[Lustre-discuss] Lustre 1.8.8 lfs getstripe does not display OBDs
Roger Sersted
rs1 at aps.anl.gov
Wed Jan 16 08:32:40 PST 2013
A colleague from another division pointed out Lustre 1.8.8's "lfs getstripe"
does not display the OBDs. I had upgraded my clients at about the same time my
array failed. It never occurred to me the command's output had changed. <face
palm> :-)
Roger S.
Roger Sersted wrote:
> I recently had OST0000 crash and the filesystem on that OST was unrecoverable.
> The rest of Lustre FS seems to function correctly. I built a new OST0000 on
> a different array and mounted it on to the same OSS as the previous OST0000.
> Along the way, I also ran tunefs.lustre on the MDSes and OSSes to setup
> failover partners.
>
> I can mount the Lustre FS on a client and as expected, an ls -l displays this:
>
> total 71975032
> ?--------- ? ? ? ? ? p15a
> -rw-r--r-- 1 root root 4194304000 Oct 30 17:48 t2
>
> In the above example, p15a was striped on to the failed OST and t2 was striped
> on to other OSTs. That's expected and I'm not concerned.
>
> However, the "lfs getstripe" command does not list the OBDs. It should list
> something like this:
>
> lfs getstripe t2
> OBDS:
> 0: lustrefc-OST0000_UUID ACTIVE
> 1: lustrefc-OST0001_UUID ACTIVE
> 2: lustrefc-OST0002_UUID ACTIVE
> 3: lustrefc-OST0003_UUID ACTIVE
> 4: lustrefc-OST0004_UUID ACTIVE
> 5: lustrefc-OST0005_UUID ACTIVE
> 6: lustrefc-OST0006_UUID ACTIVE
> 7: lustrefc-OST0007_UUID ACTIVE
> 8: lustrefc-OST0008_UUID ACTIVE
> 9: lustrefc-OST0009_UUID ACTIVE
> 10: lustrefc-OST000a_UUID ACTIVE
> 11: lustrefc-OST000b_UUID ACTIVE
> t2
> obdidx objid objid group
> 6 3 0x3 0
> 1 22090374 0x1511286 0
> 7 3 0x3 0
>
> Instead it lists:
>
> lfs getstripe p15
> p15
> lmm_stripe_count: 6
> lmm_stripe_size: 1048576
> lmm_stripe_offset: 0
> obdidx objid objid group
> 0 20333533 0x13643dd 0
> 4 20293080 0x135a5d8 0
> 1 20319599 0x1360d6f 0
> 2 20320251 0x1360ffb 0
> 5 20197030 0x1342ea6 0
> 3 16153240 0xf67a98 0
>
> Could someone shed some light on this problem? Is it safe to use the
> filesystem in this state?
>
> Thanks,
>
> Roger S.
>
>
>
>
>
> _______________________________________________
> Lustre-discuss mailing list
> Lustre-discuss at lists.lustre.org
> http://lists.lustre.org/mailman/listinfo/lustre-discuss
More information about the lustre-discuss
mailing list