[
https://issues.apache.org/jira/browse/TS-1559?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13491388#comment-13491388
]
Uri Shachar commented on TS-1559:
---------------------------------
When this patch is integrated we should give a heads-up to protocol plugin
developers/users -- it causes a slight behavior change -- In some cases we can
now get a TS_EVENT_VCONN_WRITE_READY on the client write VIO after the server
read vio has been freed (so if you'll segfault if you try to reenable). Since
the plugin will always get an EOS event beforehand it's pretty easy to adjust
for this.
> Server side termination not handled properly when a PluginVC/Protocol Plugin
> is used
> -------------------------------------------------------------------------------------
>
> Key: TS-1559
> URL: https://issues.apache.org/jira/browse/TS-1559
> Project: Traffic Server
> Issue Type: Bug
> Components: HTTP, Plugins
> Affects Versions: 3.3.1
> Reporter: Yossi Gottlieb
> Attachments: pluginvc_tunnel.diff
>
>
> Using a Protocol Plugin along with a PluginVC (HttpConnect) and HTTP CONNECT
> command to create tunnel. When the server drops the connection, the protocol
> plugin will not be notified until the connection times out.
> HttpSM ends up calling PluginVC::do_io_shutdown() which sets the appropriate
> flags but takes no action. I suspect this would affect real socket VCs as
> well, but in that case the shutdown() on the socket write side would cause
> the client to react and close its own side as well.
> The proposed fix solves my specific problem but may not address all related
> issues (with PluginVCs or other types of VCs as well).
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
