From Bugzilla Helper: User-Agent: Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.6) Gecko/20040510 Description of problem: Only one mount at a time is possible, ie the first mount, on any node, succeeds. A second mount attempt, of the same file system on any different node, blocks. Umount'ing the file system on the first node allows the mount attempt on the second node to return successfully. Node names of the nodes are initially set to mebsuta-1, mebsuta-2, mebsuta-3, mebsuta-4 When node names of all nodes are changed to mebsuta1, mebsuta2, mebsuta3, mebsuta4, and all references in configs are changed accordingly, then the mount problem disappears, ie correct cluster operation is achieved by changing only the host names (and references to host names) and re-starting. Version-Release number of selected component (if applicable): GFS-6.0.0-7 How reproducible: Always Steps to Reproduce: Situation I have a four-node cluster connected to an IBM SAN using QLogic FC adapters. I'm using GFS 6.0.0-7 and kernel 2.4.21-15.0.3.EL. Fencing is intended to be using Cisco MDS switches, but for now is simply set to fence_manual just to get things going. When I try to run a single lock manager configuration I _am_ able to create and mount GFS file systems on any one node at a time. However I am unable to mount the filesystem on more than one node at a time. All the gfs-related modules are correctly loaded on all nodes. The gfs-related service scripts all start successfully on all nodes. lock_gulmd is running on all nodes. gulm_tool nodelist shows one node as master, and others as clients. The data and ccs pools are created and visible to all nodes as shown by pool_info. I have created /etc/sysconfig/gfs on all nodes as follows: POOLS="pool0 testgfs_cca" CCS_ARCHIVE="/dev/pool/testgfs_cca" Actual Results: only one node can mount a file system Expected Results: all nodes should mount file system. Additional info: It's not clear if the problem was fixed by the removal of the hyphen in the node names, or by the shortening of the names to eight characters.
yeah, the first 8 characters of a node's name *have* to be unique right now when using gulm with gfs. making sure the first 8 chars are unique is the current work around. A real fix is none trival, but something I have planned on doing.
I've had another customer encounter this. If the fix is still pending, can we at least update the docuementation and get a KB entry describing this and the work-around?
A fix is in CVS HEAD (could be 6.1) There is a fix ready for 6.0. BUt both of these made some changes to the recovery code, and really need to get hit hard by QA first. As for the documentation or KB (what's KB?) is that soemthing I do, I do I re-assign this bug or just hope someone else does something?
KB == Knowledge Base. Contact our support staff, and they can help you out with that. I thinkit's something we can self-serve, but Support controlls access to it, and it's best to submit entries to them for inclusion at this time, AFAIK.
Verified with GFS-6.0.0-15 vs. GFS-6.0.0-7. Used hostnames linklink-13, linklink-14, linklink-15. Mounted 14 filesystems on all three nodes with GFS 6.0.0-15.
An errata has been issued which should help the problem described in this bug report. This report is therefore being closed with a resolution of ERRATA. For more information on the solution and/or where to find the updated files, please follow the link below. You may reopen this bug report if the solution does not work for you. http://rhn.redhat.com/errata/RHBA-2004-490.html