1.5. Shutting down Data Grid Server

Prevent data loss and ensure consistency of your cluster by properly shutting down and restarting nodes.

Cluster shutdown

Data Grid recommends using the shutdown cluster command to stop all nodes in a cluster while saving cluster state and persisting all data in the cache. You can use the shutdown cluster command also for clusters with a single node.

When you bring Data Grid clusters back online, all nodes and caches in the cluster will be unavailable until all nodes rejoin. To prevent inconsistencies or data loss, Data Grid restricts access to the data stored in the cluster and modifications of the cluster state until the cluster is fully operational again. Additionally, Data Grid disables cluster rebalancing and prevents local cache stores purging on startup.

During the cluster recovery process, the coordinator node logs messages for each new node joining, indicating which nodes are available and which are still missing. Other nodes in the Data Grid cluster have the view from the time they join. You can monitor availability of caches using the Data Grid Console or REST API.

However, in cases where waiting for all nodes is not necessary nor desired, it is possible to set a cache available with the current topology. This approach is possible through the CLI, see below, or the REST API.

Server shutdown

After using the shutdown server command to bring nodes down, the first node to come back online will be available immediately without waiting for other members. The remaining nodes join the cluster immediately, triggering state transfer but loading the local persistence first, which might lead to stale entries. Local cache stores configured to purge on startup will be emptied when the server starts. Local cache stores marked as purge=false will be available after a server restarts but might contain stale entries.

If you shutdown clustered nodes with the shutdown server command, you must restart each server in reverse order to avoid potential issues related to data loss and stale entries in the cache.
For example, if you shutdown server1 and then shutdown server2, you should first start server2 and then start server1. However, restarting clustered nodes in reverse order does not completely prevent data loss and stale entries.