[LU-17694] sanity-compr test_184d: last component index number got assigned even if it was not used after layout swap

• Type: Bug
• Resolution: Unresolved
• Priority: Minor
• Labels: always_except hsm layout pfl
• Severity: 3
• Assignee: WC Triage
• Reporter: Maloo

Description

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

This issue relates to the following test suite run: https://testing.whamcloud.com/test_sets/7b32c252-7710-4d4e-a06b-1a7b3509b5a7

test_184d failed with the following error:

lovea  -E 1048576 -c 1 -S 1048576 -L raid0 -i 1 -E -1 -c 1 -S 4194304 -L raid0 -i -1 !=  -E 1048576 -c 1 -S 1048576 -L raid0 -i 1 -E -1 -c 1 -S 4194304 -L raid0 -i 2

Test session details:
clients: https://build.whamcloud.com/job/lustre-reviews/102975 - 4.18.0-477.27.1.el8_8.x86_64
servers: https://build.whamcloud.com/job/lustre-reviews/102975 - 4.18.0-477.27.1.el8_lustre.x86_64

<<Please provide additional information about the failure here>>

After doing swap layout, compare the components of 2 files, the last component's index from -1 changed to a specific number even that part of component not been used yet.

== sanity test 184d: allow stripeless layouts swap ======= 12:27:47 (1709987267)
Succeed in opening file "/mnt/lustre/d184d.sanity/f184d.sanity-2"(flags=O_CREAT)
Succeed in opening file "/mnt/lustre/d184d.sanity/f184d.sanity-3"(flags=O_CREAT)
 -E 1048576 -c 1 -S 1048576 -L raid0 -i 1 -E -1 -c 1 -S 4194304 -L raid0 -i -1
 -E 1048576 -c 1 -S 1048576 -L raid0 -i 1 -E -1 -c 1 -S 4194304 -L raid0 -i 2
 sanity test_184d: @@@@@@ FAIL: lovea  -E 1048576 -c 1 -S 1048576 -L raid0 -i 1 -E -1 -c 1 -S 4194304 -L raid0 -i -1 !=  -E 1048576 -c 1 -S 1048576 -L raid0 -i 1 -E -1 -c 1 -S 4194304 -L raid0 -i 2 
  Trace dump:
  = /usr/lib64/lustre/tests/test-framework.sh:6975:error()
  = sanity.sh:20111:test_184d()
  = /usr/lib64/lustre/tests/test-framework.sh:7315:run_one()
  = /usr/lib64/lustre/tests/test-framework.sh:7375:run_one_logged()
  = /usr/lib64/lustre/tests/test-framework.sh:7201:run_test()
  = sanity.sh:20116:main()

VVVVVVV DO NOT REMOVE LINES BELOW, Added by Maloo for auto-association VVVVVVV
sanity-compr test_184d - lovea -E 1048576 -c 1 -S 1048576 -L raid0 -i 1 -E -1 -c 1 -S 4194304 -L raid0 -i -1 != -E 1048576 -c 1 -S 1048576 -L raid0 -i 1 -E -1 -c 1 -S 4194304 -L raid0 -i

Issue Links

• is related to: LU-16904 initial test script to run sanity.sh and sanityn.sh with PFL layout
• is related to: LU-9479 sanity test 184d 244: don't instantiate PFL component when taking group lock
• is related to: LU-9341 PFL: append should not instantiate full layout

Comments

[Andreas Dilger]

Skipped by patch: https://review.whamcloud.com/54797 "LU-16904 tests: Fix sanity test 34h, 184d and 101i when PFL layout is used"

[Sarah Liu]

sure, I will make a patch

this is what I got from a passed 184d with non-PFL
https://testing.whamcloud.com/sub_tests/e04863c5-5200-47b7-9462-de84452b4c99

== sanity test 184d: allow stripeless layouts swap =================================================== 14:17:28 (1712240248)
Succeed in opening file "/mnt/lustre/d184d.sanity/f184d.sanity-2"(flags=O_CREAT)
Succeed in opening file "/mnt/lustre/d184d.sanity/f184d.sanity-3"(flags=O_CREAT)
 -c 1 -S 1048576 -L raid0 -i 1
 -c 1 -S 1048576 -L raid0 -i 1
/mnt/lustre/d184d.sanity/f184d.sanity-1: trusted.lov: No such attribute

[Andreas Dilger]

sarah, for now, please make a patch to skip this test with always_except if $fs_STRIPEPARAMS is a PFL layout.

It is also strange that the files have any layout assigned, since they are opened with "O_LOV_DELAY_CREATE", and the test is explicitly checking that there is no trusted.lov xattr at the end of the swaps, so it isn't clear what get_layout_param is printing in the non-PFL test case?

[Andreas Dilger]

It looks like this is the same problem affecting the group lock (LU-9479), since "lfs swap_layouts" and "lfs migrate --block" are internally using the group lock to prevent other clients from modifying the file. This is not true with "lfs migrate" which uses a lease and can fail if the file is modified during copy, but running "lfs swap_layouts" goes from a mostly-empty PFL layout to a fully-populated PFL layout afterward.