[LU-1436] Test failure on test suite sanity, subtest test_24w

• Type: Bug
• Resolution: Fixed
• Priority: Blocker
• Fix Version/s: Lustre 2.3.0
• Affects Version/s: Lustre 2.3.0
• Severity: 3
• Assignee: Hongchao Zhang
• Reporter: Maloo

Description

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

This issue relates to the following test suite run: https://maloo.whamcloud.com/test_sets/78395658-a4b3-11e1-adce-52540035b04c.

The sub-test test_24w failed with the following error:

== sanity test 24w: Reading a file larger than 4Gb =================================================== 14:31:53 (1337635913)
1+0 records in
1+0 records out
1048576 bytes (1.0 MB) copied, 0.0721251 seconds, 14.5 MB/s
1+0 records in
1+0 records out
234852 bytes (235 kB) copied, 0.0710902 seconds, 3.3 MB/s
0+0 records in
0+0 records out
0 bytes (0 B) copied, 0.00325134 seconds, 0.0 kB/s
sanity test_24w: @@@@@@ FAIL: Error reading at the end of the file f24w

Info required for matching: sanity 24w

Issue Links

• duplicates: LU-1437 Test failure on test suite sanity, subtest test_34c
• is related to: LU-969 2.1 client stack overruns
• is related to: LU-4357 page allocation failure. mode:0x40 caused by missing __GFP_WAIT flag

Comments

[Peter Jones]

Landed for 2.3

[Hongchao Zhang]

status update:
the tests against the patch has passed and is still under inspection(currently one +)

[Alex Zhuravlev]

could someone check lustre builds with --disable-liblustre please ?

[Hongchao Zhang]

the patch is tracked at http://review.whamcloud.com/#change,3072

[Andreas Dilger]

Marking for a blocker on 2.1.2 as well.

[Peter Jones]

Hongchao

Could you please look into this one?

Thanks

Peter

[Andreas Dilger]

Unfortunately, it was my proposal that caused this problem. I'm surprised that it only caused so few issues during testing.

The patch is fairly straight forward - change libcfs_log_return() to return long long. Even with sign extension, the original value should be OK after casting the value down to the original type again.

That must mean the same problem also exists for 32-bit clients on master. Are we no longer doing any 32-bit testing on master? I thought we would drop i686 server and RHEL5 server build/test, but I thought we would keep RHEL5/i686 client testing going to ensure we have at least some 32-bit test coverage.

[Jian Yu]

The problem here was introduced by LU-969 http://review.whamcloud.com/2668

Yes. After it was committed on 11 May, the sanity tests 24w,34

{c,d,g}

, started failing on i686 client. Here are the search results of the historical Maloo reports:

test 24w: http://tinyurl.com/cf9uvj2
test 34c: http://tinyurl.com/clzdw2k

[Oleg Drokin]

The problem here was introduced by LU-969 http://review.whamcloud.com/2668

+long libcfs_log_return(struct libcfs_debug_msg_data *, long rc);

This truncates all return values passed through RETURN() macro to long, which is only 32 bit on i686.

[Andreas Dilger]

I would say that this is a blocker for the 2.1.2 release, and is likely caused by the same problem as LU-1437.

Can you check some of the earlier builds to see when it was introduced? Does it fail repeatedly, or is it intermittent?