About Nonstop Operation
The nonstop feature allows certain operations to proceed on Terracotta clients that have become disconnected from the cluster, or if an operation cannot complete by the nonstop timeout value. This is useful in meeting service-level agreement (SLA) requirements, responding to node failures, and building a more robust High Availability cluster.
One way BigMemory Max can go into nonstop mode is when a client receives a "cluster offline" event. Note that a nonstop instance can go into nonstop mode even if the client is not disconnected, such as when an operation is unable to complete within the timeout allotted by the nonstop configuration. In addition, nonstop instances running in a client that is unable to locate the TSA at startup will initiate nonstop behavior as if the client had disconnected.
Nonstop can be used in conjunction with rejoin. For information about the rejoin feature, see "Configuring Reconnection and Rejoin Properties" in the book Configuring a Terracotta Cluster for High Availability.
Use cases include:
Setting timeouts on operations.
For example, say you use BigMemory Max rather than a mainframe. The SLA calls for 3 seconds. There is a temporary network interruption that delays the response to a cache request. With the timeout you can return after 3 seconds. The lookup is then done against the mainframe. This could also be useful for write-through, writes to disk, or synchronous writes.
Automatically responding to cluster topology events to take a pre-configured action.
Allowing Availability over Consistency within the CAP theorem when a network partition occurs.
Providing graceful degradation to user applications when distributed BigMemory Max becomes unavailable.