On 07/26/2014 12:02 AM, Pranith Kumar Karampuri wrote:
On 07/26/2014 11:06 AM, Pranith Kumar Karampuri wrote:
On 07/26/2014 03:06 AM, Joe Julian wrote:
How can it come about? Is this from replacing a brick days ago? Can
I prevent it from happening?
[2014-07-25 07:00:29.287680] W [fuse-resolve.c:546:fuse_resolve_fd]
0-fuse-resolve: migration of basefd
(ptr:0x7f17cb846444 inode-gfid:87544fde-9bad-46d8-b610-1a8c93b85113)
did not complete, failing fop with
EBADF (old-subvolume:gv-nova-3 new-subvolume:gv-nova-4)
It's critical because it causes a segfault every time. :(
Joe,
This is fd migration code. When a brick layout changes (graph
change) the file needs to be re-opened in the new graph. This re-open
seemed to have failed. It leads to crash probably because extra unref
in failure code path. Could you add brick/mount logs to the bug
https://bugzilla.redhat.com/show_bug.cgi?id=1123289. What is the
configuration of the volume?
I checked the code, I don't see any extra unrefs as of now. Please
provide the details I asked for in the bug.
CC Raghavendra G, Raghavendra Bhat who know this code path a bit more.
Added the log and volume info.
I assume I can prevent any more nodes from crashing by migrating any VMs
whose image was hosted on the former brick to another compute note,
which would establish a whole new fd?
_______________________________________________
Gluster-devel mailing list
Gluster-devel@gluster.org
http://supercolony.gluster.org/mailman/listinfo/gluster-devel