[LU-1902] disable sanity-quota.sh for OFD on 2.3

• Status: Resolved
• Project: Lustre
• Affects Version/s: Lustre 2.3.0, Lustre 2.4.0
• Type: Bug
• Priority: Major
• Reporter: Maloo
• Assignee: Niu Yawei (Inactive)
• Resolution: Won't Fix
• Labels: USE_OFD
• Severity: 3

Description

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

This issue relates to the following test suite run: https://maloo.whamcloud.com/test_sets/ec6c56cc-f9ba-11e1-b8d8-52540035b04c.

Looking at the test output for a ZFS test run with USE_OFD=yes, it seems that there is some simple check needed in order to disable quota testing with OFD:

11:48:38:CMD: client-19vm4 lctl set_param lquota.lustre-OST*.stats=0
11:48:38:client-19vm4: error: set_param: /proc/{fs,sys}/{lnet,lustre}/lquota/lustre-OST*/stats: Found no match

I suspect the same will be true for ldiskfs-backed OFD tests as well.

This needs to be the case for USE_OFD=yes until the orion quota project has finished landing.

Comments

[Andreas Dilger]

Needs a relatively simple fix to the test that can hopefully still be included into b2_3.

[Johann Lombardi (Inactive)]

It is just an issue with the statistics which are not present. I think that s-q can be fixed easily.
In fact, we are supposed to run the acct tests only when ofd is in use:

only running test 33 34 35

[Minh Diep]

Johann,

could you take a look at this report https://maloo.whamcloud.com/test_sets/2d46e18e-fdb3-11e1-afe5-52540035b04c and let me know if we still only run test 33 - 35 when USE_OFD=yes with zfs?

[Johann Lombardi (Inactive)]

Well, autotest is definitely running the whole suite However, it wasn't the initial intent:

if [ "$USE_OFD" = "yes" -a -z "$ONLY" ]; then
        # only accounting tests are supported with OFD for the time being
        ONLY="33 34 35"
fi

I assume that autotest uses ONLY
Minh, is this something you can try to fix? Thanks in advance

[Minh Diep]

Johann,

sorry for not being clear. I ran this manually after removed the ONLY. I wanted to see if we can run all the tests with a little fix. If you think we only require to run 33-35. I can do that. Please let me know.

[Johann Lombardi (Inactive)]

Only accounting is supported for now, so only 33-35 should be run. Other tests can't pass for now.

[Minh Diep]

https://maloo.whamcloud.com/test_sets/b8a16670-fe7e-11e1-a707-52540035b04c
shows 33 and 34 failed. Could you please check it?

[Johann Lombardi (Inactive)]

Minh, it seems that the filesystem wasn't clean and the quota_user already had some existing files which confused the test.

[Minh Diep]

Johann, Something is wrong here. the fs started clean and quota still fail after delete

https://maloo.whamcloud.com/test_sessions/99e4c32c-0080-11e2-860a-52540035b04c

[Johann Lombardi (Inactive)]

Minh, was this on master or b2_3?
I think b2_3 misses http://review.whamcloud.com/3470, although i don't think this can explain the issue of space not released from accounting after delete.
In any case, i think we should rebase the quota accounting tests on the ones in orion. Niu, could you please handle that?

Thanks in advance.

[Minh Diep]

Assign this to Niu to investigate further - thanks

[Niu Yawei (Inactive)]

I think it might because the test_33() doesn't sync_all_data for zfs after the deletion.

Hi, Minh
Could you add a "sync; sync_all_data;" after the "wait_delete_completed" in test_33() to see if it can resolve your problem? (Sorry, I don't have zfs test environment by hand). Thanks.

For the problem of "error: set_param: /proc/

{fs,sys}

/

{lnet,lustre}

/lquota/lustre-OST*/stats: Found no match", I think we need to check if USE_OFD is set in
run_test_with_stat() and skip checking the quota stats when USE_OFD is set.

[Niu Yawei (Inactive)]

Minh, USE_OFD is obsolete now and test_33() in master do have syc_all_data after deletion. Can we close it now? Thanks.

[Andreas Dilger]

This is no longer needed.