Cache size and eviction

[Poongavanama]

Can you please clarifiy the below points.

1.Cache eviction vs cache expiry which takes precedence.
2.Size is it entries vs MB.
3. How does cache weighted eviction works?
4. Eviction is invalidation() or delete() process

Thank you.

[ben-manes]

1. The cache will remove expired entries before it performs size eviction.
2. This can be adjusted by using a weigher. The default is that every entry takes 1 unit of capacity (entry count). The weigher could instead calculate the entry's byte size as the number of units, e.g. by using JAMM or JOL. The byte size is discussed in JEP draft: Low-level Object layout introspection methods with links to Solr and Cassandra examples which use a Caffeine weigher to achieve this.
3. The weight is not used to make eviction decisions and is simply a threshold value. There was a research paper which goes into more detail, but that work hasn't been incorporated yet.
4. We define the terms as below and when notifying a listener we send along the RemovalCause.
   1. eviction means removal due to the policy
   2. invalidation means manual removal by the caller
   3. removal occurs as a consequence of invalidation or eviction

[Poongavanama]

Thank you @ben-manes

[Poongavanama]

Whcih one to recommented to use in configuration bean expireAfterWrite or expireAfterAccess, We have kubernetis nodes and multiple request invoke the API. Could you please suggesst which one is recommented to use.
Currently I am suing expireAfterAccess.

[ben-manes]

I don’t know, they are different policies, time-to-idle or time-to-live. I think TTL (expireAfterWrite) is more common, but it depends on what behavior you want.

[Poongavanama]

Thank you very much @ben-manes.

[Poongavanama]

I don’t know, they are different policies, time-to-idle or time-to-live. I think TTL (expireAfterWrite) is more common, but it depends on what behavior you want.