java - Cache inconsistency - Entity not always persisted in cached Collection -

i'm having issue validation instance added collection on step instance. declaration follows:

step class:

@entity @table @cacheable @cache(usage = cacheconcurrencystrategy.read_write) public class step extends abstractentity implements validatablestep {      @onetomany(fetch = fetchtype.lazy, cascade = cascadetype.all, orphanremoval = true)     @joincolumn(name = "step_id", nullable = false)     @cache(usage = cacheconcurrencystrategy.read_write)     private set<validation> validations = new hashset<>();      @override     public void addvalidation(validation validation) {       // stuff       ...       // add validation instance collection       getvalidations().add(validation);     }  } 

validation class:

@entity @table @cacheable @cache(usage = cacheconcurrencystrategy.read_write) @noargsconstructor(access = accesslevel.protected) public class validation extends abstractentity {   //some properties } 

both classes cacheable read_write strategy applied. unidirectional collection of validations cached same strategy.

one expect when read-write transaction invokes addvalidation(new validation('username')); commits, new validation visible in subsequent read-only transaction. weird thing work , doesn't work...

the first transaction succeeds; see new validation being persisted in database , step's version property (for optimistic locking puposes) getting incremented. sometimes, 2nd read transaction contains step instance empty validation collection...

our hibernate caching config follows:

hibernate.cache.use_second_level_cache = true hibernate.cache.use_query_cache = true hibernate.cache.region.factory_class = org.hibernate.cache.ehcache.singletonehcacheregionfactory hibernate.cache.provider_configuration_file_resource_path = classpath:ehcache.xml net.sf.ehcache.hibernate.cache_lock_timeout = 10000 

any idea what's causing weird (and random) behavior?

the hibernate collection cache invalidates existing entries , both entity , collection caches sharing same abstractreadwriteehcacheaccessstrategy, soft-lock acquired when updating data.

because using unidirectional one-to-many association, end validation table , step_validation link table too. whenever add/remove validation have hit 2 tables , that's less efficient.

i suggest adding @manytoone side in validation entity , turn @onetomany side mapped-by collection:

@onetomany(fetch = fetchtype.lazy, cascade = cascadetype.all, orphanremoval = true, mappedby = "step") @cache(usage = cacheconcurrencystrategy.read_write) private set<validation> validations = new hashset<>();