When nodes are started they detect each other and form a directed ring graph. Each node generate messages and remember them in its cache and send to its right neighbour(next node on ring). When node receives message from left node (previous node on ring) it checks such message exist in its cache, if not then retranslate(pass) further to next node, otherwise do not pass it further. When some additional node starts all nodes recompute its right neighbour and if it is changed then all cached messages passed to the right neighbour (next node).
Node have some specific internal processes residing in its dedicated module:
- Peer Monitor -- discovers nodes and computes right neighbour (next node) and sends it to Message Retranslator process (see bellow)
- Message Generator -- generates random messages and passes them to Message Retranslator process (see bellow)
- Message Retranslator -- retranslates(passes) messages to next node if they are not in cache. Sends newly generated messages (received from Message Generator) and caches them. When new next neighbour received from Peer Monitor it sends all cached messages to it.
There is a problem with findPeers in a PeerMonitor, it does not discover node down, hence come additional ping-pong protocol/mechanism required to accurately detect nodes down event.
Currently nodes started manually from shell:
$ stack exec chtt -- --host "127.0.0.1" --port "1234" --send-for 30 --wait-for 5 --with-seed 0
$ stack exec chtt -- --host "127.0.0.1" --port "1235" --send-for 30 --wait-for 5 --with-seed 0
$ stack exec chtt -- --host "127.0.0.1" --port "1236" --send-for 30 --wait-for 5 --with-seed 0
$ stack exec chtt -- --host "127.0.0.1" --port "1237" --send-for 30 --wait-for 5 --with-seed 0
It is required to clarify 2.3 Cluster configuration how to launch
cluster in a more automatic fashion.