Lustre discuss - Nov 2009 - lustre automount mounting problem with listing dir content deeper than mountpoint

Hello,
after fixing the broken hardware stuff one problem remains.
We are using lustre with automount since over one year without problems.
Since the hardware failure a few days ago (a Gigabit switch and the SATA
Backplane in one MDS) the following happens.

Client quadcore1. The lustre system is mounted under
''/misc/data'' (mountpoint) via automount.
''mount'':
mds1 at tcp0:mds2 at tcp0:/scia on /misc/data type lustre (rw)

Now doing:
''umount /misc/data''
Nov 10 10:14:24 quadcore1 LustreError:
8751:0:(ldlm_request.c:986:ldlm_cli_cancel_req()) Got rc -108 from cancel RPC:
canceling anyway
Nov 10 10:14:24 quadcore1 LustreError:
8751:0:(ldlm_request.c:1575:ldlm_cli_cancel_list()) ldlm_cli_cancel_list: -108
Nov 10 10:14:24 quadcore1 LustreError:
8751:0:(connection.c:144:ptlrpc_put_connection()) NULL connection
Nov 10 10:14:24 quadcore1 LustreError:
8751:0:(connection.c:144:ptlrpc_put_connection()) Skipped 13 previous similar
messages
Nov 10 10:14:24 quadcore1 Lustre: client ffff81001606dc00 umount complete

Trying to automount and digging one or more Dirs deeper than the mountpoint
(client console hangs after this command):
''ls -la /misc/data/OneDirDeeper''
Nov 10 10:14:33 quadcore1 automount[2797]: attempting to mount entry /misc/data
Nov 10 10:14:34 quadcore1 Lustre: Client scia-client has started
Nov 10 10:14:34 quadcore1 automount[2797]: mount(generic): mounted mds1 at
tcp0:mds2 at tcp0:/scia type lustre on /misc/data
Nov 10 10:14:34 quadcore1 automount[2797]: mounted /misc/data
Nov 10 10:14:34 quadcore1 LustreError:
3115:0:(lib-move.c:95:lnet_try_match_md()) Matching packet from
12345-192.168.16.122 at tcp, match 86814 length 1336 too big: 1272 left, 1272
allowed
Nov 10 10:14:34 quadcore1 Lustre: 3115:0:(lib-move.c:1647:lnet_parse_put())
Dropping PUT from 12345-192.168.16.122 at tcp portal 10 match 86814 offset 128
length 1336: 2

In a new console (releasing the freezed console above):
''umount /misc/data'':
Nov 10 10:15:38 quadcore1 Lustre: setting import scia-MDT0000_UUID INACTIVE by
administrator request
Nov 10 10:15:38 quadcore1 Lustre: Skipped 13 previous similar messages
Nov 10 10:15:38 quadcore1 LustreError:
3122:0:(mdc_locks.c:841:mdc_intent_getattr_async_interpret())
ldlm_cli_enqueue_fini: -4
Nov 10 10:15:38 quadcore1 LustreError:
3122:0:(mdc_locks.c:841:mdc_intent_getattr_async_interpret()) Skipped 2 previous
similar messages
Nov 10 10:15:38 quadcore1 LustreError:
8754:0:(client.c:716:ptlrpc_import_delay_req()) @@@ IMP_INVALID  req at
ffff8102244c6c00 x86895/t0 o101->scia-MDT0000_UUID at
192.168.16.122@tcp:12/10 lens 440/1400 e 0 to 100 dl 0 ref 1 fl Rpc:/0/0 rc 0/0
Nov 10 10:15:38 quadcore1 LustreError:
8754:0:(client.c:716:ptlrpc_import_delay_req()) Skipped 2 previous similar
messages
Nov 10 10:15:38 quadcore1 LustreError: 8754:0:(mdc_locks.c:586:mdc_enqueue())
ldlm_cli_enqueue: -108
Nov 10 10:15:38 quadcore1 LustreError: 8754:0:(mdc_locks.c:586:mdc_enqueue())
Skipped 76 previous similar messages
Nov 10 10:15:38 quadcore1 LustreError: 8754:0:(dir.c:261:ll_get_dir_page()) lock
enqueue: rc: -108
Nov 10 10:15:38 quadcore1 LustreError: 8754:0:(dir.c:261:ll_get_dir_page())
Skipped 2 previous similar messages
Nov 10 10:15:38 quadcore1 LustreError: 8754:0:(dir.c:415:ll_readdir()) error
reading dir 4167519/1275738219 page 6: rc -108
Nov 10 10:15:38 quadcore1 LustreError: 8754:0:(dir.c:415:ll_readdir()) Skipped 2
previous similar messages

There are no messages on the MDS or OSTs related to this.
Doing an ''ls -la /misc/data'' works ok and the lustre system
gets mounted properly on /misc/data.
The above scenario is reproducable on all clients.
The system works fine when the lustre system is mounted statically or after the
mount is done in a proper way.

lustre-1.6.6
vanilla-2.6.22.19

Thanks and Regards
Heiko

Hello,

just found that it seems to be related to a link problem inside the lustre
directory structure.
One of our users is doing this (mountpoint /misc/data):

ln -s  /misc/data/index.gz  /misc/data/WorkOrSo/index.gz

Than he overwrites the file /misc/data/index.gz thereby moving it to a different
OST.
If i delete these links (...../WorOrSo/) than lustre gets properly mounted with
the automounter even diving into deeper dirs in one go.
i.e. on an unmounted lustre (without these links) doing a ''ls -la
/misc/data/WorkOrSo/WhateverGoesHere/'' is just fine.

Does lustre has a problem with _moving_ links meaning when the file pointed to
is moved to a different OST ?

Regards
Heiko