[
https://issues.apache.org/jira/browse/TS-4507?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15344600#comment-15344600
]
ASF GitHub Bot commented on TS-4507:
------------------------------------
Github user jpeach commented on a diff in the pull request:
https://github.com/apache/trafficserver/pull/717#discussion_r68081721
--- Diff: proxy/http/Http1ClientSession.h ---
@@ -181,7 +189,12 @@ class Http1ClientSession : public ProxyClientSession
MIOBuffer *read_buffer;
IOBufferReader *sm_reader;
- C_Read_State read_state;
+
+ /*
+ * Volatile should not be necessary, but there appears to be a bug in
the 4.9 rhel gcc
+ * compiler that was using an old version of read_state to make
decisions in really_destroy
+ */
+ volatile C_Read_State read_state;
--- End diff --
If there are threaded dependencies on this, then we should force an atomic
read or a memory acquire/release. I think the volatile is too subtle to be
depended on.
> It is still possible for SSN_CLOSE hook to be called before TXN_CLOSE hook
> --------------------------------------------------------------------------
>
> Key: TS-4507
> URL: https://issues.apache.org/jira/browse/TS-4507
> Project: Traffic Server
> Issue Type: Bug
> Components: Core
> Reporter: Susan Hinrichs
> Assignee: Susan Hinrichs
> Fix For: 7.0.0
>
>
> One of our plugins will occasionally crash. It appears there is still a path
> for HTTP2 that has the SSN_CLOSE hook close before the TXN_CLOSE hook.
> Working through solutions that delay the SSN_CLOSE hook until after all the
> TXN_CLOSE hooks, but does not lose the SSN_CLOSE.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
