Replication and failover strategy for servers

[shumkov]

[till]

I wonder if this shouldn't be handled in code but by a TCP loadbalancer like haproxy?

[shumkov]

No, its 'readOnly', 'writeOnly' options for servers

[marcguyer]

What happens currently if 2+ servers are configured with consistent distribution hashing (ketama) but one of the servers becomes unavailable?

[marcguyer]

According to my research and tests, there is zero special handling for server failure. The consistent key hashing will distribute cache data across a pool of servers. However, if a member of the pool is unavailable, and the key you're operating on hashes to that server, you have a fatal situation for that particular key.