[lustre-discuss] Command "lfs find --obd" finds nothing

Tung-Han Hsieh thhsieh at twcp1.phys.ntu.edu.tw
Thu Oct 13 20:30:04 PDT 2022


Dear All,

Occasionally, we need to explicitly move some data out of a specific
OST. For example, the device of that OST should be replaced, or the
OST is almost full and we want to balence the amount of data across
the OSTs. In these cases we usually run:

	lfs find --obd <OST_UUID> <mount_point>

to get the list of files that located in the OST <OST_UUID>, and then
move them out.

But sometimes, running this command on an OST cannot get any output,
even though that the OST is full of data. So far we encounter this
problem in OSTs with ldiskfs backends, in Lustre-2.12.X and previous
versions. For example, a couple of months ago, we added a large device
with 7 partitions to an existing Lustre-2.12.6 system, so 7 new OSTs
were added. 6 of them are normal, but one of them has this problem.
We don't know why it occurs.

This kind of abnormality does not affect any other operations. In fact
we still can use the command:

	lfs getstripe <filename>

to explicitly find files located in that OST. But this way is not very
convenient.

We are wondering that whether there are missing flags in creating the
ldiskfs file system for that problematic OST ? So we have the following
questions:

1. Could anyone suggest commands to retreive the enabled feature flags
   of a running ldiskfs backend OST. Then we can compare the normal and
   abnormal OSTs, and then manually fix it.

2. If it is not due to the missing feature flags in the backend ldiskfs,
   does anyone know how to fix it ? If it is not possible to fix it for
   existing data files in that OST, it is also very appreciated to fix
   it for incoming new files.

Thank you very much for your kindly help in advance.

Best Regards,

T.H.Hsieh


More information about the lustre-discuss mailing list