[
https://issues.apache.org/jira/browse/HADOOP-9436?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13615663#comment-13615663
]
Kihwal Lee commented on HADOOP-9436:
------------------------------------
I tried to minimize contention in the patch. From a comment section in the
patch:
{code}
* There are three different points where concurrent accesses happen.
* 1) User-to-groups mapping: This is thread safe data structure, so
* putting and getting don't need to make them safe.
* 2) cachedGroups: This needs to be protected from concurrent accesses.
* Also, the content needs to be kept in sync with 1).
* Synchronize on cachedGroups whenever updating mappings.
* 3) refresh: Refresh requests need to be serialized.
* Caller to be synchronized on refreshLock.
*
* By separating into three, cache lookups are never explicitly blocked.
* Regular cache add activities and refresh can mostly overlap.
{code}
Improved:
* Refresh does not leave removed users in the cache.
* No more direct modification of values in the map. It wasn't thread safe.
* No more rebuild-from-scratch whenever a group is added.
* Addition of fine-grained locking.
> NetgroupCache does not refresh membership correctly
> ---------------------------------------------------
>
> Key: HADOOP-9436
> URL: https://issues.apache.org/jira/browse/HADOOP-9436
> Project: Hadoop Common
> Issue Type: Bug
> Affects Versions: 3.0.0, 2.0.3-alpha, 0.23.7
> Reporter: Kihwal Lee
> Assignee: Kihwal Lee
> Attachments: HADOOP-9436.patch
>
>
> NetgroupCache is used to get around the problem of inability to obtain a
> single user-to-groups mapping from netgroup. For example, the ACL code
> pre-populates this cache, so that any user-group mapping can be resolved for
> all groups defined in the service.
> However, the current refresh code only adds users to existing groups, so a
> loss of group membership won't take effect. This is because the internal
> user-groups mapping cache is never invalidated. If this is simply invalidated
> on clear(), the cache entries will build up correctly, but user-group
> resolution may fail during refresh, resulting in incorrectly denying accesses.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
