[
https://issues.apache.org/jira/browse/TS-4259?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15181983#comment-15181983
]
Alan M. Carroll commented on TS-4259:
-------------------------------------
Also, this is a bit of a hack. Once TS-3612 is done we should change the logic
in all of these cases to have the transaction / session callback to the
{{HttpSM}} instance when it is about to be destroyed so the internal pointers
can be reliably updated instead of sort of guessing where that should happen.
Logging is unfortunately another issue to consider as if these objects go out
of scope, data in them cannot be logged when the {{HttpSM}} shuts down.
Separating "close" and "destroy" is probably a good idea for this reason and
also it would provide better detection of use after close (with a nice clear
warning instead of a random crash).
> ua_session should not be NULL'd out if not closed
> -------------------------------------------------
>
> Key: TS-4259
> URL: https://issues.apache.org/jira/browse/TS-4259
> Project: Traffic Server
> Issue Type: Bug
> Reporter: Alan M. Carroll
>
> In {{HttpSM::tunnel_handler_server}} the tunnel may get shut down due to an
> issue with the server by calling {{HttpTunnel::chain_abort_all}}. The code
> presumes this will close {{ua_session}} because it is in the tunnel but it
> has been observed this is not always true. Therefore the tunnel state of
> {{ua_entry}} which corresponds to {{ua_session}} should be checked.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
