[LU-4016] metadata-updates failure: attributes check failed

• Type: Bug
• Resolution: Fixed
• Priority: Minor
• Fix Version/s: Lustre 2.6.0, Lustre 2.4.2, Lustre 2.5.1
• Affects Version/s: Lustre 2.5.0
• Environment: Build # 1687; OpenSFS cluster with combined MGS/MDS, single OSS with two OSTs, four clients; one agent (c07), one with robinhood/db running (c08) and just running as Lustre clients (c09, c10)
• Severity: 3
• Assignee: James Nunez (Inactive)
• Reporter: James Nunez (Inactive)

Description

Test results at https://maloo.whamcloud.com/test_sessions/eb9d697c-2658-11e3-8d26-52540035b04c

metadata-updates fails with "attributes check failed". Node logs don't exist, but the suite_log shows:

Checking file(s) attributes ... 
c08: /lustre/scratch/d0.metadata-updates/c07/testfile [ uid gid size mode ] expected : 0 0 1024 0644 ;  got : root 0 1024 0644 
c08: Wrong file attributes
c09: /lustre/scratch/d0.metadata-updates/c07/testfile [ uid gid size mode ] expected : 0 0 1024 0644 ;  got : root 0 1024 0644 
c09: Wrong file attributes
pdsh@c10: c08: ssh exited with exit code 56
c07: /lustre/scratch/d0.metadata-updates/c07/testfile [ uid gid size mode ] expected : 0 0 1024 0644 ;  got : root 0 1024 0644 
c07: Wrong file attributes
pdsh@c10: c09: ssh exited with exit code 56
pdsh@c10: c07: ssh exited with exit code 56
c10: /lustre/scratch/d0.metadata-updates/c07/testfile [ uid gid size mode ] expected : 0 0 1024 0644 ;  got : root 0 1024 0644 
c10: Wrong file attributes
pdsh@c10: c10: ssh exited with exit code 56
 metadata-updates : @@@@@@ FAIL: attributes check failed 
  Trace dump:
  = /usr/lib64/lustre/tests/test-framework.sh:4264:error_noexit()
  = /usr/lib64/lustre/tests/metadata-updates.sh:227:main()

Although this was run during HSM testing, the copytool was not started and HSM was not enabled on the MDS, but all HSM attributes were in the test configuration files.

Comments

[Peter Jones]

Landed for 2.4.2 and 2.6. Will land for 2.5.1 shortly

[Jian Yu]

Patch http://review.whamcloud.com/8052 was cherry-picked to Lustre b2_4 branch for 2.4.2.

[James Nunez (Inactive)]

Reopened to apply patch to 2.5 and 2.4.

2.4 patch: http://review.whamcloud.com/#/c/8333/
2.5 patch: http://review.whamcloud.com/#/c/8332/

[James Nunez (Inactive)]

The patch should apply to 2.4 and 2.5 with no problems. I'll request that it be cherry-picked against 2.4 and 2.5.

[James A Simmons]

Can we cherry pick it against 2.4.1. We will later need it cherry picked against 2.5.1

[James Nunez (Inactive)]

Landed to master.

[Stephen Champion]

http://review.whamcloud.com/8052

Against master, applies cleanly to 2.4.1 and works in my environment.

[James A Simmons]

Sure go right ahead

[Stephen Champion]

That works. I assumed that test_UID needed to be numerical, but it's not actually used anywhere else. I think we should rename it (because it's not a UID anymore) and change GID, too, for consistency.

Mind if I submit a patch with those changes, based on your idea?

[James A Simmons]

This is a old bug that has been around since the 1.6 days. I thought I was the only one that saw this problem. The way I work around it is with the below patch.

diff -urp metadata-updates.sh.orig metadata-updates.sh
— metadata-updates.sh.orig 2013-05-16 09:54:56.000000000 -0400
+++ metadata-updates.sh 2013-05-20 13:33:15.000000000 -0400
@@ -32,7 +32,7 @@
NEW_ATIME="2001-01-01 GMT"
NEW_MTIME="2005-05-05 GMT"

-test_UID=$(id -u)
+test_UID=$(id -u -n)
test_GID=$(id -g)

SUMFILE=$TESTDIR/mdsum
@@ -106,7 +106,7 @@
do_nodesv $NODES_TO_USE "set $TRACE;
for HOST in ${HOSTS//,/ } ; do
TESTFILE=$TESTDIR/\\\$HOST/$FILE;

• tmp=\\\$(stat -c \\\"%u %g %s 0%a\\\" \\\$TESTFILE);
  + tmp=\\\$(stat -c \\\"%U %g %s 0%a\\\" \\\$TESTFILE);
  echo \\\"\\\$TESTFILE [ uid gid size mode ] expected : $attr ; got : \\\$tmp \\\";
  if [ x\\\"\\\$tmp\\\" != x\\\"$attr\\\" ] ; then
  echo \\\"Wrong file attributes\\\";