[
https://issues.apache.org/jira/browse/ZOOKEEPER-1173?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17091121#comment-17091121
]
maoling commented on ZOOKEEPER-1173:
------------------------------------
[~thkoch]
_*--->The idea may have been to save memory for equal ACL lists. However
there's no code that ever removes an*_ *A**CL* _*list that is*_ *not* _*used
anymore.*_
Look at the code in *ReferenceCountedACLCache#removeUsage:*
if the reference count of that ACL list <=0 means that ACL list is useless any
longer and we will remove them, so we don't have that issue in the current
codes.
{code:java}
long newCount = referenceCounter.get(acl).decrementAndGet();
if (newCount <= 0) {
referenceCounter.remove(acl);
aclKeyMap.remove(longKeyMap.get(acl));
longKeyMap.remove(acl);
}
{code}
> Server never forgets old ACL lists
> ----------------------------------
>
> Key: ZOOKEEPER-1173
> URL: https://issues.apache.org/jira/browse/ZOOKEEPER-1173
> Project: ZooKeeper
> Issue Type: Bug
> Components: server
> Reporter: Thomas Koch
> Assignee: Thomas Koch
> Priority: Major
> Labels: acl
>
> The ACL stuff in DataTree.java reimplements a kind of reference system. The
> idea may have been to save memory for equal ACL lists. However there's no
> code that ever removes an ACL list that is not used anymore.
> Related:
> - The ACL stuff could be in a separate class so that DataTree.java is not
> such a big beast anymore.
> - It's risky to have mutable objects (list) as keys in a HashMap.
> An idea to solve this: Have ACL lists as members of the datatree nodes.
> Lookup already existing ACL lists in a java.util.WeakHashMap.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
