[Linux-cluster] problem adding new node to an existing cluster
rpeterso at redhat.com
Mon Jan 5 23:09:18 UTC 2009
----- "Joseph M. Greenseid" <Joseph.Greenseid at ngc.com> wrote:
| I have a new question. When I created this file system a year ago, I
| didn't anticipate needing any additional nodes other than the original
| 3 I set up. Consequently, I have 3 journals. Now that I've been told
| to add a fourth node, is there a way to add a journal to an existing
| file system that resides on a volume that has not been expanded (the
| docs appear to read that you can only do it to an expanded volume
| because the additional journal(s) take up additional space). My file
| system isn't full, though my volume is fully used by the formatted GFS
| file system.
| Is there anything I can do that won't involve destroying my existing
| file system?
Journals for gfs file systems are carved out during mkfs. The rest of the
space is used for data and metadata. So there are only two ways to
make journals: (1) Do another mkfs which will destroy your file system
or (2) if you're using lvm, add more storage with something like
lvresize or lvextend, then use gfs_jadd to add the new journal to the
new chunk of storage.
We realize that's a pain, and that's why we took away that restriction
in gfs2. In gfs2, journals are kept as a hidden part of the file system,
so they can be added painlessly to an existing file system without
adding storage. So I guess a third option would be to convert the file
system to gfs2 using gfs2_convert, add the journal with gfs2_jadd, then
use it as gfs2 from then on. But please be aware that gfs2_convert had some
serious problems until the 5.3 version that was committed to the cluster
git tree in December, (i.e. the very latest and greatest "RHEL5", "RHEL53",
"master", "STABLE2" or "STABLE3" versions in the cluster git (source code)
tree.) Make ABSOLUTELY CERTAIN that you have a working & recent backup and
restore option before you try this. Also, the GFS2 kernel code prior to
5.3 is considered tech preview as well, so not ready for production use.
So if you're not building from source code, you should wait until RHEL5.3
or Centos5.3 (or similar) before even considering this option.
Red Hat GFS
More information about the Linux-cluster