Restrictions

• This feature will only be available when asynchronous cross-site (aka IRAC) is enabled.
• Only for distributed caches
• Automatic state transfer needs to be enabled

Problem

With distributed caches, if num-owners nodes crash at the same time, the segments/data owned by those nodes will be lost.

Improvement

The site that lost the data can send a state transfer request to the backup site and fetch the data from there.

Corner cases

• If the backup site is offline, no request can be made. We rely on the current automatic state transfer feature to sync the data.
• If more than one backup is configured, it will send the request just to a single site. If it times out, it requests from the next site.
• Stale data will happen from the time the segments are lost until the state is received

Optimization

If the HashFunction and the num-segments are the same, the state transfer can be selective reducing the traffic. This may or not be implemented in a follow up JIRA.