[LU-1794] sanity test_51b: @@@@@@ FAIL: test_51b failed with 28

• Status: Resolved
• Project: Lustre
• Affects Version/s: Lustre 2.4.0
• Type: Bug
• Priority: Minor
• Reporter: Li Wei (Inactive)
• Assignee: WC Triage
• Resolution: Cannot Reproduce

Environment

My local single-node VM "cluster". Latest master at the time of writing. Using default local.sh MDSSIZE.

• Severity: 3

Description

It looks this is fairly easy to reproduce.

== sanity test 51b: mkdir .../t-0 --- .../t-70000 ====================== 10:55:03 (1346122503)
 - created 10000 (time 1346122511.69 total 8.47 last 8.47)
 - created 20000 (time 1346122520.57 total 17.35 last 8.88)
 - created 30000 (time 1346122529.69 total 26.46 last 9.11)
mkdir(/mnt/lustre/d51b/t-32343) error: No space left on device
total: 32343 creates in 28.63 seconds: 1129.77 creates/second
 sanity test_51b: @@@@@@ FAIL: test_51b failed with 28 
  Trace dump:
  = /root/lustre-release/lustre/tests/test-framework.sh:3638:error_noexit()
  = /root/lustre-release/lustre/tests/test-framework.sh:3660:error()
  = /root/lustre-release/lustre/tests/test-framework.sh:3893:run_one()
  = /root/lustre-release/lustre/tests/test-framework.sh:3922:run_one_logged()
  = /root/lustre-release/lustre/tests/test-framework.sh:3742:run_test()
  = sanity.sh:3150:main()
Dumping lctl log to /tmp/test_logs/1346122501/sanity.test_51b.*.1346122532.log
Dumping logs only on local client.
FAIL 51b (36s)

[root@h221f tests]# umount /mnt/mds1
[root@h221f tests]# mount -t ldiskfs -o loop /tmp/lustre-mdt1
lustre-mdt1    lustre-mdt1_2  
[root@h221f tests]# mount -t ldiskfs -o loop /tmp/lustre-mdt1 /mnt/mds1
[root@h221f tests]# df /mnt/mds1
Filesystem           1K-blocks      Used Available Use% Mounted on
/tmp/lustre-mdt1        149944     30856    109088  23% /mnt/mds1
[root@h221f tests]# df -i /mnt/mds1
Filesystem            Inodes   IUsed   IFree IUse% Mounted on
/tmp/lustre-mdt1      100000    2678   97322    3% /mnt/mds1

Comments

[Andreas Dilger]

What do lfs df and lfs df -i report for the filesystem before and after the test? I expect this is something specific to your test environment due to the size of your MDT, but it does mean something unusual is happening. The test Gould be checking in advance that there are enough inodes to complete the test, so it must be hitting some other limit and/or the free inodes are being misreported (e.g. not enough I does on te OSTs).

What stripe count is used by default?

[Andreas Dilger]

Close old ticket.