Consider this suggested deployment:
Cross reference with high availability docs:
3
or5
freno
nodes form afreno
highly available service- One of them is elected leader, and it is the only node to:
- Continuously probe backend stores
- Serve
/check
queries
HAProxy
service directs all traffic to the single active node- Clients consult with
freno
viaHAProxy
. They will implicitly connect to the leader node. - Based on
freno
's response they will either write or refrain from writing to backend stores.