Package Details: slurm-llnl 22.05.2.1-1

Git Clone URL: https://aur.archlinux.org/slurm-llnl.git (read-only, click to copy)
Package Base: slurm-llnl
Description: Simple Linux Utility for Resource Management
Upstream URL: https://www.schedmd.com
Keywords: manager resource system
Licenses: GPL
Submitter: tegenf
Maintainer: edh
Last Packager: edh
Votes: 17
Popularity: 1.20
First Submitted: 2011-01-28 14:47 (UTC)
Last Updated: 2022-06-17 14:23 (UTC)

Dependencies (22)

Required by (1)

Sources (3)

Latest Comments

edh commented on 2022-06-16 16:37 (UTC)

Starting with version 22 Slurm requires an /etc/slurm-llnl/cgroup.conf configuration file if the config switch ProctrackType=proctrack/cgroup is set. See https://slurm.schedmd.com/cgroup.conf.html for further details. With the next slurm release I will add an example file at /etc/slurm-llnl/cgroup.conf.example.

edh commented on 2021-08-06 10:08 (UTC)

@leuko

/var/log/slurm-llnl should be part of the package! It is created in package(). However, as you said, it does not need to be slurm's home directory.

I'll remove the slurm-default file and replace it with simple example files for slurmd, slurmctld and slurmdbd in /etc/sysconfig/.

leuko commented on 2021-08-06 09:45 (UTC)

Another issue: PKGBUILD installs slurm-llnl-default-conf to /etc/default/slurm-llnl, but the three service files use /etc/sysconfig.

Even more packages seem to use /etc/default, maybe it is less work to keep to upstream rather than patching or convincing upstream:

$ pacman -F '/etc/default/' | wc -l
22
$ pacman -F '/etc/sysconfig/' | wc -l
3

leuko commented on 2021-08-06 09:06 (UTC) (edited on 2021-08-06 09:07 (UTC) by leuko)

@edh :)

I also noticed that shadow.service fails with the following message:

user 'slurm': directory '/var/log/slurm-llnl' does not exist

slurm-llnl.sysusers does not create /var/log/slurm-llnl, and slurm-llnl.tmpfiles either. Z flag only updates the permissions AFAIK.

Slurm uses syslog as default. A solution would be to omit (leave empty) home-directory and shell. man sysusers.d:

It is recommended to omit this, unless software strictly requires a home directory to be set

It is recommended to omit this, unless a shell different /usr/sbin/nologin must be used

edh commented on 2021-08-05 12:19 (UTC)

@leuko Sure, that sounds like a good proposal! I will update the tmpfiles with the next release.

@leuko Thanks for the comment! Let's hope for a new release soon.

leuko commented on 2021-08-05 11:07 (UTC) (edited on 2021-08-05 11:08 (UTC) by leuko)

If slurmctld and slurmd do not start after boot, add network-online.target to the service files (instead of using `network.target). Otherwise you may get:

error: get_addr_info: getaddrinfo() failed: Name or service not known
slurmctld[...]: error: slurm_set_addr: Unable to resolve "(null)"

It is fixed in the commit for the next release.

leuko commented on 2021-08-05 09:58 (UTC)

slurm-llnl.tmpfiles creates the following state files:

d /var/spool/slurm 0755 slurm slurm -
d /var/spool/slurm/slurmctld 0755 slurm slurm -
d /var/spool/slurm/slurmd 0755 root root -
...

The default settings for these state files, i.e. when StateSaveLocation and SlurmdSpoolDir are not specified, then /var/spool directory is used. What do you think about changing slurm-llnl.tmpfiles to:

d /var/spool/slurmctld 0755 slurm slurm -
d /var/spool/slurmd 0755 root root -
...

?

This would be a more minimalist approach.

lahwaacz commented on 2021-04-13 07:26 (UTC)

@edh Hmm, that explains why kill -HUP does not apply all config changes. Thanks again for making the report!

edh commented on 2021-04-12 18:36 (UTC)

@lahwaacz This seems to be a little bit more tricky than anticipated. In #11339 Tim Wickberg explains that even though kill -HUP might no be perfect it is currently the best they can do as scontrol reconfigure sends a reconfiguration command to the whole cluster and is thus not local.

edh commented on 2021-04-10 08:57 (UTC)

@lahwaacz done #11339.

lahwaacz commented on 2021-04-10 08:22 (UTC)

@edh If you don't mind, that would be nice. I'm really out of ideas how to proceed... Thanks!

edh commented on 2021-04-10 08:07 (UTC)

@lahwaacz No, unfortunately not. The described behavior though is odd. The registriation worked just fine for me. I guess with this one person you mean Tim tim@schedmd.com. If you want I can report your issue for you on their bugtracker.

lahwaacz commented on 2021-04-10 07:44 (UTC)

@edh Do you have some connection to the SchedMD people? I tried to create an account on their bug tracker several times, but the automatic email never arrived. I also tried to contact one person (maybe the maintainer of the tracker) for which I found an email, but also did not get any reply...

edh commented on 2021-03-08 12:30 (UTC)

@lahwaacz Great! Thanks for taking care of this!

lahwaacz commented on 2021-03-08 11:57 (UTC)

@edh: Doh, I'm still at the older version... I'll try to register at SchedMD and report it upstream. Thanks for the info.

edh commented on 2021-03-06 21:50 (UTC) (edited on 2021-03-06 21:57 (UTC) by edh)

@lahwaacz In a recent commit I switched to the service files distributed by the SchedMD upstream. Thus, I think this problem is a more general one and all slurm users regardless of distribution would benefit if either slurmctld would handle /bin/kill -HUP properly or at least the ExecReload command in the service file be replaced. IMHO it would be best to raise the issue upstream at their buptracker. In case you don't want to deal with it upstream, I can also open a ticket in your place.

lahwaacz commented on 2021-03-06 18:43 (UTC)

The slurmctld.service contains ExecReload=/bin/kill -HUP $MAINPID which should instruct slurmctld to reload its configuration. But some changes, such as changing the time limit of a partition, are not reflected by this. I think the reload command should be changed to /usr/bin/scontrol reconfigure which just solved the issue for me.

edh commented on 2020-11-23 20:02 (UTC)

@lahwaacz Great! I'll add those flags with the next release of the software.

lahwaacz commented on 2020-11-22 22:50 (UTC)

Yes, with those flags the part -ggdb3 -Wall -g -O1 -fno-strict-aliasing is gone.

edh commented on 2020-11-22 22:14 (UTC)

Can you try adding --disable-developer --disable-debug --enable-optimizations to the configure call and check whether that resolves your issue? If so, I will add those flags with the next release.

lahwaacz commented on 2020-11-22 11:09 (UTC)

The package is built with funny flags:

-march=x86-64 -mtune=generic -O2 -pipe  -fno-omit-frame-pointer -pthread -ggdb3 -Wall -g -O1 -fno-strict-aliasing ...

The first part comes from makepkg.conf, the rest are Slurm's defaults. There is no reason for -g and -ggdb3 since makepkg strips debugging symbols anyway. Also, -O2 should not be overridden with -O1.

edh commented on 2020-11-17 20:11 (UTC)

With the latest release (20.11.0.1) I changed the way the user creation and the permissions of temporary files are handled. The package now fully embraces systemd-sysusers and systemd-tmpfiles.

lahwaacz commented on 2019-12-21 09:12 (UTC)

--with-pmix=/usr is needed instead of just --with-pmix. Run srun --mpi=list to see the difference.

chenweiguang commented on 2019-07-24 08:25 (UTC)

@edh Thanks a lot. I installed it sucessfully by modifingy PKGBUILD with-hdf5=no. I don't know why sh5util can not be compiled while hdf5 als0 be installed by pacman previousely.

edh commented on 2019-07-24 07:56 (UTC)

@chenweiguang Please do not post such lengthy output into a comment but instead use some pastebin server to host it.

Unfortunately, I can not reproduce your error. Slurm builds just fine on my machine and in a clean chroot. Please do some proper test yourself and check whether the error is due to e.g. building in a polluted build directory etc.

chenweiguang commented on 2019-07-24 00:59 (UTC)

I used AUR helper yaourt to install it. yaourt -S slurm-llnl But an error occur while compiling sh5util, as follow: libtool: link: gcc -DNUMA_VERSION1_COMPATIBILITY -march=x86-64 -mtune=generic -O2 -pipe -pthread -ggdb3 -Wall -g -O1 -fno-strict-aliasing -Wl,-O1 -Wl,--sort-common -Wl,--as-needed -Wl ,-z -Wl,relro -Wl,-z -Wl,lazy -o .libs/sh5util sh5util.o -Wl,-rpath=/usr/lib/slurm -Wl,--export-dynamic -L/home/zznu/miniconda3/lib -L../../../../../src/api/.libs /tmp/yaourt-tmp-zznu/aur-slurm-llnl/src/slurm-slurm-19-05-0-1/src/api/.libs/libslurmfull.so ../.libs/libhdf5_api.a -lm -ldl -lz -lpthread -lrt -lhdf5 -pthread -Wl,-rpath -Wl,/usr/lib/slurm /usr/bin/ld: sh5util.o: in function _extract_series_table': /tmp/yaourt-tmp-zznu/aur-slurm-llnl/src/slurm-slurm-19-05-0-1/src/plugins/acct_gather_profile/hdf5/sh5util/sh5util.c:1079: undefined reference toH5PTopen' /usr/bin/ld: sh5util.o: in function _extract_totals': /tmp/yaourt-tmp-zznu/aur-slurm-llnl/src/slurm-slurm-19-05-0-1/src/plugins/acct_gather_profile/hdf5/sh5util/sh5util.c:935: undefined reference toH5PTget_num_packets' /usr/bin/ld: /tmp/yaourt-tmp-zznu/aur-slurm-llnl/src/slurm-slurm-19-05-0-1/src/plugins/acct_gather_profile/hdf5/sh5util/sh5util.c:939: undefined reference to H5PTget_next' /usr/bin/ld: sh5util.o: in function_extract_series_table': /tmp/yaourt-tmp-zznu/aur-slurm-llnl/src/slurm-slurm-19-05-0-1/src/plugins/acct_gather_profile/hdf5/sh5util/sh5util.c:1089: undefined reference to H5PTget_num_packets' /usr/bin/ld: /tmp/yaourt-tmp-zznu/aur-slurm-llnl/src/slurm-slurm-19-05-0-1/src/plugins/acct_gather_profile/hdf5/sh5util/sh5util.c:1094: undefined reference toH5PTget_next' /usr/bin/ld: /tmp/yaourt-tmp-zznu/aur-slurm-llnl/src/slurm-slurm-19-05-0-1/src/plugins/acct_gather_profile/hdf5/sh5util/sh5util.c:1125: undefined reference to H5PTclose' /usr/bin/ld: /tmp/yaourt-tmp-zznu/aur-slurm-llnl/src/slurm-slurm-19-05-0-1/src/plugins/acct_gather_profile/hdf5/sh5util/sh5util.c:1126: undefined reference toH5PTclose' /usr/bin/ld: /tmp/yaourt-tmp-zznu/aur-slurm-llnl/src/slurm-slurm-19-05-0-1/src/plugins/acct_gather_profile/hdf5/sh5util/sh5util.c:1116: undefined reference to H5PTclose' /usr/bin/ld: sh5util.o: in function_extract_item_step': /tmp/yaourt-tmp-zznu/aur-slurm-llnl/src/slurm-slurm-19-05-0-1/src/plugins/acct_gather_profile/hdf5/sh5util/sh5util.c:1551: undefined reference to H5PTopen' /usr/bin/ld: /tmp/yaourt-tmp-zznu/aur-slurm-llnl/src/slurm-slurm-19-05-0-1/src/plugins/acct_gather_profile/hdf5/sh5util/sh5util.c:1555: undefined reference toH5PTget_num_packets' /usr/bin/ld: sh5util.o: in function _item_analysis_uint': /tmp/yaourt-tmp-zznu/aur-slurm-llnl/src/slurm-slurm-19-05-0-1/src/plugins/acct_gather_profile/hdf5/sh5util/sh5util.c:1275: undefined reference toH5PTget_next' /usr/bin/ld: sh5util.o: in function _extract_item_step': /tmp/yaourt-tmp-zznu/aur-slurm-llnl/src/slurm-slurm-19-05-0-1/src/plugins/acct_gather_profile/hdf5/sh5util/sh5util.c:1590: undefined reference toH5PTclose' /usr/bin/ld: sh5util.o: in function _item_analysis_double': /tmp/yaourt-tmp-zznu/aur-slurm-llnl/src/slurm-slurm-19-05-0-1/src/plugins/acct_gather_profile/hdf5/sh5util/sh5util.c:1376: undefined reference toH5PTget_next' /usr/bin/ld: sh5util.o: in function _extract_item_step': /tmp/yaourt-tmp-zznu/aur-slurm-llnl/src/slurm-slurm-19-05-0-1/src/plugins/acct_gather_profile/hdf5/sh5util/sh5util.c:1605: undefined reference toH5PTclose' collect2: error: ld returned 1 exit status
make[6]: [Makefile:572: sh5util] Error 1 make[6]: Leaving directory '/tmp/yaourt-tmp-zznu/aur-slurm-llnl/src/slurm-slurm-19-05-0-1/src/plugins/acct_gather_profile/hdf5/sh5util' make[5]: [Makefile:722: all-recursive] Error 1 make[5]: Leaving directory '/tmp/yaourt-tmp-zznu/aur-slurm-llnl/src/slurm-slurm-19-05-0-1/src/plugins/acct_gather_profile/hdf5' make[4]: [Makefile:521: all-recursive] Error 1 make[4]: Leaving directory '/tmp/yaourt-tmp-zznu/aur-slurm-llnl/src/slurm-slurm-19-05-0-1/src/plugins/acct_gather_profile' make[3]: [Makefile:552: all-recursive] Error 1 make[3]: Leaving directory '/tmp/yaourt-tmp-zznu/aur-slurm-llnl/src/slurm-slurm-19-05-0-1/src/plugins' make[2]: [Makefile:546: all-recursive] Error 1 make[2]: Leaving directory '/tmp/yaourt-tmp-zznu/aur-slurm-llnl/src/slurm-slurm-19-05-0-1/src' make[1]: [Makefile:674: all-recursive] Error 1 make[1]: Leaving directory '/tmp/yaourt-tmp-zznu/aur-slurm-llnl/src/slurm-slurm-19-05-0-1'

Thanks.

cmcc commented on 2019-05-29 10:58 (UTC)

@edh

yes, solved just after posting my previous comment. Thanks a lot!

edh commented on 2019-05-29 10:30 (UTC)

@cmcc

Sorry about that. I forgot to deprecate autogen.sh in favor of autoreconf like I did for the -git and -dev version.

cmcc commented on 2019-05-29 10:21 (UTC) (edited on 2019-05-29 10:26 (UTC) by cmcc)

Installing the latest version (slurm-llnl 19.05.0.1-1) with yay AUR helper, I get the following error:

[...]

==> Starting build()...

/home/XXXX/.cache/yay/slurm-llnl/PKGBUILD: line 64: ./autogen.sh: No such file or directory

==> ERROR: A failure occurred in build(). Aborting...

Error making: slurm-llnl

[...]

handrake0724 commented on 2018-12-26 22:45 (UTC)

@cbpark Thanks for the comments. I have checked pacman -Ql pmix and it shows pmix /usr/

slurm config.log with --with-pmix=/usr showed unabled to locate pmix due to missing -lpmix

Since libpmix.so was not in /usr/lib indeed, I have pmix reinstalled and now slurm finds pmix.

cbpark commented on 2018-12-26 18:16 (UTC)

@handrake0724 Setting --with-pmix=/usr seems to work well.

checking for pmix installation... /usr

Could you check the message from autoconf? And check the location of pmix.

$ pacman -Ql pmix
pmix /usr/

handrake0724 commented on 2018-12-26 10:53 (UTC)

I have one question about pmix support. pmix AUR is already installed in /usr.

When I run makepkg slurm-18.08.4.1-1, I found that the configuration was unable to locate pmix installation.

--with-pmix=/usr, --with-pmix=/usr/lib, --with-pmix=/usr/lib/pmix, --with-pmix=/usr/lib64, --with-pmix=/usr/lib64/pmix, --with-pmix=/usr/include did not work.

Any idea?

cbpark commented on 2018-04-20 00:28 (UTC)

Thank you for your time. I think it's fair enough for now.

edh commented on 2018-04-19 20:50 (UTC)

@cbpark Thanks for the notch. Done. I assume it should be fine to just add it to the opendepends array it is is no essential for slum to start. Furthermore I did not increase the pkgver as a new optdepends which might not be installed would not effect the build in any way.

cbpark commented on 2018-04-18 14:01 (UTC)

Could you add pmix to depends (or optdepends)?

https://aur.archlinux.org/packages/pmix/

It is to build slurm with openmpi support.

https://slurm.schedmd.com/mpi_guide.html#open_mpi

Adding --with-pmix would be enough.

edh commented on 2018-02-08 17:46 (UTC) (edited on 2018-02-08 17:58 (UTC) by edh)

@handrake0724 Once again, please make sure you use the recent version of the PKGBUILD and its associated files. The problem you describe has at least been addressed partially. I added the removal of the '-fno-plt' flag which somehow slipped through. Thanks for that!

handrake0724 commented on 2018-02-08 13:25 (UTC) (edited on 2018-02-08 13:25 (UTC) by handrake0724)

@edh I found slum-llnl.install was not updated. So the msg related log has gone. BTW, after compiling, I got the following error when starting slumd daemon.

error: plugin_load_from_file: dlopen(/usr/lib/slurm/select_linear.so):/usr/lib/slurm/select_linear.so: undefined symbol: slurm_job_preempt_mode

After googling similar behavior, I got the following link https://bbs.archlinux.org/viewtopic.php?id=233261

following the link, I could fix the above issue by modifying in build() in PKGBUILD as follows:

export CFLAGS="${CFLAGS//-z,now/-z,lazy}"
export CFLAGS="${CFLAGS//-fno-plt/}"

edh commented on 2018-02-01 16:42 (UTC)

@handrake0724 I solved this issue some time ago. Are you sure you are building a recent version of slurm? By the way this message, besides not being desirable, is harmless. The package should work either way. Back in the old-days you could use makepkg's internal functions (like msg, msg2, ...) to print stuff in a fashionable way.

handrake0724 commented on 2018-02-01 14:12 (UTC)

I got the following error while installing

(1/1) installing slurm-llnl [######################] 100% /tmp/alpm_PbzPUP/.INSTALL: line 3: msg: command not found /tmp/alpm_PbzPUP/.INSTALL: line 6: msg: command not found /tmp/alpm_PbzPUP/.INSTALL: line 9: msg: command not found /tmp/alpm_PbzPUP/.INSTALL: line 18: msg: command not found error: command failed to execute correctly

In my linux box, the command msg does not exist. Is there any package needed?

edh commented on 2017-07-12 08:04 (UTC)

@serhatcevikel Thanks, for nudge! While switching the source to GitHub I forgot to change the directory name.

serhatcevikel commented on 2017-07-11 22:38 (UTC)

The build gives the following error: /home/xx/.cache/pacaur/slurm-llnl/PKGBUILD: line 49: cd: /home/xxx/.cache/pacaur/slurm-llnl/src/slurm-17-02-6-1: No such file or directory The mentioned line is orginally: cd "${srcdir}/slurm-${pkgver//./-}" It should be changed to: cd "${srcdir}/slurm-slurm-${pkgver//./-}" to make it confirmant with the directory structure of the build files.

edh commented on 2016-10-29 16:40 (UTC)

@richli Thanks for bringing this to my attention. I addressed the pidfile inconsistency in a recent commit and simply moved them to /run instead of a creating a subfolder. Plus I removed /var/log/slurm/ from the systemd-tmpfile since it is not being used (/var/log/slurm-llnl is and is contained within the package).

richli commented on 2016-10-28 19:52 (UTC)

FYI, the PID file setting in slurmctld.service (/run/slurm-llnl/slurmctld.pid) conflicts with the defaults in slurm.conf.example (/var/run/slurmctld.pid). Same with slurmd.service. Also, shouldn't the /run/slurm-llnl directory be created by slurm-tmpfiles.conf?

edh commented on 2016-04-13 15:18 (UTC) (edited on 2016-04-13 15:20 (UTC) by edh)

@idedios Unfortunately I can't reproduce your error. Assuming that versions prior to 15.08.10 build fine, try to revert any changes made by this commit [1]. If this doesn't help either I would suggest contacting the developers. However in case altering the PKGBUILD helped, please be sure to let me now of the changes which were required. [1] https://aur.archlinux.org/cgit/aur.git/commit/?h=slurm-llnl&id=20854b5963fe5337df9a3072f8838989efa2e786

idedios commented on 2016-04-13 01:30 (UTC)

I'm getting the following error building the latest version on Arch with no GUI: configure.ac:280: error: possibly undefined macro: AM_PATH_GTK_2_0 If this token and others are legitimate, please use m4_pattern_allow. See the Autoconf documentation.

edh commented on 2016-01-21 19:18 (UTC)

@richli Thanks! Good catch.

richli commented on 2016-01-21 19:09 (UTC)

Could you add to the PKGBUILD: backup=('etc/default/slurm-llnl') Otherwise any changes made to that file get wiped when upgrading. Thanks!

edh commented on 2015-03-20 17:05 (UTC)

slurm-llnl version 14.11.5 is out! Furthermore I did some tweaking to the service files and fixed one inconsistency. By the way: Just recently I wrote a short wiki entry. The basic installation precess is described and the fundamental handling of the service files. Feel free to add whatever seems appropiate.

edh commented on 2015-03-08 19:07 (UTC)

The default User ID and Group ID for slurm is now 64030. Prior to this version the UID and GID were the smallest number available, hence this change adds persistence. On systems where slurm is already configured, nothing is changed!

edh commented on 2015-03-08 18:28 (UTC)

Redesigned PKGBUILD, with new sysconfdir residing under /etc/slurm-llnl instead of /etc/slurm. In addation a much cleaner handling of service files was implemented. Furthermore the dependency array got an incremental cleanup and some unnecessary patches were removed.

edh commented on 2015-02-19 12:32 (UTC)

Updated PKGBUILD version which classifies quite every dependency as optional, as this is how slurm-llnl is intended to be build. Only the essential deps are included, although munge could also be replaced however this is not recommended by the developers. One further improvement is that important notes which may be of interest to the user are now highlighted.

wookietreiber commented on 2015-02-19 07:19 (UTC)

Nice, with the optional dependencies. You have however added some both to depends as well as optdepends.

edh commented on 2015-02-18 19:12 (UTC)

Added a whole lot of optional dependencies, which may or may not be compiled into slurm if present prior to the compilation.

tegenf commented on 2015-01-19 09:04 (UTC)

Sorry for the lack of updates. Here is the latest version of slurm-llnl. I'll be more attentive in the future...

tegenf commented on 2014-05-12 08:28 (UTC)

Updated to 14.03.3-2. This one is not yet in archive. An idea is to maintain the latest version in 'archive' instead (thx balwierz). However this will not be the latest version available. The advantage though is that the link wont be invalid each time there is an update. Would such an approach be preferable?

balwierz commented on 2014-05-09 14:55 (UTC)

If you change in the URL "latest" --> "archive" the link won't get broken when there is a new version released. In "latest" is only the most recent version kept. And now the URL is invalid again.

tegenf commented on 2014-04-24 08:38 (UTC)

Note I have only tested it to the point that it compiles and that namcap does not make too serious complains.

tegenf commented on 2014-04-24 08:37 (UTC)

New version 14.03.1-2

tegenf commented on 2014-04-10 10:49 (UTC)

Update to latest slurm v 14.03.0

balwierz commented on 2014-04-09 12:24 (UTC)

Broken! curl: (22) The requested URL returned error: 404 Not Found ==> ERROR: Failure while downloading slurm-2.6.7.tar.bz2 Aborting... BTW. What is the difference between this package and "slurm" package? Different forks of the same software. Is any one better than the other in some aspects?

milo.nimrod commented on 2014-03-30 14:02 (UTC)

need to Update version 14.03.0

tegenf commented on 2014-03-17 13:55 (UTC)

Slurm 2.6.7 * added patch for man2html (thx munyamunya) * removed slurmctld as requirement for slurmd in slurmd.service file Note: if MySQL is used, this version (and earlier) does not work with MySQL v5.6.16. For now I use 14.03-rc1 which contains a fix for this bug.

munyamunya commented on 2014-03-17 07:35 (UTC)

Ok, "make -j1" does not actually cure freeze, might be really random. I had to remove man and html from docs makefiles and rerun autogen. Does anybody else experience this weird thing?

munyamunya commented on 2014-03-17 06:15 (UTC)

You should also patch ./doc/man/man2html.py to use python2. Also build process strangely randomly hangs when it is converting files with said script (with multithreaded make). I had to force make -j1 to make it work.

tegenf commented on 2014-03-04 09:23 (UTC)

sorry for the delay, here is slurm 2.6.6-2

teetee1 commented on 2013-12-01 06:15 (UTC)

2.6.3 is available at http://www.schedmd.com/download/archive/slurm-2.6.3.tar.bz2

richli commented on 2013-11-23 23:56 (UTC)

The source URL returns 404. Looks like 2.6.4 is now available.

tegenf commented on 2013-10-15 08:58 (UTC)

New package for 2.6.3. FYI, namcap reports the following missing dependencies: lua51, rrdtool, hdf5, numactl, gtk2 I will fix that later.

tegenf commented on 2013-09-24 11:43 (UTC)

update for 2.6.2

aorth commented on 2013-09-12 16:54 (UTC)

Updated PKGBUILD for 2.6.2: https://github.com/alanorth/archbuilds/commit/b4406d7e5efd2510009cdd188ec09bcd700b3971

aorth commented on 2013-08-17 19:45 (UTC)

I've just updated my PKGBUILD for new upstream release (2.6.1). It's in my github repo here: https://github.com/alanorth/archbuilds/commit/56062595a9a871a3338290718602fa5bc9fe8780

aorth commented on 2013-07-14 19:51 (UTC)

I've just updated my PKGBUILD. Changes include: - SLURM 2.6.0[0] - libmysqlclient -> libmariadbclient[1] - Remove DB libs from depends[2]; technically they're optional (the build system will detect and use them if they're installed) [0] https://github.com/alanorth/archbuilds/commit/3b1fc4ab7ff1d95a9d754904056587cce0835270 [1] https://github.com/alanorth/archbuilds/commit/48fe6efefcde4a547ba68969f9c28659d9b505e0 [2] https://github.com/alanorth/archbuilds/commit/b2adb5604d46544f3d687172fdb64fe14bc2d0c7

tegenf commented on 2013-06-24 09:43 (UTC)

Ok, sorry for the delay but now I have updated to 2.5.7 using the improvements and updates from aorth. Thank you very much for providing us with the update!

aorth commented on 2013-06-08 19:21 (UTC)

I've updated my version of this package to 2.5.7[0]. If you don't want to clone my whole archbuilds git repo[1] to get this, I've posted a tarball of this package as well[2]. [0] https://github.com/alanorth/archbuilds/commit/defe04a5e54bc0dc2795d6dfd05a0b1f009b5cfb [1] https://github.com/alanorth/archbuilds [2] https://copy.com/vGG2TjlxTVOp (md5sum: b1c42fd75225ef8b8fb5f0a468492d1d)

wookietreiber commented on 2013-06-08 13:10 (UTC)

yeah, would be awesome to have these changes incorporated

aorth commented on 2013-06-02 19:29 (UTC)

Ok, I've greatly improved, reorganized and simplified this package: - move all binaries from /usr/sbin to /usr/bin - rename service files from slurm*-llnl.service to slurm*.service - fix the dependencies for slurmdbd (it's sufficient for slurmdbd.service to say "Before=slurmctld.service", which won't affect people who don't use slurmdbd). - more?! It's available on my github if you want to take a look at the commit history: https://github.com/alanorth/archbuilds/tree/master/slurm-llnl

aorth commented on 2013-06-01 07:56 (UTC)

Heads up, you'll need to move /usr/sbin/slurm* to /usr/bin. See Allan's comments regarding finishing the /usr move soon in [testing]: https://mailman.archlinux.org/pipermail/arch-dev-public/2013-May/025003.html

wookietreiber commented on 2013-05-12 20:56 (UTC)

Please check the package with namcap and fix the issues (RPATH, dependencies, ...).

wookietreiber commented on 2013-04-26 09:08 (UTC)

thx for updating. is the a reason why you have chosen http://www.schedmd.com/ over the direct slurm page http://slurm.schedmd.com/

tegenf commented on 2013-04-26 08:18 (UTC)

* updated to 2.5.6 thx wookietreiber for pointing out the faulty upstream url.

wookietreiber commented on 2013-04-25 03:53 (UTC)

please also update upstream url to http://slurm.schedmd.com/

aorth commented on 2013-03-13 19:50 (UTC)

@tegenf: You right, I hadn't thought about that. The machine I run Arch + slurm on does have slurmctld. That's tricky... On a related note, today I was upgrading my production slurm cluster (on CentOS) to 2.5.4 and I noticed that there are only init files for slurmd and slurmdbd; I'm not sure how they did it, but the slurmd service starts the slurmctld daemon. I will look to see if any systemd distros like Fedora have slurm packages with service files we can compare.

tegenf commented on 2013-03-12 08:41 (UTC)

aorth: I have not yet added your changes. For the slurmdbd service file I understand the change. However, for the slurmd, why the "Requires=slurmctld-llnl.service" ? What happens when you install the slurmd on a node where there is no slurmctld ?

tegenf commented on 2013-03-12 08:39 (UTC)

updated package to 2.5.4

wookietreiber commented on 2013-03-11 23:25 (UTC)

2.5.4 is out

aorth commented on 2013-02-17 10:27 (UTC)

I recently switched to using SlurmDBD + MySQL for job accounting and I was having problems with the services not starting up. I've fixed the dependencies in the systemd service files to fix this. Here's a new source package with my changes: https://mega.co.nz/#!0slEwbAB!GARgCQBn6roG10G_Y5bwXih9Iv1TodjZDes-GSh-BnE I don't think the dependency changes will affect users who aren't using SlurmDBD, but I'd love it if someone could test. :)

tegenf commented on 2013-02-11 09:44 (UTC)

Here is the update for 2.5.3

aorth commented on 2013-02-10 13:41 (UTC)

PKGBUILD for 2.5.3: http://ix.io/4lD Built on top of this 2.5.2 AUR tarball.

giddie commented on 2013-02-06 10:12 (UTC)

The issue was quickly fixed. It'll be resolved in the next version, and the patch is here: https://github.com/SchedMD/slurm/commit/1e16b4bf25e94745e395d4450ea3d6a47d690257

big_gie commented on 2013-02-05 17:27 (UTC)

@giddie: yes, github is fine for reporting issues. At least it was when I did some months ago ;) I'm still registered to the mailing list, I could forward an email on somebody's behalf.

giddie commented on 2013-02-05 16:10 (UTC)

I've created an issue in the SchedMD Github repo, but I've no idea if that's the "real" upsteam? https://github.com/SchedMD/slurm/issues/37

giddie commented on 2013-02-05 15:52 (UTC)

My patch is a rather ugly workaround to force SLURM to build without sview. It's not really the proper solution to this bug, and it's only useful to people who want to build on a headless server, without GTK. The current PKGBUILD will build fine if GTK is installed, I expect. I can't see a link to a bug tracker for SLURM, only a mailing list, and I don't want to have to deal with a mailing list to report this. Maybe if it doesn't go away with the next release...

wookietreiber commented on 2013-02-05 15:25 (UTC)

Ok. Then, I guess, we can work with the patch instead of the sed command in PKGBUILD. @tegenf: Care to apply the patch to PKGBUILD and push a 2.5.2-2? @giddie: Care to send https://github.com/SchedMD/slurm a pull request with your patch?

giddie commented on 2013-02-05 14:46 (UTC)

I believe what is required is a little more complex. Sorry if it wasn't clear: it's not a case of swapping one line for another. Patch is here: https://gist.github.com/giddie/4714872

wookietreiber commented on 2013-02-05 13:51 (UTC)

I have not tested it. But it basically simulates your manual workflow: makepkg -o $EDITOR configure makepkg -e ... by hacking the PKGBUILD rather than the source directly and thus should be more convenient. In case the editing part you did was more complex, the sed would need to be more sophisticated, of course.

giddie commented on 2013-02-05 13:32 (UTC)

Have you tested that? I'm not sure it'll work. Those variables are used to conditionally comment-out portions of a Makefile. Ugh; I really don't like Autotools.

wookietreiber commented on 2013-02-05 13:27 (UTC)

add this to the build() function in PKGBUILD: sed -i "s/BUILD_SVIEW_TRUE='#'/BUILD_SVIEW_FALSE/" configure

giddie commented on 2013-02-05 12:22 (UTC)

It seems that the latest SLURM has a bug that means that it still tries to build sview even when GTK isn't present. Note that the following page mentions that the intention is for this not to happen: https://computing.llnl.gov/linux/slurm/quickstart_admin.html I can't find a bug tracker upstream, so I couldn't report the bug. The simplest workaround I could find was to do "makepkg -o", open "configure" at line 20417, and remove everything except the second branch: BUILD_SVIEW_TRUE='#' BUILD_SVIEW_FALSE= After that, "makepkg -e", and all seems good.

wookietreiber commented on 2013-02-01 10:49 (UTC)

... correction: munged isn't exactly required. Actually it should be set to optional with 'Wants=' because one does not necessarily need it. But a) iirc it is recommended by the slurm guys and b) this build does not make it an optional dependency, so possibly for the current hard dependency it might be 'Required='.

wookietreiber commented on 2013-02-01 10:15 (UTC)

I guess the service files should also include "Requires=munged.serviceX", not just "After=..." as explained here: https://wiki.archlinux.org/index.php/Systemd#Handling_dependencies I modified these locally (added comments for reasons why): tail /etc/systemd/system/slurm* # requires munged to be loaded before ==> /etc/systemd/system/slurmctld-llnl.service <== .include /usr/lib/systemd/system/slurmctld-llnl.service [Unit] Requires=munged.service # requires slurmctld-llnl to be loaded before (should it be the other way around? that slurmctld has an optional (Wants=) dependency to slurmdbd) ==> /etc/systemd/system/slurmdbd-llnl.service <== .include /usr/lib/systemd/system/slurmdbd-llnl.service [Unit] Requires=slurmctld-llnl.service # requires munged to be loaded before # optional dependency slurmctld (Wants=) ==> /etc/systemd/system/slurmd-llnl.service <== .include /usr/lib/systemd/system/slurmd-llnl.service [Unit] Requires=munged.service Wants=slurmctld-llnl.service As explained above I guess but don't know exactly that slurmdbd is an optional dependency of slurmctld.

tegenf commented on 2013-02-01 09:42 (UTC)

Update for 2.5.2. Also updated service file for slurmd - was set run after slurmctld only which is fine on the main server but not on the nodes.

big_gie commented on 2013-01-22 16:30 (UTC)

And v2.5.1 is out ;)

big_gie commented on 2013-01-22 16:30 (UTC)

That's probably because v2.4.5 is not the "latest" version. After reporting, they added a folder containing all releases. See https://bugs.gentoo.org/show_bug.cgi?id=431664#c8 The URL is: http://schedmd.com/download/total/ instead of http://schedmd.com/download/latest/

giddie commented on 2013-01-22 15:48 (UTC)

FYI: the pkgbuild currently fails to build because of a 404 Not Found when trying to download the upstream source.

wookietreiber commented on 2013-01-08 12:12 (UTC)

... but I guess one could add separate packages: slurm-llnl-slurmd (compute node) slurm-llnl-slurmctld (master node) slurm-llnl-slurmdbd (database daemon)

wookietreiber commented on 2013-01-08 12:10 (UTC)

@giddie: +1 would also be nice to choose whether to use postgresql / mysql depending on whether or not which db backend is chosen

giddie commented on 2013-01-08 10:45 (UTC)

I wonder if gtk2 could be made an optdepends? If this were a binary package, and it had been built against gtk2, it would obviously need to be a hard dependency, but since this is on the AUR and it actually builds fine without GTK2, maybe it could be optional? (I always tweak the PKGBUILD to remove gtk2 manually as my cluster is headless, but would be nice if this weren't necessary.)

tegenf commented on 2012-12-11 13:02 (UTC)

Update for 2.4.5

wookietreiber commented on 2012-11-06 12:20 (UTC)

iirc the old rc.d slurm file started slurmd as user slurm and that was no problem afaik i figure munge is responsible for the auth stuff anyway other resource and job management systems have no problem with running the node-daemons without root so i figure why should slurm

tegenf commented on 2012-11-06 11:42 (UTC)

The slurmd daemon runs on all nodes in a cluster. In order for it to be able to launch jobs as an arbitrary user, it needs to be run as root. On the other hand, slurmctld does not need root privileges and runs as user slurm.

wookietreiber commented on 2012-11-06 11:13 (UTC)

would you please elaborate why slurmd-llnl.service has "User=root"

tegenf commented on 2012-11-06 08:46 (UTC)

Update to 2.4.4.

tegenf commented on 2012-09-19 12:37 (UTC)

Update to 2.4.3 Tested on: Linux bee 3.3.7-1-ARCH #1 SMP PREEMPT Tue May 22 00:26:26 CEST 2012 x86_64 GNU/Linux gcc version: cc (GCC) 4.7.0 20120505 (prerelease)

wookietreiber commented on 2012-08-28 12:54 (UTC)

same error as "Comment by: jeffdn on Mon, 30 Jul 2012 09:45:15 +0000" libtool: compile: gcc -DHAVE_CONFIG_H -I. -I../.. -I../../slurm -I../.. -DGRES_CONFIG_FILE=\"/etc/slurm/gres.conf\" -march=x86-64 -mtune=generic -O2 -pipe -fstack-protector --param=ssp-buffer-size=4 -D_FORTIFY_SOURCE=2 -pthread -fno-gcse -MT jobacct_common.lo -MD -MP -MF .deps/jobacct_common.Tpo -c jobacct_common.c -fPIC -DPIC -o .libs/jobacct_common.o jobacct_common.c: In function 'jobacct_common_setinfo': jobacct_common.c:150:33: error: dereferencing pointer to incomplete type jobacct_common.c:151:34: error: dereferencing pointer to incomplete type jobacct_common.c:152:32: error: dereferencing pointer to incomplete type jobacct_common.c:153:33: error: dereferencing pointer to incomplete type jobacct_common.c: In function 'jobacct_common_getinfo': jobacct_common.c:221:28: error: invalid application of 'sizeof' to incomplete type 'struct rusage' jobacct_common.c:222:9: error: dereferencing pointer to incomplete type jobacct_common.c:223:9: error: dereferencing pointer to incomplete type jobacct_common.c:224:9: error: dereferencing pointer to incomplete type jobacct_common.c:225:9: error: dereferencing pointer to incomplete type make[3]: *** [jobacct_common.lo] Error 1 $ uname -a Linux mobster 3.4.9-1-ARCH #1 SMP PREEMPT Wed Aug 15 18:59:31 CEST 2012 x86_64 GNU/Linux $ gcc --version gcc (GCC) 4.7.1 20120721 (prerelease)

commented on 2012-08-13 02:19 (UTC)

Hi tegenf -- I'm on an Intel Centrino, so it's i686. My gcc version is 4.7.1 20120712. I'm going to try the new version now.

commented on 2012-08-07 04:16 (UTC)

I was getting the same error as jeffdn. It looks like it's due to header files moving around due to kernel updates. I was able to get it to compile with this patch https://gist.github.com/3281505.

tegenf commented on 2012-08-02 08:54 (UTC)

Updated to 2.4.2. jeffdn: I don't get your error. What is your architecture? And gcc version? I have not looked into that yet.

commented on 2012-07-30 09:45 (UTC)

I got this error on during slurm's compile today: libtool: compile: gcc -DHAVE_CONFIG_H -I. -I../.. -I../../slurm -I../.. -DGRES_CONFIG_FILE=\"/etc/slurm/gres.conf\" -march=i686 -mtune=generic -O2 -pipe -fstack-protector --param=ssp-buffer-size=4 -D_FORTIFY_SOURCE=2 -pthread -fno-gcse -MT jobacct_common.lo -MD -MP -MF .deps/jobacct_common.Tpo -c jobacct_common.c -fPIC -DPIC -o .libs/jobacct_common.o jobacct_common.c: In function 'jobacct_common_setinfo': jobacct_common.c:150:33: error: dereferencing pointer to incomplete type jobacct_common.c:151:34: error: dereferencing pointer to incomplete type jobacct_common.c:152:32: error: dereferencing pointer to incomplete type jobacct_common.c:153:33: error: dereferencing pointer to incomplete type jobacct_common.c: In function 'jobacct_common_getinfo': jobacct_common.c:221:28: error: invalid application of 'sizeof' to incomplete type 'struct rusage' jobacct_common.c:222:9: error: dereferencing pointer to incomplete type jobacct_common.c:223:9: error: dereferencing pointer to incomplete type jobacct_common.c:224:9: error: dereferencing pointer to incomplete type jobacct_common.c:225:9: error: dereferencing pointer to incomplete type --- snip --- make: *** [all] Error 2 ==> ERROR: A failure occurred in build(). Aborting... Just wanted to let you know!

tegenf commented on 2012-07-13 07:37 (UTC)

Updated to 2.4.1. With namcap I got one new warning: slurm-llnl W: Referenced library 'libslurm.so.24' is an uninstalled dependency I have not yet been able to test this so I don't know if this is a problem. I will test this asap but at the moment I cannot.

tegenf commented on 2012-05-23 15:35 (UTC)

Fixed MD5SUM. The file to download actually changed between the date I first retrieved it (May 17) and today! I verified that the MD5SUM is correct on http://www.schedmd.com/#repos .

giddie commented on 2012-05-23 15:24 (UTC)

I'm getting a failed validity check; could the source package have changed for any reason? ==> Validating source files with md5sums... arch_rc.d.slurm ... Passed arch_rc.d.slurmdbd ... Passed shtml2html.py.patch ... Passed slurm-2.3.5.tar.bz2 ... FAILED ==> ERROR: One or more files did not pass the validity check! ==> ERROR: Makepkg was unable to build slurm-llnl.

tegenf commented on 2012-05-17 09:29 (UTC)

update for 2.3.5

tegenf commented on 2012-03-28 09:08 (UTC)

* slurm version 2.3.4 Note that this package will break as soon there is a new release. The reason for this is that only the latest version is in the download directory.

fhs commented on 2012-03-22 07:45 (UTC)

2.3.4 is released and it seems the 2.3.3 archive is gone now

tegenf commented on 2012-02-23 08:08 (UTC)

:) no reason - here it is, 2.3.3 .

big_gie commented on 2012-02-22 19:15 (UTC)

Any reason for not bumping to 2.3.3?

tegenf commented on 2012-02-08 10:19 (UTC)

Fixed post_install() - was defined in both PKGBUILD and .install file

tegenf commented on 2012-02-08 09:33 (UTC)

I have added a new /etc/rc.d/slurmdbd adapted for arch linux. In previous packages it did not work. Note that I do not use slurmdbd so I have not been able to test this 100%. Please note the comment at the end of the install (change /var/run to /run in slurmdbd.conf and create /var/log/slurm).

tegenf commented on 2012-02-08 08:18 (UTC)

Hi, I see the problem and I have a solution. Sorry, I have not used slurmdbd so I did not notice it was broken. A new version will appear soon. Thanks for pointing this out!

commented on 2012-02-07 20:20 (UTC)

Hi, when starting the daemon 'slurmdbd' I get the error: "Could not find /etc/rc.d/init.d/functions. Is some other daemon launch mechanism used?" I symlinked /etc/rc.d/init.d to /etc/rc.d but it still doesn't work: "starting slurmdbd: /etc/rc.d/slurmdbd: line 80: daemon: command not found touch: cannot touch `/var/lock/subsys/slurmdbd': No such file or directory" Has somebody worked this out yet?

tegenf commented on 2012-01-16 13:53 (UTC)

update to new /run dir structure

commented on 2011-11-28 21:56 (UTC)

I'm getting "munge" as missing dependency.

tegenf commented on 2011-06-07 14:47 (UTC)

* change in makedepends mysql to libmysqlclient * uses slurm 2.2.6

tegenf commented on 2011-05-23 11:58 (UTC)

added a check in the install for the environment variable SLURM_USER_ID If set, it will use this as a numerical ID.

tegenf commented on 2011-05-19 09:38 (UTC)

removed two cases in rc.d script where the slurm name was hardcoded

tegenf commented on 2011-02-07 15:44 (UTC)

fixed rc.d script