[
https://issues.apache.org/jira/browse/PROTON-907?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Gordon Sim updated PROTON-907:
--
Attachment: PROTON-907-workaround.patch
The issue appears to be that on the affected platforms, when unable to connect,
the file descriptor is not marked as writeable.
Though it hits the read error, messenger only closes the 'tail' of the
transport as a result. The head is closed when an error is returned from send,
but as the socket is not writeable, send is never called.
I don't know what the real fix for this is, messenger is an area of the code
I'm even less familiar with. Fwiw the attached patch works around the issue and
passes all the existing tests. It works by explicitly closing the head of the
transport if there is an error on reading from the socket and the connection
has not been closed by the peer.
Qpid Proton Point to Point Hang on CentOS 6 pn_messenger_send
-
Key: PROTON-907
URL: https://issues.apache.org/jira/browse/PROTON-907
Project: Qpid Proton
Issue Type: Bug
Components: proton-c
Affects Versions: 0.8, 0.9.1
Environment: CentOS 6 (both VM and native 64-bit) and RHEL 6
Reporter: Frank Quinn
Priority: Critical
Attachments: PROTON-907-workaround.patch
See thread at
http://qpid.2158936.n2.nabble.com/Strange-behaviour-for-pn-messenger-send-on-CentOS-6-td7625846.html.
Key points:
* pn_messenger_send will hang on CentOS 6 if the destination is not yet up
* Works fine on Fedora 21 and 22 (by 'fine', i mean it will attempt to send,
fail and move on)
* Can be recreated by running the send.c application when recv.c is not yet
running
* Proton burns CPU as it hangs
This effectively deadlocks our application. So far, I’ve tried compiling qpid
proton c myself (both 0.8 and 0.9.1), setting pn_messenger_send timeout to 1
(it was previously -1), turning off iptables entirely and disabling selinux
and rebooting but no luck.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
