This lists the known issues with clustering (running multiple instances in a load-balancer setup, or even just running multiple queue processors). More are to be investigated.

• The queue system's job retry handler (RecoverOrPrepareForExitAsync) doesn't handle crashes properly (especially with multiple queue processor workers)
• WebSocket connections will only receive events processed by the node they’re connected to
• So far, there is no option to start a node in pure "queue processor" mode
• Cron tasks shouldn't be executed by all workers & should be executed in a way that's resilient to a worker going down
• Push notifications get duplicated once per full/web worker (queue workers don't affect this problem) (maybe push notification delivery should be a background-task job in cluster mode?)
• Duplicate work prevention (AsyncKeyedLocker) is not easily adaptable to multiple workers