On Sep 21, 2006 07:55 -0400, Theodore Tso wrote: > On Wed, Sep 20, 2006 at 02:37:17PM +0100, Alastair McKinstry wrote: > > Attached are some patches used by Lustre; the patches are written > > by cluster filesystems, inc. and are apparently expected to be in the > > next release of e2fsprogs upstream, 1.40. > > > > Lustre is a cluster filesystem that uses a modified ext3 fs underneath; > > including additional ext3 attributes not supported by standard ext3. > > For lustre you use this modified e2fsck. > > 1) These patches will not enter e2fsprogs upstream with some > significant changes, which I am currently working on, and > > 2) These patches are not enough to enable e2fsprogs check Lustre > filesystems. Clusterfs maintains their own set of changes that goes > beyond this set of patches. These patches are designed to get the > Lustre tree closer in sync with e2fsprogs, but is not sufficient for > e2fsprogs to check Lustre filesystems.
To clarify: - many of these patches have already been accepted upstream - it is not absolutely required to have the CFS-patched e2fsprogs, IF the Lustre OST filesystems are not using the extents feature - the extents support that is expected to be in e2fsprogs-1.40 will work with the lustre extents patches, which is the only incompatible on-disk format change used by lustre - the lfsck tool (which is one of the few CFS patches that has not been submitted upstream yet) is not required in order for e2fsck to work with a Lustre filesystem. If the individual ext3 filesystems are internally consistent (i.e. e2fsck completed, in case of corruption) then Lustre itself will still work in the face of inter-filesystem inconsistencies (i.e. objects missing/extra on OSTs compared to MDS filesystem), possibly with EIO or ENOENT errors returned to applications or leakage of space on the OST filesystems. I would of course be happy to get as many patches into the upstream e2fsprogs package. Cheers, Andreas -- Andreas Dilger Principal Software Engineer Cluster File Systems, Inc. -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
