[
https://issues.apache.org/jira/browse/HDFS-13488?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16449534#comment-16449534
]
Yiqun Lin commented on HDFS-13488:
----------------------------------
Haven't looked into this, one question:
{code:java}
+ if (conf.getBoolean(
+ RBFConfigKeys.DFS_ROUTER_CLIENT_REJECT_OVERLOAD,
+ RBFConfigKeys.DFS_ROUTER_CLIENT_REJECT_OVERLOAD_DEFAULT)) {
+ workQueue = new ArrayBlockingQueue<>(numThreads);
+ } else {
+ workQueue = new LinkedBlockingQueue<>();
+ }
{code}
Why we use {{ArrayBlockingQueue}} as the work queue type? Any preference reason
for this? From my understanding of this, {{ArrayBlockingQueue}} uses the same
ReentrantLock to control the put/take operations, but {{LinkedBlockingQueue}}
uses two separate ReentrantLock(takeLock and putLock). And the latter will have
a high efficiency for executing requests.
> RBF: Reject requests when a Router is overloaded
> ------------------------------------------------
>
> Key: HDFS-13488
> URL: https://issues.apache.org/jira/browse/HDFS-13488
> Project: Hadoop HDFS
> Issue Type: Sub-task
> Reporter: Íñigo Goiri
> Assignee: Íñigo Goiri
> Priority: Major
> Attachments: HDFS-13488.000.patch, HDFS-13488.001.patch
>
>
> A Router might be overloaded when handling special cases (e.g. a slow
> subcluster). The Router could reject the requests and the client could try
> with another Router. We should leverage the Standby mechanism for this.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
