[lustre-discuss] Errors stat-ing file

[Alastair Basden]

Hi all,

Wondering whether anyone can help resolve this.

We have some files which are known about by the MDS, but it has lost links to the OSTs, i.e.
A /bin/ls works on the file, but when a stat is require (e.g. ls -al)
ls -al restart.99
ls: cannot access 'restart.99': No such file or directory
-?????????   ? ?        ?              ?            ? restart.99

So the MDS knows about the file, but the links to it's components on the 
OSSs seem to be broken.  The files aren't striped (i.e. single stripe), 
and the backend is ZFS for the OSS, and ldiskfs for the MDS.  It is a 
2.12.6 system.

An lfsck doesn't fix the problem.

However, we are wondering about various options for the lctl lfsck_start 
command, e.g.
the -o option (i.e. lctl lfsck_start -A -o).

Which should "Repair orphan OST-objects for layout LFSCK".

I don't think we want the -c option ("create the lost OST-object for
dangling LOV EA"), as I think that might just create an empty file?

My feeling is that the data is there on the OST, just the MDT no longer 
knows how to find it, and so presumably there is a simple repair operation 
to get it back.

Any thoughts?

THanks,
Alastair.