[jboss-jira] [JBoss JIRA] (JGRP-1792) "No buffer space available" when sending on Mac OS

[Bela Ban (JIRA)]

    [ https://issues.jboss.org/browse/JGRP-1792?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12983494#comment-12983494 ] 

Bela Ban edited comment on JGRP-1792 at 7/10/14 5:21 AM:
---------------------------------------------------------

OK, the reason the error didn't happen on 3.4 is probably that some changes in 3.5 (no copying of looped back messages etc) caused the senders to be send more data in a shorter amount of time than in 3.4, and therefore the pressure on the send buffer is higher.
Also, Mac OS does throw an exception when a packet is discarded due to a full buffer, whereas other OSes silently discard the packet.
In 3.4, the problem can be reproduced without JGroups (attached {{mcast2}}).
Since JGroups retransmits dropped messages, the adopted solution will be to suppress the warning (or show it only once a minute): the performance is actually good. It is the user's resposibility to provide adequate send buffer space on Mac OS.


was (Author: belaban):
OK, the reason the error didn't happen on 3.4 is probably that some changes in 3.5 (no copying of looped back messages etc) caused the senders to be send more data in a shorter amount of time than in 3.4, and therefore the pressure on the send buffer is higher.
In 3.4, the problem can be reproduced without JGroups (attached {{mcast2}}).
Since JGroups retransmits dropped messages, the adopted solution will be to suppress the warning (or show it only once a minute): the performance is actually good. It is the user's resposibility to provide adequate send buffer space on Mac OS.

> "No buffer space available" when sending on Mac OS
> --------------------------------------------------
>
>                 Key: JGRP-1792
>                 URL: https://issues.jboss.org/browse/JGRP-1792
>             Project: JGroups
>          Issue Type: Bug
>      Security Level: Public(Everyone can see) 
>            Reporter: Bela Ban
>            Assignee: Bela Ban
>            Priority: Minor
>             Fix For: 3.5
>
>         Attachments: mcast2.java
>
>
> When sending multicast messages (with MPerf & fast.xml), the following message appears:
> {noformat}
> 8981 [ERROR] UDP: JGRP000029: A: failed sending message to cluster (1053 bytes): java.io.IOException: No buffer space available, 
> headers: MPerf: DATA999, NAKACK2: [MSG, seqno=1003], UDP: [cluster_name=mperf]
> {noformat}
> The messages *are* received, but the error message occurs on almost every message (1 sender thread sending 1000 messages).
> With {{UdpPerf}} (also shipped with JGroups), which also multicasts messages, the error doesn't happen.
> With {{UPerf}}, which uses unicasts, the error doesn't occur either.
> Todo: compare UdpPerf to MPerf (UDP protocol) to see what the diff is.



--
This message was sent by Atlassian JIRA
(v6.2.6#6264)