[LU-14017] Interop: sanityn tests 43a and 45a fail with 'mkdir must succeed'

• Status: Resolved
• Project: Lustre
• Affects Version/s: Lustre 2.14.0, Lustre 2.15.0
• Type: Bug
• Priority: Minor
• Reporter: James Nunez (Inactive)
• Assignee: WC Triage
• Resolution: Done
• Labels: interop

Environment

2.13.52.52 or later servers with earlier (< 2.13.52.52 ) clients

• Severity: 3

Description

sanityn tests 43a and 45a fail with 'mkdir must succeed' after the patch https://review.whamcloud.com/30880 for LU-10235 landed to master in FEB 2020.

Starting on 23 FEB 2020, interop testing between master servers and older clients, sanityn test_43a and test_45a fail with

== sanityn test 43a: pdirops: unlink vs mkdir ======================================================== 12:50:41 (1582462241)
CMD: trevis-14vm4 /usr/sbin/lctl set_param -n ldlm.namespaces.*mdt*.lru_size=clear
CMD: trevis-14vm4 /usr/sbin/lctl get_param ldlm.namespaces.*mdt*.lock_unused_count ldlm.namespaces.*mdt*.lock_count
ldlm.namespaces.mdt-lustre-MDT0000_UUID.lock_count=35
CMD: trevis-14vm4 lctl set_param fail_loc=0x80000145
fail_loc=0x80000145
mkdir: cannot create directory '/mnt/lustre2/f43a.sanityn': File exists
 sanityn test_43a: @@@@@@ FAIL: mkdir must succeed 
  Trace dump:
  = /usr/lib64/lustre/tests/test-framework.sh:5900:error()
  = /usr/lib64/lustre/tests/sanityn.sh:1887:test_43a()

and

== sanityn test 45a: pdirops: rename src vs mkdir ==================================================== 12:55:13 (1582462513)
CMD: trevis-14vm4 /usr/sbin/lctl set_param -n ldlm.namespaces.*mdt*.lru_size=clear
CMD: trevis-14vm4 /usr/sbin/lctl get_param ldlm.namespaces.*mdt*.lock_unused_count ldlm.namespaces.*mdt*.lock_count
ldlm.namespaces.mdt-lustre-MDT0000_UUID.lock_count=35
CMD: trevis-14vm4 lctl set_param fail_loc=0x80000145
fail_loc=0x80000145
mkdir: cannot create directory '/mnt/lustre2/f45a.sanityn': File exists
 sanityn test_45a: @@@@@@ FAIL: mkdir must succeed 
  Trace dump:
  = /usr/lib64/lustre/tests/test-framework.sh:5900:error()
  = /usr/lib64/lustre/tests/sanityn.sh:2170:test_45a()

We may just need a Lustre version check to skip or run different tests in this situation.

Here are a few link to logs for past 138 failures:
2.13.52.52 servers/2.12.4 clients - https://testing.whamcloud.com/test_sets/048a979e-1a4a-4178-8bdf-568314985c82
2.13.52.52 servers/2.13.0 clients - https://testing.whamcloud.com/test_sets/c390a7a8-c0d8-456b-9708-e24b07508f2c
2.13.56.6 servers/2.12.5 clients - https://testing.whamcloud.com/test_sets/95c71443-8c73-41c7-8b9e-c1ada7bc7b48
2.13.56.6 servers/2.13.0 clients - https://testing.whamcloud.com/test_sets/4e2ff8da-0590-4175-818f-658a777b3aea

Issue Links

• is related to: LU-10235 mkdir should check for directory exis...

Comments

[Andreas Dilger]

Both 43a and 45a are modified by that patch, so I suspect that we just need to skip these subtests in interop testing before 2.13.53.

[Andreas Dilger]

The patch https://review.whamcloud.com/41674 "LU-10235 mdt: mdt_create: check EEXIST without lock" was landed to b2_12 and changes these tests to match master, so they will no longer fail in interop testing with 2.15/master. However, they will presumably start failing with 2.10 clients, but I don't think we care to fix that.

NB: the most recent interop test failure was 2.12.8.6, but the patch landed as 2.12.8.21.