgentFamily
implementation
What if Wolverine had the ability to make the message listener for an endpoint (a Rabbit queue, an Azure Service Bus queue, a Kafka topic) only active on a single node within a Wolverine cluster? This potentially gives Wolverine the ability for: