[LU-9790] ost-pools test_23a: Quota failed to enable in ZFS 0.7.0-rc5

• Status: Resolved
• Project: Lustre
• Type: Bug
• Priority: Critical
• Reporter: Maloo
• Assignee: Jinshan Xiong (Inactive)
• Resolution: Duplicate
• Severity: 3

Description

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

This issue relates to the following test suite run: https://testing.hpdd.intel.com/test_sets/02cee546-6dc5-11e7-bb38-5254006e85c2.

The sub-test test_23a failed with the following error:

test_23a failed with 2

Please provide additional information about the failure here.

Info required for matching: ost-pools 23a

MDT Console Log:

02:26:22:[10452.862564] Lustre: DEBUG MARKER: /usr/sbin/lctl conf_param lustre.quota.ost=ug
02:26:22:[10455.041842] LustreError: 0-0: lustre-MDT0000: can't enable quota enforcement since space accounting isn't functional. Please run tunefs.lustre --quota on an unmounted filesystem if not done already

Issue Links

• duplicates: LU-7991 Add project quota for ZFS

Comments

[Nathaniel Clark]

Also affecting conf-sanity/test_32b
https://testing.hpdd.intel.com/test_sets/c60e413e-6dd7-11e7-9a2c-5254006e85c2
https://testing.hpdd.intel.com/test_sets/a1f429ce-68cf-11e7-baf7-5254006e85c2

[Peter Jones]

Jinshan

Can you please advise on this one?

Thanks

Peter

[Nathaniel Clark]

jay,

I thought about that, but I'm seeing this on ost-pools also, not just conf-sanity.

[Jinshan Xiong (Inactive)]

It turned out this is about project quota, which is supported by ZFS yet.

Patch https://github.com/zfsonlinux/zfs/pull/6290 need to be landed to ZFS then we can support it.

[Jinshan Xiong (Inactive)]

We need a fix to check if ZFS supports project quota, otherwise Lustre shouldn't bother initializing it.

The problem is this test case doesn't use project quota at all, let's investigate it further

[Nathaniel Clark]

Other ost-pool failures by one of the ZFS 0.7.0 patches: https://review.whamcloud.com/#/c/22569
https://testing.hpdd.intel.com/test_sets/6e0d9806-6452-11e7-baf7-5254006e85c2
https://testing.hpdd.intel.com/test_sets/bfd472fa-637a-11e7-baf7-5254006e85c2
https://testing.hpdd.intel.com/test_sets/31711f74-4587-11e7-b558-5254006e85c2

This failure is also ZFS 0.7.0-rc2 https://build.hpdd.intel.com/job/lustre-ppc/781:
https://testing.hpdd.intel.com/test_sets/eecf38c0-457e-11e7-91f4-5254006e85c2

[Jinshan Xiong (Inactive)]

It seems like there are some problem for quota information to transfer to OSTs. As I can see,

$LFS quota -v -u $RUNAS_ID $dir

can't sometimes get expected results.

I'm not an expert of quota; this issue need to be fixed by somebody else.

[Peter Jones]

Niu

Are you able to advise on this one?

Thanks

Peter

[Niu Yawei (Inactive)]

This should be a regression caused by:

    LU-2435 osd-zfs: use zfs native dnode accounting

    If ZFS userobj_accounting feature exists, use it for osd-zfs object
    accounting; otherwise estimate object use by block space.

    Disable quota check in conf-sanity:32b and sanity-quota:7e if the
    underlying zfs version is below 0.7.0. For conf-sanity test, if the
    running ZFS is version 0.7.0 or higher, upgrade the zpool image to
    enable native dnode accounting. That ensures we also test the upgrade
    procedure.

Which mistakenly changed osd_acct_index_lookup() to return -ENOENT when some ID not being found in the accounting file. This regression was addressed by https://review.whamcloud.com/#/c/27661/ (which hasn't been landed yet).

[Peter Jones]

Thanks for the analysis Niu

[Jinshan Xiong (Inactive)]

I have rebased the 0.7.0-rc5 patch on patch 27661 to see how it goes.

[Peter Jones]

AFAIK the patch Niu mentioned in https://jira.hpdd.intel.com/browse/LU-9790?focusedCommentId=203296&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-203296 has now landed but the problem persists. What are the next steps?