Phil,
Could you please file a ticket? We can hopefully get this done in more detail, but in gist:
Eureka:
- AP (weak consistency), state is replicated with “best effort”
- Services are registered with one server, which attempts to replicate to other servers
- Service registrations have a TTL, and clients must heartbeat
- Reads are routed to any server, can be stale or missing data
- Scales well due to low coordination, especially when server failures relatively rare
- Fails if all servers down
Consul
- CP (strong consistency), state is replicated using Raft
- Services registered with any server, but written via Raft to a quorum
- Registrations have complex health checks, including gossip failure detection instead of heartbeating
- Reads routed to any server, consistent by default but stale reads can be requested
- Stale reads scale well, Consistent reads scale to tens of thousands per second
- Consistency offers locking and cluster coordination
- Lots more features (health checking, locking, KV, federation, ACLs)
- Fails if a majority of servers down
Hope that helps!
Best Regards,
Armon Dadgar