Issue Summary

Performance improvement in request access operation on a restricted page

Steps to Reproduce

1. When a user click on Request access on a restricted page, Confluence runs the following query for each group that has been assgined space admin permissions

   2020-11-25 19:59:01,775 DEBUG [http-nio-8094-exec-8] [org.hibernate.SQL] logStatement select internalus1_.user_name as col_0_0_ from cwd_membership hibernatem0_ cross join cwd_user internalus1_ cross join cwd_group internalgr2_ where hibernatem0_.child_user_id=internalus1_.id and hibernatem0_.parent_id=internalgr2_.id and (hibernatem0_.child_group_id is null) and internalgr2_.lower_group_name=? and internalgr2_.directory_id=? and internalgr2_.group_type=? order by internalus1_.lower_user_name
   

2. If there are multiple very large groups added, it can result in a higher CPU utilization on the instance and DB
3. Looking at the thread dump, the thread is busy with flushing and loading the user caches

   http-nio-8080-exec-96 url:/confluence/res...48942/request/view username:HOE1IMB" #446 daemon prio=5 tid=0x00007f23480b2800 nid=0x72b2 runnable [0x00007f1d9b075000]
      java.lang.Thread.State: RUNNABLE
       at sun.reflect.GeneratedMethodAccessor1160.invoke(Unknown Source)
       at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43)
       at java.lang.reflect.Method.invoke(Method.java:498)
       at org.hibernate.property.access.spi.GetterMethodImpl.get(GetterMethodImpl.java:41)
       at org.hibernate.tuple.entity.AbstractEntityTuplizer.getPropertyValues(AbstractEntityTuplizer.java:604)
       at org.hibernate.tuple.entity.PojoEntityTuplizer.getPropertyValues(PojoEntityTuplizer.java:215)
       at org.hibernate.persister.entity.AbstractEntityPersister.getPropertyValues(AbstractEntityPersister.java:4707)
       at org.hibernate.event.internal.DefaultFlushEntityEventListener.getValues(DefaultFlushEntityEventListener.java:178)
       at org.hibernate.event.internal.DefaultFlushEntityEventListener.onFlushEntity(DefaultFlushEntityEventListener.java:135)
       at org.hibernate.event.internal.AbstractFlushingEventListener.flushEntities(AbstractFlushingEventListener.java:216)
       at org.hibernate.event.internal.AbstractFlushingEventListener.flushEverythingToExecutions(AbstractFlushingEventListener.java:85)
       at org.hibernate.event.internal.DefaultAutoFlushEventListener.onAutoFlush(DefaultAutoFlushEventListener.java:44)
       at org.hibernate.internal.SessionImpl.autoFlushIfRequired(SessionImpl.java:1389)
       at org.hibernate.internal.SessionImpl.list(SessionImpl.java:1882)
       at org.hibernate.internal.CriteriaImpl.list(CriteriaImpl.java:366)
       at org.hibernate.internal.CriteriaImpl.uniqueResult(CriteriaImpl.java:388)
       at com.atlassian.crowd.embedded.hibernate2.HibernateUserDao.lambda$internalFindUser$21(HibernateUserDao.java:559)
       at com.atlassian.crowd.embedded.hibernate2.HibernateUserDao$$Lambda$1777/720364257.doInHibernate(Unknown Source)
       at org.springframework.orm.hibernate5.HibernateTemplate.doExecute(HibernateTemplate.java:385)
       at org.springframework.orm.hibernate5.HibernateTemplate.execute(HibernateTemplate.java:337)
       at com.atlassian.crowd.embedded.hibernate2.HibernateUserDao.internalFindUser(HibernateUserDao.java:556)
       at com.atlassian.crowd.embedded.hibernate2.HibernateUserDao.internalFindByName(HibernateUserDao.java:470)
       at com.atlassian.crowd.embedded.hibernate2.HibernateUserDao.findByName(HibernateUserDao.java:457)
       at com.atlassian.confluence.user.crowd.CachedCrowdUserDao.findUserInternal(CachedCrowdUserDao.java:95)
       at com.atlassian.confluence.user.crowd.CachedCrowdUserDao.lambda$findUser$0(CachedCrowdUserDao.java:131)
       at com.atlassian.confluence.user.crowd.CachedCrowdUserDao$$Lambda$1776/501151391.get(Unknown
   

4. Since request mail has to be sent to only 5 users,its seems a waste to load all the users

This will affect our largest customers with a huge no. of groups and users

Expected Results

Confluence shouldn't load all users from the groups to determine 5 users to send notification to

Actual Results

Confluence loads all users from all the groups that have been assigned admin space permission

Workaround

Currently, there is no known workaround for this behavior. A workaround will be added here when available