[LU-16673] sanity test_125: failures with aarch64 servers

• Type: Bug
• Resolution: Fixed
• Priority: Minor
• Fix Version/s: Lustre 2.16.0, Lustre 2.15.5
• Severity: 3
• Assignee: Xinliang Liu
• Reporter: Maloo

Description

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

This issue relates to the following test suite run with aarch64 servers:
https://testing.whamcloud.com/test_sets/38335ace-3096-4f6a-a18a-f252f4d7bf6c

test_125 failed with the following error:

setfacl /mnt/lustre/d125.sanity failed

Test session details:
clients: 2.15.54.79 - 4.18.0-348.23.1.el8_lustre331.aarch64
servers: 2.15.54.79 - 4.18.0-348.23.1.el8_lustre331.aarch64

Also failed the following subtests (listing all here for searching):
test_150f
test_150g
test_151
test_154a
test_154b
test_156

VVVVVVV DO NOT REMOVE LINES BELOW, Added by Maloo for auto-association VVVVVVV
sanity test_125 - setfacl /mnt/lustre/d125.sanity failed

Issue Links

• is related to: LU-16541 sanity test_64f: buffered io, not write rpc: grants mismatch: 12656640, expected 4218880
• is related to: LU-15259 SLES15.2 sanity test_103a test_125 test_154a: setfacl: Option -m: Invalid argument near character 3

Comments

[Gerrit Updater]

"Oleg Drokin <[e-mail]>" merged in patch https://review.whamcloud.com/c/fs/lustre-release/+/55433/
Subject: LU-16673 tests: add checking for user USER0
Project: fs/lustre-release
Branch: b2_15
Current Patch Set:
Commit: e92212c52d53f16d2de8d8d67988b2e0fc3c9223

[Gerrit Updater]

"Frederick Dilger <[e-mail]>" uploaded a new patch: https://review.whamcloud.com/c/fs/lustre-release/+/55433
Subject: LU-16673 tests: add checking for user USER0
Project: fs/lustre-release
Branch: b2_15
Current Patch Set: 1
Commit: a455867b7c95c7c4cd2adf32517aff16454c9138

[Peter Jones]

Landed for 2.16

[Gerrit Updater]

"Oleg Drokin <[e-mail]>" merged in patch https://review.whamcloud.com/c/fs/lustre-release/+/50501/
Subject: LU-16673 tests: add checking for user USER0
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: b7b55534b9b5d4d58025082ae5403938853b168c

[Xinliang Liu]

BTW, RHEL 9 uses 4K PAGE_SIZE.

[Xinliang Liu]

Our aarch64 CI clusters run on CentOS 8 which use 64KB PAGE_SIZE and thus use 64KB blocksize for ldiskfs.

It sounds a good idea to store files < 64KB on MDT with DoM for 64KB PAGE_SIZE cluster.

 

We also do test locally on openEuler which use 4K PAGE_SIZE.

 

[Andreas Dilger]

xinliang, out of curiosity, are you formatting the OSTs with 64KB blocksize, or still 4KB blocksize (assuming ldiskfs)? There are definitely some interesting benefits from 64KB blocksize:

• larger directories with fewer htree levels
• 256x fewer block groups due to 16x bits per bitmap x 16x larger blocks
• more efficient allocation for large files

This would mean less space efficiency for files < 64KB in size and for the fraction of a block at the end of each file, but for files <= 64KB it would be possible to use DoM to store them on the MDT.

[Xinliang Liu]

And for those tests 150f,150g,151,156, failing might relate to 64K page size, because these tests pass on aarch64 4k page size cluster.

[Gerrit Updater]

"xinliang <[e-mail]>" uploaded a new patch: https://review.whamcloud.com/c/fs/lustre-release/+/50501
Subject: LU-16673 tests: add checking for user USER0
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 7559cd4ae840828720b5ae5f1ccf927ca96280e2

[Xinliang Liu]

Should add a check for user USER0 in the test.