[LU-3720] Test failure on test suite sanity, subtest test_130d: FIEMAP failed

• Status: Resolved
• Project: Lustre
• Fix Version/s: Lustre 2.5.0
• Type: Bug
• Priority: Minor
• Reporter: Maloo
• Assignee: Emoly Liu
• Resolution: Fixed
• Severity: 3

Description

This issue was created by maloo for James Nunez <[e-mail]>

This issue relates to the following test suite run: http://maloo.whamcloud.com/test_sets/e20c0452-ff4a-11e2-a3fb-52540035b04c.

The sub-test test_130d failed with the following error:

FIEMAP on /mnt/lustre/f.sanity.130d failed; returned len 1216 for OST 0006 instead of 1024

Output from test log:

== sanity test 130d: FIEMAP (N-stripe file) == 16:38:06 (1375832286)
7+0 records in
7+0 records out
7340032 bytes (7.3 MB) copied, 0.103369 s, 71.0 MB/s
Filesystem type is: bd00bd0
File size of /mnt/lustre/f.sanity.130d is 7340032 (7168 blocks of 1024 bytes)
 ext:     device_logical:        physical_offset: length:  dev: flags:
   0:        0..    1023:     136192..    137215:   1024: 0006: network
   1:     1024..    1215:     213492..    213683:    192: 0006: network
   2:        0..    1023:     221184..    222207:   1024: 0000: network
   3:     1024..    1215:     142344..    142535:    192: 0000: network
   4:        0..    1023:     148480..    149503:   1024: 0002: network
   5:     1024..    1215:     140892..    141083:    192: 0002: network
   6:        0..    1023:     136192..    137215:   1024: 0003: network
   7:     1024..    1215:     141108..    141299:    192: 0003: network
   8:        0..    1023:     155648..    156671:   1024: 0004: network
   9:     1024..    1151:     163688..    163815:    128: 0004: network
  10:        0..    1023:     136192..    137215:   1024: 0005: network
  11:     1024..    1151:     141124..    141251:    128: 0005: network
/mnt/lustre/f.sanity.130d: 7 extents found
 sanity test_130d: @@@@@@ FAIL: FIEMAP on /mnt/lustre/f.sanity.130d failed; returned len 1216 for OST  0006 instead of 1024 

Issue Links

• is related to: LU-3633 sanity.sh test_101d failed for 'dd fa...

Comments

[James Nunez (Inactive)]

I've looked at the last 10 failures of this subtest and it is always preceded with 101d/102b/102c/116a failures, i.e. a full OST. So, this may be caused by a single OST filling and would be fixed by the patch for LU-3633.

[Andreas Dilger]

This test could be made a bit more robust to use the actual number of stripes allocated (via "lfs getstripe -c") instead of the requested number of stripes. The test is expecting each OST to get a specific amount if data based in the size and stripe count, but since an OST is full more data is written to each of the other OSTs. This same problem might also be hit if there are no preallocated objects on an OST, so it is worthwhile to fix it even outside if the ENOSPC problem.

[Emoly Liu]

I can reproduce this bug easily if an OST is full.

The patch is at http://review.whamcloud.com/7303 .

[Emoly Liu]

patch landed for 2.5

[Gerrit Updater]

"Andreas Dilger <[e-mail]>" uploaded a new patch: https://review.whamcloud.com/c/fs/lustre-release/+/53072
Subject: LU-3720 tests: use actual stripe count in sanity/130b
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 6f58bd703cf2d47540f631721a4da7361e9ed3cd