Not seeing expected improvement in throughput of RaftCluster.ReplicateAsync method when cluster minority is inaccessible

[LarsWithCA]

Hello there!

This plots shows the timing (in ms) of RaftCluster.ReplicateAsync - at the vertical green line 1 node is disconnected (out of a cluster of 6 nodes in total):
(Linux ARM + .NET6 + DotNext.Net.Cluster 4.14.1)

From the change log of DotNext.Net.Cluster 4.15.0:

Raft performance: improved throughput of IRaftCluster.ReplicateAsync method when cluster minority is not accessible (faulty node). Now the leader waits for replication from majority of nodes instead of all nodes

This made us hope that we would no longer see these kinds of longer timings in case of inaccessible cluster minority. However, we see a pretty similar plot - at the green line 1 node is disconnected (out of a cluster of 6 nodes):
(Linux ARM + .NET8 + DotNext.Net.Cluster 5.3.0)

Did we have wrong expectations, or are we doing something wrong?

[sakno]

I'll check this out.

[sakno]

For now, appended log entries are marked as committed when replicated to the majority. ForceReplicationAsync remains suspended while receiving responses from all (even unavailable) members. So information about increased throughput is partially true. Will take a look how resume callers earlier.

[sakno]

@LarsWithCA , could you please create an issue?

[LarsWithCA]

Hi @sakno I created #233

[LarsWithCA]

Fixed in version 5.5.0