You are viewing a plain text version of this content. The canonical link for it is here.
Posted to common-issues@hadoop.apache.org by "Daryn Sharp (JIRA)" <ji...@apache.org> on 2018/11/02 20:09:00 UTC
[jira] [Commented] (HADOOP-15897) Port range binding fails due to
socket bind race condition
[ https://issues.apache.org/jira/browse/HADOOP-15897?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16673615#comment-16673615 ]
Daryn Sharp commented on HADOOP-15897:
--------------------------------------
We're testing this internally. Can't write a test case because java doesn't expose the ability to independently bind and listen, thus preventing the interleaving of bind/listen necessary to produce the issue on demand. Ie. t1:bind, t2:bind, t1:listen, t2:listen.
> Port range binding fails due to socket bind race condition
> ----------------------------------------------------------
>
> Key: HADOOP-15897
> URL: https://issues.apache.org/jira/browse/HADOOP-15897
> Project: Hadoop Common
> Issue Type: Bug
> Components: ipc
> Affects Versions: 2.0.2-alpha
> Reporter: Daryn Sharp
> Assignee: Daryn Sharp
> Priority: Major
> Attachments: HADOOP-15897.patch
>
>
> Java's {{ServerSocket#bind}} does both a bind and listen. At a system level, multiple processes may bind to the same port but only one may listen. Java sockets are left in an unrecoverable state when a process loses the race to listen first.
> Servers that compete over a listening port range (ex. App Master) will fail the entire range after a collision. The IPC layer should make a better effort to recover from failed binds.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: common-issues-unsubscribe@hadoop.apache.org
For additional commands, e-mail: common-issues-help@hadoop.apache.org