[LU-2428] ost-pools 18: Degradation with wide pool is 43% > 30%

• Type: Bug
• Resolution: Won't Fix
• Priority: Blocker
• Affects Version/s: Lustre 2.4.0, Lustre 2.4.1, Lustre 2.5.0
• Labels: yuc2
• Severity: 3
• Assignee: Alex Zhuravlev
• Reporter: Li Wei (Inactive)

Description

https://maloo.whamcloud.com/test_sets/cedea7ac-e715-11e1-ac43-52540035b04c

Comments

[Jian Yu]

More instances on Lustre b2_4 branch:
https://maloo.whamcloud.com/test_sets/64f55a14-fd78-11e2-9fdb-52540035b04c
https://maloo.whamcloud.com/test_sets/e8d89476-ce1c-11e2-96ef-52540035b04c
https://maloo.whamcloud.com/test_sets/3f122714-c60a-11e2-9bf1-52540035b04c

[Sarah Liu]

Hit this issue on lustre-master build #1560 zfs
https://maloo.whamcloud.com/test_sets/021e1e98-ed43-11e2-99b4-52540035b04c

[Sarah Liu]

Also seen in lustre-2_4RC2 RHEL6 client:
https://maloo.whamcloud.com/test_sets/3f122714-c60a-11e2-9bf1-52540035b04c

[Sarah Liu]

Hit this issue when testing 2.4.50-RC1 on SLES11 SP2 client:
https://maloo.whamcloud.com/test_sets/100d9812-bf30-11e2-88e0-52540035b04c

[Sarah Liu]

another failure seen in lustre-master with ldiskfs

https://maloo.whamcloud.com/test_sets/59aff5bc-3d73-11e2-b377-52540035b04c

client-27vm3: Pool lustre.testpool destroyed
CMD: client-27vm6.lab.whamcloud.com lctl get_param -n lov.lustre-*.pools.testpool       2>/dev/null || echo foo
total: 7059 creates in 14.63 seconds: 482.45 creates/second
iter 3: 7059 creates with missing pool

Avg files created in 15 seconds without pool: 3493
Avg files created in 15 seconds with pool: 2673
Avg files created in 15 seconds missing pool: 2353
No pool / wide pool: 23 %.
No pool / missing pool: 32 %.
 ost-pools test_18: @@@@@@ IGNORE (bz23408): Degradation with wide pool is 32% > 30% 

[Ian Colle (Inactive)]

https://maloo.whamcloud.com/test_sets/cf495660-fcc3-11e1-b09c-52540035b04c

[Andreas Dilger]

This test has failed rarely on master as in the past, which is why it is marked "error_ignore".

That said, it does appear that this is a steady regression on orion. I don't think this is a blocker for orion_head_sync, but we'll need to understand why the OST pools performance is significantly impacted.

[Li Wei (Inactive)]

https://maloo.whamcloud.com/test_sets/49bb7fa8-ecad-11e1-ba25-52540035b04c