[LU-17747] interop: sanity test_130g: filefrag printed 175 < 700 extents

• Type: Bug
• Resolution: Unresolved
• Priority: Minor
• Affects Version/s: Lustre 2.14.0, Lustre 2.15.4
• Severity: 3
• Assignee: Dongyang Li
• Reporter: Andreas Dilger

Description

This issue was created by maloo for Alena Nikitenko <[e-mail]>

This issue relates to the following test suite run: https://testing.whamcloud.com/test_sets/5907cf45-fd1e-4f8b-a959-32150206c06d

test_130g failed with the following error:

filefrag printed 175 < 700 extents

Test session details:
clients: https://build.whamcloud.com/job/lustre-b_2_14/8 - 4.18.0-348.7.1.el8_5.x86_64
servers: https://build.whamcloud.com/job/lustre-master/4518 - 5.14.0-284.30.1_lustre.el9.x86_64

trevis-27vm1 - Client 1 (2.14.0.21, x86_64)
trevis-27vm2 - Client 2 (2.14.0.21, x86_64)
trevis-27vm3 - OST 1, OST 2, OST 3, OST 4, OST 5, OST 6, OST 7 (2.15.62, x86_64)
trevis-27vm6 - MDS 1 (2.15.62, x86_64)

VVVVVVV DO NOT REMOVE LINES BELOW, Added by Maloo for auto-association VVVVVVV
sanity test_130g - filefrag printed 175 < 700 extents

Issue Links

• is related to: LU-14310 Interop: sanity test 130g fails with error 'failed to setstripe -C 700 /mnt/lustre/f130g.sanity'
• is related to: LU-10465 increase default stripe size to 4MB
• is related to: LU-11912 reduce number of OST objects created per MDS Sequence

Comments

[Dongyang Li]

https://review.whamcloud.com/c/fs/lustre-release/+/54851

[Andreas Dilger]

Looking at one of the test failures:
https://testing.whamcloud.com/test_logs/d9eaa5a7-f091-4711-8427-08d3fb0cd0bb/show_text

  sanity test_130g: @@@@@@ FAIL: filefrag printed 175 < 700 extents 

It shows the file was created with 700 stripes, and just returned 175 stripes with filefrag, so I don't think the problem is on the object creation side, especially since the stripe counts are always the same:

 filefrag list 175 extents in file with stripecount 700
 /mnt/lustre/f130g.sanity
 lmm_stripe_count:  700
 lmm_stripe_size:   4194304
 lmm_pattern:       raid0,overstriped 

However, I do see that lmm_stripe_size = 4MB, so this would mean the "dd" is not writing enough data to the file to put 1MB on each stripe, so this is likely causing the test to fail. It looks like "-S1M" is added on master in commit v2_15_58-43-gea18d7da59, but it isn't on b2_14 or b2_15 clients where the the test is running:

    onyx-51vm4 - Client 1 (2.14.0.21, x86_64)
    onyx-51vm5 - Client 2 (2.14.0.21, x86_64)
    onyx-55vm4 - OST 1-7 (2.15.61.194, x86_64)
    onyx-55vm7 - MDS 1 (2.15.61.194, x86_64)

It looks like this (and other) subtests need to be fixed on b2_15 to handle the change in default stripe size, and this subtest excluded by autotest from running with b2_14 clients by filing an ATM ticket to request this for clients < 2.15.5 (assuming this patch will land on b2_15 for that release). This needs to be handled at the Autotest level because we can't retroactively fix the 2.14.0 release to work with 4MB default stripe count.