[LU-16053] osd-zfs build system update

• Status: Open
• Project: Lustre
• Fix Version/s: Upstream
• Type: Improvement
• Priority: Minor
• Reporter: Brian Behlendorf
• Assignee: Brian Behlendorf
• Resolution: Unresolved
• Labels: build, llnl, zfs
• Epic/Theme: zfs

Description

While reviewing the OpenZFS kernel interface checks I noticed that there were some improvements which could be made.  I propose making the following changes (patches to follow).

1. Enforce a minimum ZFS version of 0.7.0 (released in July 2017).  This version provides the required interfaces for all major features expected by Lustre (quotas, multihost), and dropping support for older versions let's us remove some dead code.  Setting a minimum version also allows us to fail cleanly at configure time if a required interface isn't available.
2. Update the lustre-build-zfs.m4 checks to use the parallel build functionality from LU-13485.  This reduces the ZFS configure check time to just a few seconds.
3. Update all the checks to have consistent output and be a little more strict to catch as much as possible during configure.

Issue Links

• is related to: LU-13485 Enable parallel compile tests during ...

Comments

[Brian Behlendorf]

"Brian Behlendorf <[e-mail]>" uploaded a new patch: https://review.whamcloud.com/48068
Subject: LU-16035 build: osd-zfs use parallel configure macros
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 9abee6ba2e88b9dbf966cdf115046dca4e212569

[Andreas Dilger]

Hmm, strange that Gerrit didn't automatically add comments here for all of the patches... There was a system outage yesterday, so that might be the reason.

[Brian Behlendorf]

That was actually my fault, I accidentally used the wrong LU number when I first pushed the patches.  I pushed a second version to correct it, but Gerrit linked them to 16035 instead of 16053.  If possible you might want to just delete those Gerrit comments in the other issue to avoid confusion.

[Gerrit Updater]

"Brian Behlendorf <[e-mail]>" uploaded a new patch: https://review.whamcloud.com/48076
Subject: LU-16053 build: add project quota configure check
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 73e002b407edb1abd6e98012c06300a460ea768c

[Gerrit Updater]

"Brian Behlendorf <[e-mail]>" uploaded a new patch: https://review.whamcloud.com/48077
Subject: LU-16053 build: remove obsolete ZFS configure checks
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: b0c2b5dcb7d3a275d00820f8d3bb3b30075fafeb

[Gerrit Updater]

"Shaun Tancheff <[e-mail]>" uploaded a new patch: https://review.whamcloud.com/c/fs/lustre-release/+/51082
Subject: LU-16053 zfs: fix dmu_offset_next() check
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 73c7c72204da1618b0d90d86d66d07be9e227535

[Shaun Tancheff]

Splitting the configure fix for dmu_offset_next to a separate patch because the associated sanity test for SEEK_HOLE and SEEK_DATA are failing for ZFS.

My testing suggests that dmu_offset_next() never returns EBUSY. I updated SEEK_HOLE and SEEK_DATA operations to explicitly sync before attempting dmu_offset_next() which improves the situation but there is still issues. The server side sync before lseek is not sufficient for reliable operation.

Writing a new data segment to a sparse file also fails in most cases.

[Gerrit Updater]

"Oleg Drokin <[e-mail]>" merged in patch https://review.whamcloud.com/c/fs/lustre-release/+/48089/
Subject: LU-16053 build: Update zfs configure checks
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: 46938c53461d136b71a32c4951b1776e2d226648