This is a harness for deterministically simulating the execution of distributed algorithms.
See DistributedPlayground.Algorithms/PingPongSimple for a simple example algorithm
and DistributedPlayground.Algorithms.Tests/PingPongSimple for some simple tests.
The tests can instantiate processes, connect them together via a network and inspect the result.
The network can simulate
- the passage of time
- network latency
- network partitions
- network regions (coming soon)
- lost messages
- duplicated messages
- out of order messages
This is achieved via a configurable pipeline of interceptors for the scheduling and then delivery
of messages sent in the network. See Interceptors static class.
Processes derive from LogicalProcess which provides them with some helpers for easily sending messages
to themselves and others. When handling messages they are dispatched internally using dynamic for
brevity but I may reconsider this.
- Other algorithms: Paxos State Machine Replication, VR, Raft, MESI, ...
- some way to avoid dynamic dispatch without having to write a lot of subscribing code
- perhaps a random delay that doesn't reorder messages between two processes
- some process to randomly partition the network