Memory leak during DDP cursor cleanup on low-tier instances

When hosting low-traffic Meteor applications on minimal-memory cloud instances, the publish-subscribe mechanism often leads to unexpected memory exhaustion. The Distributed Data Protocol (DDP) maintains persistent connections to facilitate real-time synchronization, which creates a server-side state footprint for every active subscription.

Specifically, when clients frequently disconnect and reconnect (flapping), the de-hydration and rehydration of the OMongo cache state can cause CPU spikes. If the 'unpublish' method is not explicitly invoked or if the framework fails to clean up cursors promptly, memory usage grows, eventually exceeding the available resources on small instances.

Are there specific configuration parameters to ensure that DDP cursors are purged immediately upon client heartbeat timeout, and how can the server-side state footprint be minimized for non-critical data without abandoning the reactive-modeling paradigm?