Triton v0.2.2 Triton.Supervisor View Source
There is one supervisor per cluster, which is composed of a Xandra worker + Triton monitor. The Supervisor uses a one_for_all strategy to ensure that clusters are restarted when the monitor fails.
This design is necessary because when Xandra / DBConnection receives a disconnection error it cannot be restarted without terminating & restarting Xandra.
Separation of clusters ensure that one cluster going down doesn’t affect the other.