[LU-7134] Ensure ZFS hostid protection if servicenode/failover options given to mkfs.lustre

• Type: Improvement
• Resolution: Fixed
• Priority: Minor
• Fix Version/s: Lustre 2.9.0
• Labels: zfs
• Assignee: Nathaniel Clark
• Reporter: Nathaniel Clark

Description

SPL import protection only exists if hostid is set to a non-zero value. If servicenode or failover is given as an option, then mkfs.lustre should at least warn that hostid is set to zero and import protection won't work.

Proposed solution:

• Have mkfs.lustre fail if --servicenode / --failover is specified and hostid is zero (will check spl_hostid value).
• Add override option --force-nohostid to ignore this case

Issue Links

• is related to: LU-7797 Can't mount zpools after OSS restart
• is related to: LU-8816 test-framework needs to reload SPL module after set hostid
• is related to: LU-3163 conf-sanity test_61 reformats MDT but not OST
• is related to: LU-6889 new kernel [SLES11 SP4 3.0.101-65]
• links to: ZFS Issue #745: Multi mount support

Comments

[Peter Jones]

Landed for 2.9

[Gerrit Updater]

Oleg Drokin ([e-mail]) merged in patch http://review.whamcloud.com/16611/
Subject: LU-7134 utils: Ensure hostid set for ZFS during mkfs
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: 7b654fc673a2570ab4a828c2a7b213cee8b5b0cc

[Nathaniel Clark]

Upon closer reading of the code in ZFS, I'm not sure enabling this will help in a fail-over scenario or provide adequate protection.

The code compares the host's hostid to the one written to the zpool and then prevents import if they don't match (unless one of the hostids is 0). This isn't true multi-mount protection. A more robust solution tracked in issue ZFS #745.

[Gerrit Updater]

Nathaniel Clark ([e-mail]) uploaded a new patch: http://review.whamcloud.com/16611
Subject: LU-7134 utils: Ensure MMP support in SPL/ZFS during mkfs
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 15b26271b2ce4ccd87b424a75765edbf9f5f82eb

[Nathaniel Clark]

SPL supports setting HostID in two ways:

option spl spl_hostid to manually specify the HostID to use
option spl spl_hostid_path to specify an alternate HostID file to use (default: /etc/hostid)

What if I check the spl hostid /sys/module/spl/parameters/spl_hostid to ensure it's non-zero? It show the HostID is is using (even when read from /etc/hostid)