[LU-10460] Evaluate two LLNL patches for upstream

• Type: Task
• Resolution: Fixed
• Priority: Minor
• Fix Version/s: Lustre 2.11.0, Lustre 2.10.4
• Assignee: Peter Jones
• Reporter: Olaf Faaland

Description

In LU-10392, a list of local LLNL patches applied to 2.10.2 during our testing came up, and Peter requested we consider pushing them upstream.

https://jira.hpdd.intel.com/browse/LU-10392?focusedCommentId=216657&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-216657

This ticket is to track that.  The two patches are:

* 7823f77 LU-4009 osd-zfs: Add tunables to disable sync (DEBUG)
* 1826c76 Don't install lustre init script on systemd systems
(AKA https://review.whamcloud.com/#/c/7761/)

Comments

[Peter Jones]

IIUC we do not intend to land the systemd patch. If you agree can you abandon the patch and we'll close out this ticket now that the sync patch has landed.

[Gerrit Updater]

Oleg Drokin ([e-mail]) merged in patch https://review.whamcloud.com/7761/
Subject: LU-10460 osd-zfs: Add tunables to disable sync
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: 00c4bd85b372f68c3e75fdab9658c8f0074113be

[Gerrit Updater]

Olaf Faaland-LLNL ([e-mail]) uploaded a new patch: https://review.whamcloud.com/30830
Subject: LU-10460 rpm: Skip packaging lustre init for systemd
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 236ff56515c4263a3af61d316644f44f4a553b20

[Andreas Dilger]

I can't find systemd patch in our Gerrit repo, so it would be useful to get a version pushed for review, with a description of what problem it is solving. These patches were added for a reason, so we need to understand why they are harmful just being installed on systemd systems, and if we can somehow make them conditional/deactivated when HA agents are involved.

[Peter Jones]

Olaf

We discussed this and agree with your assessment that the lustre-init scripts change is more suitable to keep as an LLNL-only change but we will move forward to land adding https://review.whamcloud.com/#/c/7761/ using this tracking reference.

Peter

[Olaf Faaland]

Let's discuss these one at a time.

I propose the patch re: lustre init scripts is not appropriate for upstream.

* 1826c76 Don't install lustre init script on systemd systems

All that patch does, is avoid including the following files in the rpm if with_systemd is not set:

• lustre/scripts/lustre
• lustre/conf/lustre

This works for us, because we use pacemaker and a resource agent to start, stop, and query the status of lustre targets on our servers. The resource agent contains the start/stop/query logic and pacemaker keeps track of what is running where.

However this seems inappropriate for the general user running under systemd, who might be merely testing lustre, or might be using diskful nodes without shared storage, where ordinary systemctl commands seem like the right mechanism for managing these functions.

We don't need this ourselves, though, and so don't have plans to allocate time to work out the right service definitions.