[LU-10627] tests/utils: lhsmtool_posix now runs lt-lhsmtool_posix and breaks tests

• Type: Bug
• Resolution: Fixed
• Priority: Minor
• Fix Version/s: Lustre 2.12.0
• Labels: HSM patch test
• Severity: 3
• Assignee: James A Simmons
• Reporter: CEA

Description

When running tests locally lhsmtool_posix is a bash script that internally runs lustre/utils/.libs/lt-lhsmtool_posix. I assume this is due to the use of libtool in the compilation process.

Running pgrep "lhsmtool_posix" now yields no pid since the actual process is named "lt-lhsmtool_posix".
This breaks quite a lot of tests in sanity-hsm.

Issue Links

• is related to: LU-11742 ERROR: RPATH is not allowed
• is related to: LU-10514 all metadata operations take 1+ minutes thanks to libtool's l_getidentity
• is related to: LU-11039 Wrong HSMTOOL_BASE name when kill copytools

Comments

[Gerrit Updater]

Oleg Drokin ([e-mail]) merged in patch https://review.whamcloud.com/32835/
Subject: LU-10627 tests: don't use libtool wrapper for applications
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: fd472f9427931092bab8d676798bc3f9744a552b

[Gerrit Updater]

James Simmons ([e-mail]) uploaded a new patch: https://review.whamcloud.com/32835
Subject: LU-10627 tests: don't use libtool wrapper for applications
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 3e6fcf41eb85af32899fefcccc303b07eb61c111

[James A Simmons]

Okay I looked into this problem and found the solution. If you add the following it disables creating libtool executable wrappers.
prog_LDFLAGS = -no-install

[James A Simmons]

Now that I'm testing Ubuntu 18 on my local system I will be testing lustre without installing it so I will look to fix this. The easy way to fix this is change $LUSTRE/utils/ to $LUSTRE/utils/.libs but I'm also looking at using env to properly set the PATH variable so the right application would be used.

[Quentin Bouget (Inactive)]

Hi John,

Your workaround works for me, thanks!

Hi James,

I tried what you suggested (before using ./configure --disable-shared) but it did not work. The main issue is that sanity-hsm uses killall "lhsmtool_posix" to kill copytools (unlike pkill, killall only affects processes whose name fully matches its arguments). Since internally, the libtool-generated lhsmtool_posix calls exec lt-lhsmtool_posix, killall fails to find any matching process.

[James A Simmons]

I have an idea. Can you try and change 

export LD_LIBRARY_PATH=${LUSTRE}/utils:${LD_LIBRARY_PATH}

in test-framework.sh to

export LD_LIBRARY_PATH=${LUSTRE}/utils/.libs:${LD_LIBRARY_PATH}

Let me know if that works.

[John Hammond]

Hi Quentin,

Yes this is due to libtoolization of lustre/utils/ done by https://review.whamcloud.com/30562 LU-5541 build: move libcfs and liblustreapi over to libtool.

I have been using ./configure --disable-shared as a workaround for this issue when I am doing local testing. There may be a better fix for this but I do not know it.