[LU-14912] client picking other checksum type over T10PI

• Type: Improvement
• Resolution: Fixed
• Priority: Minor
• Fix Version/s: Lustre 2.15.0
• Assignee: Dongyang Li
• Reporter: Dongyang Li

Description

When server indicates it supports T10 checksums, the client could still pick other check types over T10PI as the other types getting better numbers during client benchmark.

Issue Links

• is related to: LU-11011 checksum type can not be selected permanently
• is related to: LU-14924 LustreError: 133-1: nbp17-OST0064-osc-ffff9bf24bc99800: BAD READ CHECKSUM
• is related to: LU-10472 Data Integrity(T10PI) support for Lustre
• is related to: LU-14895 dump T10 guard tags on checksum error and flush pages

Comments

[Peter Jones]

Landed for 2.15

[Gerrit Updater]

"Oleg Drokin <[e-mail]>" merged in patch https://review.whamcloud.com/44657/
Subject: LU-14912 obdclass: prefer T10 checksum if the target supports it
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: 5e9059e08aec6fb36de6788465978de0e962b956

[Gerrit Updater]

"Li Dongyang <[e-mail]>" uploaded a new patch: https://review.whamcloud.com/44657
Subject: LU-14912 obdclass: prefer T10 checksum if the target supports it
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 0c6d2cf5b4cc449253965cbd527e6a2f6ef0ee3e

[Andreas Dilger]

T10-PI checksums were only added in 2.12, so I don't think it is reasonable for the server to return only a single t10crc4K checksum type, because this would break interoperability for 2.10.x clients.

One option for selecting the proper T10-PI checksum type automatically on the client, is to detect if only a single OBD_CKSUM_T10* type is returned to the client, then this indicates the server has T10-PI enabled (otherwise it would return multiple OBD_CKSUM_T10* types), and it should prefer this over other checksum types.

For example, on my test VM (no T10-PI support in the storage) it reports multiple t10* checksum types enabled:

# lctl get_param osc.*.checksum_type
osc.testfs-OST0000-osc-ffff9ac928b3d800.checksum_type=
crc32 adler crc32c t10ip512 [t10ip4K] t10crc512 t10crc4K 

while on a client where T10-PI hardware support is enabled on the server the client reports:

# lctl get_param osc.*.checksum_type
osc.testfs-OST0000-osc-ffff9ac928b3d800.checksum_type=
crc32 adler [crc32c] t10crc4K

In this case, the client should prefer the single t10crc4K checksum type (if no type was specified) that matches the server hardware checksum type, even though crc32c may be somewhat faster.