Cache eviction strategies of Liferay Clusters

Issue

1. When a cache eviction order is emitted on a node, is an eviction message also sent to the other nodes?
2. Is replication the default behavior in a clustered environment?
3. When the cache expiration time is reached on a node, is the eviction message also sent to the other nodes?

Environment

• Liferay DXP 7.0+

Resolution

1. Liferay doesn't replicate evictions across to other nodes, as they may still be using the cached asset and need it to handle requests/responses.
2. No, but we send the update messages to the other nodes so they can mark cached entities locally as dirty/stale, so the other nodes should get a cache miss and perform a subsequent re-retrieve of the data.
3. Cache expiration messages don't replicate across all nodes. Just because one node isn’t using the cached data to fulfill requests and ends up expiring a cached asset, that doesn’t mean other nodes are not actively using those cached entities to fulfill their own requests.

Additional Information

• Caching is really to support the activity on a single node, it is not a cluster-wide solution.
• Each node will maintain its own cache and populate it with retrieved data that it needs to service its requests. Most of the time the cluster nodes will often have a highly similar cache because they are all serving similar requests, but especially with sticky node support, a single node may be retrieving data for an admin user that would never be retrieved/used on other nodes.
• The only activity the nodes need to message is the updated data so that all caches can mark the entity as stale and not then try to use stale/dirty data in the process of serving subsequent requests.