[Ocfs2-users] Unstable Cluster Node

[Mark Fasheh]

On Fri, Nov 30, 2007 at 03:25:27AM -0800, rain c wrote:
> uname -a
> Linux webhost1 2.6.18-028stab039 #2 SMP Tue Aug 21 17:49:05 UTC 2007 i686 GNU/Linux
> 
> Both nodes are in the same bladecenter an directly connected with 1Gbit/s by the baldecenters internal ethernet switch.
> 
> One of the nodes stops working at least once a day with the following messages:
> 
> Nov 23 19:05:02 webhost2 kernel: (4424,3):o2net_sendpage:827 ERROR: sendpage of size 24 to node webhost1 (num 0) at 10.2.0.70:7777 failed with 4294967264
> Nov 23 19:05:02 webhost2 kernel: (6774,0):dlm_send_remote_convert_request:395 ERROR: status = -107
> Nov 23 19:05:02 webhost2 kernel: (4997,2):dlm_send_remote_convert_request:395 ERROR: status = -107
> Nov 23 19:05:02 webhost2 kernel: (4997,2):dlm_wait_for_node_death:374 225202289F954729807AACECEBB2D2AC: waiting 5000ms for notification of death of node 0
> Nov 23 19:05:02 webhost2 kernel: (6774,0):dlm_wait_for_node_death:374 225202289F954729807AACECEBB2D2AC: waiting 5000ms for notification of death of node 0
> 
> 
> After that the node hangs and even does not reboot although /proc/sys/kernel/panic and /proc/sys/kernel/panic_on_oops are set to 1.
> 
> Can anybody please help me to understand the error messages and make that node more stable?

This was fixed in 2.6.23, and there's a version of the patch backported to
2.6.22:

http://www.kernel.org/pub/linux/kernel/people/mfasheh/ocfs2/backports/2.6.22.14/broken-out/0006-ocfs2-Retry-sendpage-if-it-returns-EAGAIN.patch


Unfortunately, It doesn't look like the patch would apply to 2.6.18 without
some work.


Can you upgrade to the latest stable 2.6.23 kernel? If you do, take the
time to apply the other patches in our 2.6.23 backports:

http://www.kernel.org/pub/linux/kernel/people/mfasheh/ocfs2/backports/2.6.23.7/
	--Mark

--
Mark Fasheh
Senior Software Developer, Oracle
mark.fasheh at oracle.com