Update serf to pick up graceful leave fix #4278
Conversation
| @@ -1245,6 +1245,10 @@ func (s *Server) setupSerf(conf *serf.Config, ch chan serf.Event, path string) ( | |||
| } | |||
| conf.ProtocolVersion = protocolVersionMap[s.config.ProtocolVersion] | |||
| conf.RejoinAfterLeave = true | |||
| // LeavePropagateDelay is used to make sure broadcasted leave intents propagate | |||
| // This value was tuned using https://www.serf.io/docs/internals/simulator.html to | |||
| // allow for convergence in 99.9% of nodes in a 10 node cluster | |||
There was a problem hiding this comment.
I'm confused by this comment -- how can there be 99.9% of nodes in a 10 node cluster? (either 90% or 100%?)
There was a problem hiding this comment.
That serf simulator is based on a convergence factor calculated with logarithmic scale that never reaches 100%, so I didn't round it up for the comment. In practice, for such a small cluster (10 nodes), this means all nodes will converge with this value.
In a 100K node cluster, 99.9% is 99900 nodes, so the fact that its 99.9% and not 100% matters.
|
I'm going to lock this pull request because it has been closed for 120 days ⏳. This helps our maintainers find and focus on the active contributions. |
Update serf library for improvements to ensure that leave intents are handled correctly