[LU-3828] Default HSM Archive Identifier is not Persistent

• Status: Resolved
• Project: Lustre
• Affects Version/s: Lustre 2.5.0
• Fix Version/s: Lustre 2.5.0
• Type: Technical task
• Priority: Blocker
• Reporter: Malcolm Cowe (Inactive)
• Assignee: John Hammond
• Resolution: Fixed
• Labels: HSM

Description

The procfs tuneable parameter that is used to define the default HSM archive ID is cannot be set persistently using lctl conf_param or lctl set_param -P. This is by design and is in part because the "hsm" procfs subtree is not created unless and until the HSM feature is itself activated.

The parameter variable is: mdt.<fsname>-MDT*.hsm.archive_id

Please consider a mechanism whereby the system-wide default for the archive ID is stored persistently by the MGS, as it simplifies administration overhead and makes other features such as server failover behave more predictably.

Comments

[Malcolm Cowe (Inactive)]

Original test case:

If I try to change the default archive ID using lctl conf_param instead of lctl set_param, the setting does not change.
Tested as follows:

On mount, default archive is set to 1 and archive commands work as expected (when no archive is specified, default is selected automatically).

Using lctl set_param to change the default works as expected and lfs hsm_archive command picks up the change.

Using lctl hsm_archive with --archive works as expected.

lctl conf_param does not work:

[root@m64-1 /]# lctl set_param mdt.scratch-MDT*.hsm.archive_id=2
mdt.scratch-MDT0000.hsm.archive_id=2
[root@m64-1 /]# lctl get_param mdt.scratch-MDT*.hsm.archive_id
mdt.scratch-MDT0000.hsm.archive_id=2
[root@m64-1 /]# lctl conf_param scratch-MDT*.mdt.hsm.archive_id=1
[root@m64-1 /]# lctl get_param mdt.scratch-MDT*.hsm.archive_id
mdt.scratch-MDT0000.hsm.archive_id=2

I am assuming that this is the path one should use for conf_param. If I use conf_param the setting does not change.

[jacques-charles lafoucriere]

This behavior is also true for all HSM tunable (except hsm_control). A solution is to make the proc hsm directory always visible even if HSM is off. If you agree on it I will work on a patch.

[Malcolm Cowe (Inactive)]

I agree that the feature is useful and would like to see it implemented. If making the hsm procfs directory visible is a straightforward fix, then it has my support but someone else has to make the call.

[jacques-charles lafoucriere]

It is not a "straightforward fix", but the simplest I see. I will work on a patch. Can you validate the other patch, so I will base this new one on the other one (will simplify the review)

[jacques-charles lafoucriere]

Patch at http://review.whamcloud.com/7557

[Peter Jones]

Landed for 2.5.0