[LU-4123] lfsck: @@@@@@ FAIL: /data/test/output isn't a shared directory

• Status: Resolved
• Project: Lustre
• Affects Version/s: Lustre 2.4.1
• Fix Version/s: Lustre 2.6.0, Lustre 2.5.1
• Type: Bug
• Priority: Minor
• Reporter: Stephen Champion
• Assignee: Jian Yu
• Resolution: Fixed
• Labels: patch

Environment

Lustre 1.8 -> master
multi rail IB cluster

• Severity: 3

Description

check_write_access() compares the local node name to the host name passed via
xxx_HOST env variables. This breaks when there is a mismatch.

My cluster uses the nodename as the host name of the management interface
(ethernet). The Infiniband interface(s) use 'hostname-ibX'. To use an IB
interface for acceptance tests, the ib host name must be used in xxx_HOST
env variables.

I have a patch to use the remote nodename in check_write_access().

Comments

[Stephen Champion]

http://review.whamcloud.com/#/c/8009/

[Stephen Champion]

A little more detail - my startup script runs from an NFS exported directory. It specifies hosts by name of an IB interface:

1. grep _HOST= run-acc-accfs.sh
   mgs_HOST=n013-ib1
   mds_HOST=n013-ib1
   mds1_HOST=n013-ib1
   ost1_HOST=n008-ib1
   ost_HOST=n008-ib1
   ost2_HOST=n009-ib1

check_logdir() creates the files:
touch $dir/check_file.$(hostname -s)

but must run on the node to actually be useful:
do_rpc_nodes "$list" check_logdir $dir
check_write_access $dir "$list" || return 1

This works just fine:

1. ls
   check_file.n008 rpmlist-client.n006 rpmlist-server.n009 shared
   check_file.n009 rpmlist-client.n007 rpmlist-server.n013
   check_file.n013 rpmlist-server.n008 run-acc-accfs.sh

But because 'n013' != 'n013-ib1', the check for shared access fails.

The patch insures that it will check for the same file name that is created, regardless of host/interface name arrangement.

[Jian Yu]

Patch landed on master branch.

[Jian Yu]

Patch was back-ported to Lustre b2_4 branch: http://review.whamcloud.com/8343