[
https://issues.apache.org/jira/browse/HADOOP-16504?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16904546#comment-16904546
]
Lisheng Sun commented on HADOOP-16504:
--------------------------------------
[~jojochuang]
The problem now is that since listen queue is full, TCP can drop many packet
and resend packet after a while. So tcp connection to be slow or even timed
out.
Changing this default value can reduce client connection timeout when the
request is large for nn or dn.
Update the v1 patch and upload it. Thank you.
> Increase ipc.server.listen.queue.size default from 128 to 256
> -------------------------------------------------------------
>
> Key: HADOOP-16504
> URL: https://issues.apache.org/jira/browse/HADOOP-16504
> Project: Hadoop Common
> Issue Type: Improvement
> Reporter: Lisheng Sun
> Assignee: Lisheng Sun
> Priority: Major
> Attachments: HADOOP-16504.000.patch, HADOOP-16504.001.patch
>
>
> Because ipc.server.listen.queue.size default value is too small, TCP's
> ListenDrop indicator along with the rpc request large.
> The upper limit of the system's semi-join queue is 65636 and maximum number
> of fully connected queues is 1024.
> {code:java}
> [work@c3-hadoop-srv-talos27 ~]$ cat /proc/sys/net/ipv4/tcp_max_syn_backlog
> 65536
> [work@c3-hadoop-srv-talos27 ~]$ cat /proc/sys/net/core/somaxconn
> 1024
> {code}
> I think this default value should be adjusted.
--
This message was sent by Atlassian JIRA
(v7.6.14#76016)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
