questions about HA and key distribution

[siddo420]

For HA, I am thinking of setting up multiple nutcracker servers (that may be used together in parallel). All nutcracker servers may be connected to the same redis+memcached servers on the backend. My questions are:

• Is this supported?
• In the scenario above, if I plan on using data partitioning using hash_tag:
  • Will partitioning work using any distribution mode or hash function or I have a limited set to choose from?
  • If one or more nutcracker servers die (for any reason), will the remaining nutcracker server(s) correctly route clients to the right backend servers based on the hash_tag values? In other words, the hash function and distribution modes are enough to tell nutcracker which server to use or it stores some data locally about key distribution etc?

Please let me know

Thanks.

[siddo420]

Also, if we want to add new redis servers to an existing pool then what happens to key distribution? Will nutcracker be querying the older servers for old data or adding new servers in an existing pool will mess up the key distribution algorithm?

[vlm]

The key distribution will be skewed, so some of the keys which used to connect to the older servers will be dealt with by the newer server. Moreover, depending on distribution, there can be some keys which will move from older servers to another set of the older servers.