[
https://issues.apache.org/jira/browse/CLK-636?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12841113#action_12841113
]
Henry Saputra commented on CLK-636:
-----------------------------------
Actually the code in the PropertyUtils could still allow two threads could go
in to call Map.get and both may return null so the operation will execute to
override the value in the Map cache.
I am proposing to change the Map instead of <String, Object> but to <String,
Future>. Then the first thread call to Map.get will check if there exist task
to get the cache value.
If already exist just call get on it otherwise create one and by leveraging
ConcurrentHashMap.putIfAbsent we could guarantee only one Future object exist
for the String key.
> Replace EDU.oswego.cs.dl.util.concurrent.ConcurrentReaderHashMap with
> java.util.concurrent.ConcurrentHashMap
> ------------------------------------------------------------------------------------------------------------
>
> Key: CLK-636
> URL: https://issues.apache.org/jira/browse/CLK-636
> Project: Click
> Issue Type: Improvement
> Components: core
> Affects Versions: 2.2.0
> Reporter: Henry Saputra
> Attachments: concurrentreader_patch.diff
>
>
> Since Click required Java SDK 1.5 or later, we could leverage the
> java.util.concurrent.ConcurrentHashMap class to replace
> EDU.oswego.cs.dl.util.concurrent.ConcurrentReaderHashMap class so reducing
> the Click runtime dependency.
> In my opinion here are some good reasons why:
> 1. The ConcurrentHashMap class in Java SDK is more efficient since it
> utilizes internal hash classes to support better granularity and concurrency
> compare to simple syncrhonized on the instance like in
> DU.oswego.cs.dl.util.concurrent.ConcurrentReaderHashMap.
> 2. Looking at the use case ConcurrentReaderHashMap in Click, it used to cache
> the OGNL expression (please correct me if I am wrong). This scenario does not
> need exclusive lock on update which is the intended/ preferred use case for
> ConcurrentReaderHashMap. If there is a miss on OGNL expression on a name in
> the cache, it will cerate one and put it to the map if no other thread has
> not. So it will still perform as well as or better locking entire table.
> However, if we do need exclusive lock on update, we can simulate
> ConcurrentReaderHashMap with ConcurrentHashMap by setting concurrencyLevel to
> one.
> 3. The ConcurrentHashMap support generic which is part of task being done to
> move Click code to Java generics.
> 4. Looks like the EDU.oswego.cs.dl.util.concurrent.ConcurrentReaderHashMap
> class is created by Doug Lea before contributions to java.util.concurrent
> packages in Java 1.5 SDK so the code may no longer optimized.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
