The calculation methods of quorum and superquorum are as follows:
/// Xline/curp/src/client/unary.rs
/// Calculate the super quorum
fn super_quorum(size: usize) -> usize {
let fault_tolerance = size.wrapping_div(2);
fault_tolerance
.wrapping_add(fault_tolerance.wrapping_add(1).wrapping_div(2))
.wrapping_add(1)
}
/// Calculate the quorum
fn quorum(size: usize) -> usize {
size.wrapping_div(2).wrapping_add(1)
}
Actually， for any distributed cluster with a node count of N, the sum of its fault tolerance and quorum should always be equal to N. But the current calculation method for fault tolerance breaks this assumption. This assumption holds true when and only when the N is odd.
For any even number N (N = 2K), the calculation using the above method yields:
fault tolerance = N / 2 = K
quorum = N/2 + 1 = K + 1
Obviously, N = fault tolerance + quorum = 2K + 1 contradicts N being even. The super quorum, derived from the incorrect fault tolerance, also cannot be correct.
The correct calculation methods are as follow:
quorum = N/2 + 1
fault quorum = N - quorum
recovery quorum = quorum / 2 + 1
super quorum = fault tolerance + recovery quorum
Version
0.6.0 (Default)
Relevant log output
No response
Code of Conduct
[X] I agree to follow this project's Code of Conduct
Description about the bug
The calculation methods of
quorumandsuperquorumare as follows:Actually， for any distributed cluster with a node count of N, the sum of its fault tolerance and quorum should always be equal to N. But the current calculation method for fault tolerance breaks this assumption. This assumption holds true when and only when the N is odd.
For any even number N (N = 2K), the calculation using the above method yields:
Obviously, N = fault tolerance + quorum = 2K + 1 contradicts N being even. The super quorum, derived from the incorrect fault tolerance, also cannot be correct.
The correct calculation methods are as follow:
Version
0.6.0 (Default)
Relevant log output
No response
Code of Conduct