[LU-8135] sanity test_101g fails with 'not all RPCs are 16 MiB BRW rpcs'

• Type: Bug
• Resolution: Fixed
• Priority: Minor
• Fix Version/s: Lustre 2.9.0
• Affects Version/s: Lustre 2.9.0
• Environment: autotest review-zfs
• Severity: 3
• Assignee: Jinshan Xiong (Inactive)
• Reporter: James Nunez (Inactive)

Description

sanity test 101g fails with

'not all RPCs are 16 MiB BRW rpcs' 

In the test log on the client, we can see the ‘File too large’ error returned when dd tries to write to the file system

== sanity test 101g: Big bulk(4/16 MiB) readahead ==================================================== 11:11:42 (1462903902)
CMD: onyx-50vm8 /usr/sbin/lctl get_param obdfilter.*.brw_size |
			while read s; do echo ost1 \$s; done
CMD: onyx-50vm8 /usr/sbin/lctl get_param obdfilter.*.brw_size |
			while read s; do echo ost2 \$s; done
CMD: onyx-50vm8 /usr/sbin/lctl set_param -n obdfilter.lustre-OST*.brw_size=16M 		osd-*.lustre-OST*.brw_size=16M 2>&1
remount client to enable large RPC size
CMD: onyx-50vm1.onyx.hpdd.intel.com grep -c /mnt/lustre' ' /proc/mounts
Stopping client onyx-50vm1.onyx.hpdd.intel.com /mnt/lustre (opts:)
CMD: onyx-50vm1.onyx.hpdd.intel.com lsof -t /mnt/lustre
CMD: onyx-50vm1.onyx.hpdd.intel.com umount  /mnt/lustre 2>&1
Starting client: onyx-50vm1.onyx.hpdd.intel.com:  -o user_xattr,flock onyx-50vm7@tcp:/lustre /mnt/lustre
CMD: onyx-50vm1.onyx.hpdd.intel.com mkdir -p /mnt/lustre
CMD: onyx-50vm1.onyx.hpdd.intel.com mount -t lustre -o user_xattr,flock onyx-50vm7@tcp:/lustre /mnt/lustre
dd: error writing '/mnt/lustre/f101g.sanity': File too large
2+0 records in
1+0 records out
16777216 bytes (17 MB) copied, 0.978339 s, 17.1 MB/s
0+0 records in
0+0 records out
0 bytes (0 B) copied, 0.00599718 s, 0.0 kB/s
0 RPCs
 sanity test_101g: @@@@@@ FAIL: not all RPCs are 16 MiB BRW rpcs 

We’ve had three occurrences of this failure since the test was introduced by http://review.whamcloud.com/19368 a little over a week ago and, so far, only seen on review-zfs-part-1:

2016-05-09 - https://testing.hpdd.intel.com/test_sets/145be8f2-1661-11e6-b5f1-5254006e85c2
2016-05-10 - https://testing.hpdd.intel.com/test_sets/f2b9b50c-1701-11e6-b5f1-5254006e85c2
2016-05-11 - https://testing.hpdd.intel.com/test_sets/73c7a9a4-179f-11e6-b5f1-5254006e85c2

Issue Links

• is related to: LU-10239 Lustre crash (client): The first extent to be fit in a RPC contains 17 chunks, which is over the limit 16.
• is related to: LU-8632 Address of function 'page_count' is used instead of a local variable
• is related to: LU-7181 Submitting random writes using 4MB RPC
• is related to: LU-8139 sanity test_101g: 'max_pages_per_rpc not correctly set'