[
https://issues.apache.org/jira/browse/HADOOP-10169?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13850514#comment-13850514
]
Jason Lowe commented on HADOOP-10169:
-------------------------------------
As Jing pointed out, the function needs to be synchronized as written. It's
true that once initialized the code will no longer modify the map, but there's
a race during the initialization itself. If two or more threads call this
function before it's initialized then they can both attempt to put() at the
same time. Alternatively, a thread could be late to the party and be calling
get() just as the initializing thread is calling put() which is also a thread
safety violation.
One possible way to avoid the synchronization bottleneck is to use a volatile
boolean to indicate whether it's initialized so the vast majority of callers
don't have to grab a lock. Something like this pseudo-code:
{code}
volatile boolean isInitialized = false;
if (!isInitialized) {
synchronized (gcInfoCache) {
if (!isInitialized) {
.....
gcInfoCache.put(...);
}
}
isInitialized = true;
}
return gcInfoCache.get(...);
{code}
> remove the unnecessary synchronized in JvmMetrics class
> --------------------------------------------------------
>
> Key: HADOOP-10169
> URL: https://issues.apache.org/jira/browse/HADOOP-10169
> Project: Hadoop Common
> Issue Type: Improvement
> Components: metrics
> Affects Versions: 3.0.0, 2.2.0
> Reporter: Liang Xie
> Assignee: Liang Xie
> Priority: Minor
> Attachments: HADOOP-10169.txt
>
>
> When i looked into a HBase JvmMetric impl, just found this synchronized seems
> not essential.
--
This message was sent by Atlassian JIRA
(v6.1.4#6159)
