Struct InMemoryGroup 

pub struct InMemoryGroup { /* private fields */ }

An n-node fully-connected in-memory replication mesh.

InMemoryGroup owns one InMemoryEndpoint per ordered (from, to) peer pair (with from != to). The endpoint at (from, to) is from’s view of its socket to to; sending on that endpoint is observed by the endpoint at (to, from).

Higher layers typically consume the group by handing each node the row of channels [group.channel(my_id, peer)] for peer in 0..n.

Crash and recovery

InMemoryGroup::simulate_crash closes every channel touching a node, modelling a hard crash or partition. After a crash, InMemoryGroup::reconnect rewires that node’s row of channels (paired against the other live nodes) to model a node restart or healed partition. A crashed node may be reconnected at most once per crash; the implementation tolerates repeated calls.

Implementations

impl InMemoryGroup

pub fn size(&self) -> usize

Number of nodes in the mesh.

pub fn channel(&self, from: usize, to: usize) -> Arc<dyn Channel>

Return a cheap Arc<dyn Channel> handle to the directed channel from from to to.

Panics

Panics if from or to is out of range, if from == to, or if the channel has been removed by Self::simulate_crash without a subsequent Self::reconnect.

pub fn try_channel(&self, from: usize, to: usize) -> Option<Arc<dyn Channel>>

Try to acquire the directed channel from from to to, returning None if the channel has been crashed.

Panics

Panics on out-of-range indices or from == to.

pub fn simulate_crash(&self, node: usize)

Simulate a hard crash of node: every channel originating from or terminating at node is closed and dropped from the mesh. Subsequent send / receive on any handle that was previously cloned out of those channels returns crate::error::RepError::ChannelClosed, matching a real socket disconnect.

Idempotent: calling simulate_crash on an already-crashed node is a no-op.

Panics

Panics if node is out of range.

pub fn reconnect(&self, node: usize)

Rewire node’s row of channels against every peer that is still live (i.e., still has a channel slot to node). Models a node restart or a healed partition.

Channels whose remote end is itself currently crashed are left disconnected; the caller should reconnect them in a separate pass once the remote node has come back.

Panics

Panics if node is out of range.

pub fn is_node_live(&self, node: usize) -> bool

Return true iff every directed channel touching node is currently open.