[LU-10556] lustre client rebuild not building lnetctl

• Type: Bug
• Resolution: Fixed
• Priority: Minor
• Fix Version/s: Lustre 2.11.0
• Affects Version/s: Lustre 2.10.2
• Environment: OS: SLF 6.8; Kernel: 2.6.32-642.15.1.el6.x86_64
• Severity: 3
• Assignee: James A Simmons
• Reporter: Alex Kulyavtsev

Description

Similar to LU-7887.

lnetctl not present in lustre-client rpm rpm rebuild from source. lnetctl is present in prebuilt rpm.

It will be extremely helpful to have "rpmbuild --rebuild" to build lustre client rpm with the same content and functionality as lustre-client rpm from distro.

The lustre client rpm rebuild is required time to time through lifecycle of the cluster. It shall be possible to do this type of work by administrators without knowledge of lustre internals.
There are several mails on lustre-discuss list reporting lnetctl is not in rpm.

I had to explicitly use flags as listed below to get lnetctl:
rpmbuild --rebuild --without servers --with lnet-dlc --with lustre-utils ./lustre-2.10.2-1.src.rpm

I did not try rpm rebuild for SLF 7.4 as distro version worked for me.

The other aspect, we have to document rpm rebuild procedure.
There are clear and detailed instructions on client rebuild on Cray website for their distro. I can not easily find "rebuild" in lustre manual linked from lustre.org.
There are detailed instructions on lustre wiki; they have to be updated with lnet-dlc lustre-utils until rpm build fixed.

Issue Links

• is related to: LU-10858 lustre-initialization-1 lustre-initialization fails for SLES12 SP2 and SP3
• is related to: LU-10003 lnetctl error "cannot add network: invalid argument"
• is related to: LU-9897 Tracking of lustre library improvements.

Comments

[Gerrit Updater]

James Nunez ([e-mail]) uploaded a new patch: https://review.whamcloud.com/31800
Subject: Revert "LU-10556 build: add require libyaml and zlib"
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 8c206fdb6a2deae0cf60c4da4900518dad8360d4

[Peter Jones]

Landed for 2.11

[Gerrit Updater]

Oleg Drokin ([e-mail]) merged in patch https://review.whamcloud.com/31710/
Subject: LU-10556 build: add require libyaml and zlib
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: ab3185eb2b54ed2f616022956dca9bd5a849eb80

[James A Simmons]

Alex I have a patch for the gss issues : https://review.whamcloud.com/#/c/3175. That is tracked under LU-10752

[Alex Kulyavtsev]

There is unrelated gss issue on SLF 7 node (it can be also some rpms are installed/missing).

I had to add flags on SLF7 node to resolve error:

--define 'configure_args --disable-gss --disable-gss-keyring'

File not found: /root/rpmbuild/BUILDROOT/lustre-2.11.0_RC1_1_gac4da53-1.x86_64/etc/request-key.d/lgssc.conf

 

Built completed OK on SLF 6.8 node.

 

I guess gss is disabled by default in lustre.spec.in . If by some reason gss is enabled lustre build shall provide default on build host or do not complain.

I read lgssc.conf is not used on client host until gss is configured:

LUDOC-197:

• * is the /etc/request-key.d/lgssc.conf file included and installed with the Lustre RPMs? It seems that would be OK, since it would never be used unless there is a security flavor for a connection. The fewer things that users need to configure, the better.

 

 

[Alex Kulyavtsev]

I confirm /usr/sbin/lnetctl present in lustre-client rpm built from  lustre-2.11.0_RC1_1_gac4da53-1.src.rpm

on SLF 7.4:  3.10.0-693.11.1.el7.x86_64

and SLF 6.8: 2.6.32-642.15.1.el6.x86_64

 

I grabbed source rpm from build server (artifact of build triggered by patch 31710)

wget http://build.lustre.org/downloads/31710/1/centos/7.2/SRPM/lustre-2.11.0_RC1_1_gac4da53-1.src.rpm

DIR=../downloads/build.lustre.org

SRPM=lustre-2.11.0_RC1_1_gac4da53-1.src.rpm 

rpmbuild  --rebuild --without servers --define 'configure_args --disable-gss --disable-gss-keyring' $DIR/$SRPM

 

[Gerrit Updater]

Minh Diep ([e-mail]) uploaded a new patch: https://review.whamcloud.com/31710
Subject: LU-10556 build: add require libyaml and zlib
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: ac4da53c0f66b2f77b9f502bac6f4b26a45f2ceb

[James A Simmons]

Yep. I missed adding that dependency. BTW while working with the LU-9897 patch I noticed Debian/Ubuntu have no dependencies in its build files!!!! I was waiting for Martin's Ubuntu server patches to land first before I fix that.

[Andreas Dilger]

If we are marking the lctl versions of the commands deprecated in 2.11 (per LU-10003) it makes sense to have enable_dlc set by default.

Alex, if you already have a patch for this, could you please submit it to Gerrit so we can land that for 2.11 and backport it for a 2.10 maintenance release.

James, I see that in LU-9897 there is a change to the lustre.spec.in file that removes the conditional build of liblnetconfig.so but it doesn't add libyaml or yaml-devel (or whatever) package to the Build-requires: line.  That should probably be added.

[Alex Kulyavtsev]

Yes, both libyaml libyaml-devel installed - I learned that during the server build.
config log explicitly shows that "configure" got libyaml* during rpm build.

I rebuilt rpm with / without disable-dlc flag and it made the difference so it is not related to host configuration (with libyaml present).

config.status shows for the plain rebuild "--without servers" and without explicitly setting flags lnet-dlc that enable_dlc is set no (disabled) by default and this prevents building lnetctl.

aclocal.m4 :

# LN_CONFIG_DLC
#
# Configure dlc if enabled
#
# if libyaml is set (IE libyaml installed) and enable_dlc = yes then build
# dlc other wise (IE if libyaml is not set or enable_dlc = no) then don't
# build dlc.

Because distro rpm has dlc enabled, we shall have set enable_dlc = yes by default (at least when libyaml installed) and explicitly disable dlc when it is not needed and/or libyaml not installed.

At this point I'm OK with workaround;

It is important to have client rpm to be rebuild with the same configuration as rpm in distro without unpacking and digging into source rpm.