commit: dc471fd436829cd2029175e0e9cd4c61cf9d5e1d Author: Benda Xu <heroxbd <AT> gentoo <DOT> org> AuthorDate: Sat Aug 31 14:08:51 2024 +0000 Commit: Benda XU <heroxbd <AT> gentoo <DOT> org> CommitDate: Fri Oct 18 13:37:05 2024 +0000 URL: https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=dc471fd4
sys-cluster/slurm: revive with version 24.05.3. Bug: https://bugs.gentoo.org/631552 Bug: https://bugs.gentoo.org/920104 Signed-off-by: Benda Xu <heroxbd <AT> gentoo.org> sys-cluster/slurm/Manifest | 1 + sys-cluster/slurm/files/logrotate | 20 +++ sys-cluster/slurm/files/slurm.confd | 6 + sys-cluster/slurm/files/slurm.tmpfiles | 1 + sys-cluster/slurm/files/slurmctld.initd | 76 +++++++++ sys-cluster/slurm/files/slurmd.initd | 79 +++++++++ sys-cluster/slurm/files/slurmdbd.initd | 74 +++++++++ sys-cluster/slurm/metadata.xml | 27 ++++ sys-cluster/slurm/slurm-24.05.3.ebuild | 276 ++++++++++++++++++++++++++++++++ 9 files changed, 560 insertions(+) diff --git a/sys-cluster/slurm/Manifest b/sys-cluster/slurm/Manifest new file mode 100644 index 000000000000..6e8549336d1c --- /dev/null +++ b/sys-cluster/slurm/Manifest @@ -0,0 +1 @@ +DIST slurm-24.05.3.tar.bz2 7189600 BLAKE2B 4da4c49a7db334494ea03529f45016741e93603680f592d0baf34657e12079f3766e32f67725d16c17336b28206cf920ae2314c9b7076bda22ee1cb239932229 SHA512 fa903aea77b667911d90e9bd7aec3a21b43d4f642e2fdf3f6eb309fd6c8ca0892fba36f371f67e944a4198248ff44d5b6ac6c0edd03cfd4f3e1828b03c12a539 diff --git a/sys-cluster/slurm/files/logrotate b/sys-cluster/slurm/files/logrotate new file mode 100644 index 000000000000..978a869eb993 --- /dev/null +++ b/sys-cluster/slurm/files/logrotate @@ -0,0 +1,20 @@ +/var/log/slurm/*.log { + weekly + compress + missingok + nocopytruncate + nocreate + nodelaycompress + nomail + notifempty + noolddir + rotate 5 + sharedscripts + size=5M + create 640 slurm slurm + postrotate + /etc/init.d/slurmd reload + /etc/init.d/slurmctld reload + /etc/init.d/slurmdbd reload + endscript +} diff --git a/sys-cluster/slurm/files/slurm.confd b/sys-cluster/slurm/files/slurm.confd new file mode 100644 index 000000000000..c406dc3f2a2c --- /dev/null +++ b/sys-cluster/slurm/files/slurm.confd @@ -0,0 +1,6 @@ +# /etc/conf.d/slurm: config file for /etc/init.d/slurm* +SLURMCTLD_OPTIONS="" +SLURMDBD_OPTIONS="" +SLURMD_OPTIONS="" +SLURM_USE_MUNGE="0" +SLURM_USE_YPBIND="0" \ No newline at end of file diff --git a/sys-cluster/slurm/files/slurm.tmpfiles b/sys-cluster/slurm/files/slurm.tmpfiles new file mode 100644 index 000000000000..c77457db5576 --- /dev/null +++ b/sys-cluster/slurm/files/slurm.tmpfiles @@ -0,0 +1 @@ +d /run/slurm 755 slurm slurm - diff --git a/sys-cluster/slurm/files/slurmctld.initd b/sys-cluster/slurm/files/slurmctld.initd new file mode 100644 index 000000000000..5cb5888fa153 --- /dev/null +++ b/sys-cluster/slurm/files/slurmctld.initd @@ -0,0 +1,76 @@ +#!/sbin/openrc-run +# Copyright 1999-2018 Gentoo Authors +# Distributed under the terms of the GNU General Public License v2 + +extra_commands="reload" + +[[ -f /etc/conf.d/slurm ]] && . /etc/conf.d/slurm + +depend() { + local _need + [[ ${SLURM_USE_MUNGE} -ne 0 ]] && _need="${_need} munged" + [[ ${SLURM_USE_YPBIND} -ne 0 ]] && _need="${_need} ypbind" + after logger slurmdbd + need ${_need} +} + +create_folder_and_set_permission() { + folder=${1} + # Fix permissions (/tmp is wiped at boot) + if [[ "${folder}" == "/tmp" ]]; then + eerror "SLURM: folder == ${folder} == /tmp" + exit + fi + if [[ "${folder}" == "/var/tmp" ]]; then + eerror "SLURM: folder == ${folder} == /var/tmp" + exit + fi + checkpath -d -o slurm:slurm "${folder}" +} + +checkconfig() { + if [ ! -e "/etc/slurm/slurm.conf" ]; then + eerror "Missing config /etc/slurm/slurm.conf" + eerror "Customize sample one or generate new by configurator.html" + eerror "in slurm doc directory" + return 1 + fi + + # Make sure folders exists + . /etc/slurm/slurm.conf + folders=( + ${SlurmdSpoolDir} + $(dirname ${SlurmctldLogFile}) + $(dirname ${SlurmctldPidFile}) + $(dirname ${SlurmdLogFile}) + $(dirname ${SlurmdPidFile}) + ${SlurmdSpoolDir} + ${StateSaveLocation} + ) + for folder in ${folders[@]}; do + create_folder_and_set_permission "${folder}" + done +} + +start() { + ebegin "Starting slurm control daemon" + checkconfig + start-stop-daemon --start --quiet --background \ + --pidfile /run/slurm/slurmctld.pid \ + --user slurm \ + --group slurm \ + --exec /usr/sbin/slurmctld -- ${SLURMCTLD_OPTS} + eend $? +} + +stop() { + ebegin "Stopping slurm control daemon" + start-stop-daemon --stop --pidfile /run/slurm/slurmctld.pid + eend $? +} + +reload() { + ebegin "Reloading slurm daemon configuration" + start-stop-daemon --signal HUP --pidfile /run/slurm/slurmctld.pid + eend $? +} diff --git a/sys-cluster/slurm/files/slurmd.initd b/sys-cluster/slurm/files/slurmd.initd new file mode 100644 index 000000000000..dd9797c0c7d2 --- /dev/null +++ b/sys-cluster/slurm/files/slurmd.initd @@ -0,0 +1,79 @@ +#!/sbin/openrc-run +# Copyright 1999-2018 Gentoo Authors +# Distributed under the terms of the GNU General Public License v2 + +extra_commands="reload" + +[[ -f /etc/conf.d/slurm ]] && . /etc/conf.d/slurm + +depend() { + local _need + [[ ${SLURM_USE_MUNGE} -ne 0 ]] && _need="${_need} munged" + [[ ${SLURM_USE_YPBIND} -ne 0 ]] && _need="${_need} ypbind" + after logger + need ${_need} +} + +create_folder_and_set_permission() { + folder=${1} + # Fix permissions (/tmp is wiped at boot) + if [[ "${folder}" == "/tmp" ]]; then + eerror "SLURM: folder == ${folder} == /tmp" + exit + fi + if [[ "${folder}" == "/var/tmp" ]]; then + eerror "SLURM: folder == ${folder} == /var/tmp" + exit + fi + checkpath -d -o slurm:slurm ${folder} +} + +checkconfig() { + if [ ! -e "/etc/slurm/slurm.conf" ]; then + eerror "Missing config /etc/slurm/slurm.conf" + eerror "Customize sample one or generate new by configurator.html" + eerror "in slurm doc directory" + return 1 + fi + # Make sure folders exists + . /etc/slurm/slurm.conf + folders=( + ${SlurmdSpoolDir} + $(dirname ${SlurmctldLogFile}) + $(dirname ${SlurmctldPidFile}) + $(dirname ${SlurmdLogFile}) + $(dirname ${SlurmdPidFile}) + ${SlurmdSpoolDir} + ${StateSaveLocation} + ) + for folder in ${folders[@]}; do + create_folder_and_set_permission "${folder}" + done +} + +start() { + ebegin "Starting slurm daemon" + checkconfig + # Slurm does not propagate limits, which can brakes InfiniBand + # See http://www.open-mpi.org/faq/?category=openfabrics + # and https://computing.llnl.gov/linux/slurm/faq.html#rlimit + ulimit -l unlimited + start-stop-daemon --start --quiet --background \ + --pidfile /run/slurm/slurmd.pid \ + --user root \ + --group root \ + --exec /usr/sbin/slurmd -- ${SLURMD_OPTS} + eend $? +} + +stop() { + ebegin "Stopping slurm daemon" + start-stop-daemon --stop --pidfile /run/slurm/slurmd.pid + eend $? +} + +reload() { + ebegin "Reloading slurm daemon configuration" + start-stop-daemon --signal HUP --pidfile /run/slurm/slurmd.pid + eend $? +} diff --git a/sys-cluster/slurm/files/slurmdbd.initd b/sys-cluster/slurm/files/slurmdbd.initd new file mode 100644 index 000000000000..27b5102101cb --- /dev/null +++ b/sys-cluster/slurm/files/slurmdbd.initd @@ -0,0 +1,74 @@ +#!/sbin/openrc-run +# Copyright 1999-2018 Gentoo Authors +# Distributed under the terms of the GNU General Public License v2 + +extra_commands="reload" + +[[ -f /etc/conf.d/slurm ]] && . /etc/conf.d/slurm + +depend() { + local _need + [[ ${SLURM_USE_MUNGE} -ne 0 ]] && _need="${_need} munged" + [[ ${SLURM_USE_YPBIND} -ne 0 ]] && _need="${_need} ypbind" + after logger + need ${_need} +} + +create_folder_and_set_permission() { + folder=${1} + # Fix permissions (/tmp is wiped at boot) + if [[ "${folder}" == "/tmp" ]]; then + eerror "SLURM: folder == ${folder} == /tmp" + exit + fi + if [[ "${folder}" == "/var/tmp" ]]; then + eerror "SLURM: folder == ${folder} == /var/tmp" + exit + fi + checkpath -d -o slurm:slurm ${folder} +} + +checkconfig() { + if [ ! -e "/etc/slurm/slurmdbd.conf" ]; then + eerror "Missing config /etc/slurm/slurmdbd.conf" + return 1 + fi + + # Make sure folders exists + . /etc/slurm/slurm.conf + folders=( + ${SlurmdSpoolDir} + $(dirname ${SlurmctldLogFile}) + $(dirname ${SlurmctldPidFile}) + $(dirname ${SlurmdLogFile}) + $(dirname ${SlurmdPidFile}) + ${SlurmdSpoolDir} + ${StateSaveLocation} + ) + for folder in ${folders[@]}; do + create_folder_and_set_permission "${folder}" + done +} + +start() { + ebegin "Starting slurmdb daemon" + checkconfig + start-stop-daemon --start --quiet --background \ + --pidfile /run/slurm/slurmdbd.pid \ + --user slurm \ + --group slurm \ + --exec /usr/sbin/slurmdbd -- ${SLURMD_OPTS} + eend $? +} + +stop() { + ebegin "Stopping slurmdb daemon" + start-stop-daemon --stop --pidfile /run/slurm/slurmdbd.pid + eend $? +} + +reload() { + ebegin "Reloading slurmdb daemon configuration" + start-stop-daemon --signal HUP --pidfile /run/slurm/slurmdbd.pid + eend $? +} diff --git a/sys-cluster/slurm/metadata.xml b/sys-cluster/slurm/metadata.xml new file mode 100644 index 000000000000..9ddcae0895da --- /dev/null +++ b/sys-cluster/slurm/metadata.xml @@ -0,0 +1,27 @@ +<?xml version="1.0" encoding="UTF-8"?> +<!DOCTYPE pkgmetadata SYSTEM "https://www.gentoo.org/dtd/metadata.dtd";> +<pkgmetadata> + <maintainer type="person"> + <email>[email protected]</email> + <name>Alexey Shvetsov</name> + </maintainer> + <maintainer type="project"> + <email>[email protected]</email> + <name>Gentoo Cluster Project</name> + </maintainer> + <use> + <flag name="html">Build html documentation</flag> + <flag name="ipmi">Build support for collecting some ipmi stats</flag> + <flag name="json">Add support for json-persing via json-c</flag> + <flag name="munge">Enable authentication via munge</flag> + <flag name="multiple-slurmd">Allow multiple slurmd to run</flag> + <flag name="numa">Add NUMA awareness</flag> + <flag name="nvml">Enable the NVML device discovery</flag> + <flag name="ofed">Add Infiniband support via ofed</flag> + <flag name="slurmdbd">Install SQL server for slurmdbd</flag> + <flag name="slurmrestd">Enable interface to Slurm via REST API</flag> + <flag name="torque">Enable perl scripts that emulates pbs (qstat, qsub etc.)</flag> + <flag name="ucx">Enable Unified Communication X library support</flag> + <flag name="yaml">Enable YAML support</flag> + </use> +</pkgmetadata> diff --git a/sys-cluster/slurm/slurm-24.05.3.ebuild b/sys-cluster/slurm/slurm-24.05.3.ebuild new file mode 100644 index 000000000000..1078be9a6b38 --- /dev/null +++ b/sys-cluster/slurm/slurm-24.05.3.ebuild @@ -0,0 +1,276 @@ +# Copyright 1999-2022 Gentoo Authors +# Distributed under the terms of the GNU General Public License v2 + +EAPI=8 + +LUA_COMPAT=( lua5-{1..4} ) + +if [[ ${PV} == *9999* ]]; then + EGIT_REPO_URI="https://github.com/SchedMD/slurm.git"; + INHERIT_GIT="git-r3" + MY_P="${P}" +else + if [[ ${PV} == *pre* || ${PV} == *rc* ]]; then + MY_PV=$(ver_rs '-0.') # pre-releases or release-candidate + else + MY_PV=$(ver_rs 1-4 '-') # stable releases + fi + MY_P="${P}" + INHERIT_GIT="" + SRC_URI="https://download.schedmd.com/slurm/${MY_P}.tar.bz2"; + KEYWORDS="~amd64 ~arm64 ~riscv ~x86" +fi + +inherit autotools bash-completion-r1 lua-single pam perl-module prefix toolchain-funcs systemd ${INHERIT_GIT} tmpfiles + +DESCRIPTION="A Highly Scalable Resource Manager" +HOMEPAGE="https://www.schedmd.com https://github.com/SchedMD/slurm"; + +LICENSE="GPL-2" +SLOT="0" +IUSE="X debug hdf5 html ipmi json lua multiple-slurmd +munge mysql numa nvml ofed pam perl slurmdbd slurmrestd static-libs torque ucx yaml" + +# See bug #813924 for hdf5lib < dep, needs proper fix +COMMON_DEPEND=" + !sys-cluster/torque + !net-analyzer/slurm + !net-analyzer/sinfo + || ( sys-cluster/pmix[-pmi] >=sys-cluster/openmpi-2.0.0 ) + mysql? ( + || ( dev-db/mariadb-connector-c dev-db/mysql-connector-c ) + slurmdbd? ( || ( dev-db/mariadb:* dev-db/mysql:* ) ) + ) + slurmrestd? ( + net-libs/http-parser + ) + munge? ( sys-auth/munge ) + pam? ( sys-libs/pam ) + lua? ( ${LUA_DEPS} ) + ipmi? ( sys-libs/freeipmi ) + json? ( dev-libs/json-c:= ) + hdf5? ( <sci-libs/hdf5-1.12:= ) + numa? ( sys-process/numactl ) + nvml? ( dev-util/nvidia-cuda-toolkit x11-drivers/nvidia-drivers ) + ofed? ( sys-cluster/rdma-core ) + ucx? ( sys-cluster/ucx ) + yaml? ( dev-libs/libyaml ) + X? ( net-libs/libssh2 ) + >=sys-apps/hwloc-1.1.1-r1:= + sys-libs/ncurses:0= + app-arch/lz4:0= + dev-libs/glib:2= + sys-apps/dbus + sys-libs/readline:0=" + +DEPEND="${COMMON_DEPEND} + html? ( sys-apps/man2html )" + +BDEPEND="acct-user/slurm + acct-group/slurm" + +RDEPEND="${COMMON_DEPEND} + dev-libs/libcgroup" + +REQUIRED_USE="lua? ( ${LUA_REQUIRED_USE} ) + torque? ( perl ) + slurmrestd? ( json ) " + +S="${WORKDIR}/${MY_P}" + +LIBSLURM_PERL_S="${S}/contribs/perlapi/libslurm/perl" +LIBSLURMDB_PERL_S="${S}/contribs/perlapi/libslurmdb/perl" + +RESTRICT="test" + +pkg_setup() { + use lua && lua-single_pkg_setup +} + +src_unpack() { + if [[ ${PV} == *9999* ]]; then + git-r3_src_unpack + else + default + fi +} + +src_prepare() { + tc-ld-disable-gold + default + + # pids should go to /var/run/slurm + sed \ + -e 's:/tmp:/var/tmp:g' \ + -e "s:/var/run/slurmctld.pid:${EPREFIX}/run/slurm/slurmctld.pid:g" \ + -e "s:/var/run/slurmd.pid:${EPREFIX}/run/slurm/slurmd.pid:g" \ + -e "s:StateSaveLocation=.*:StateSaveLocation=${EPREFIX}/var/spool/slurm:g" \ + -e "s:SlurmdSpoolDir=.*:SlurmdSpoolDir=${EPREFIX}/var/spool/slurm/slurmd:g" \ + -i "${S}/etc/slurm.conf.example" \ + || die "Can't sed for /var/run/slurmctld.pid" + sed \ + -e "s:/var/run/slurmdbd.pid:${EPREFIX}/run/slurm/slurmdbd.pid:g" \ + -i "${S}/etc/slurmdbd.conf.example" \ + || die "Can't sed for /var/run/slurmdbd.pid" + # gentooify systemd services + sed \ + -e 's:sysconfig/.*:conf.d/slurm:g' \ + -e 's:var/run/:run/slurm/:g' \ + -e '/^EnvironmentFile=.*/d' \ + -i "${S}/etc"/*.service.in \ + || die "Can't sed systemd services for sysconfig or var/run/" + + sed -e '/AM_PATH_GTK_2_0/d' -i configure.ac || die + + hprefixify auxdir/{ax_check_zlib,x_ac_{lz4,ofed,munge}}.m4 + eautoreconf +} + +src_configure() { + local myconf=( + CPPFLAGS="-I/opt/cuda/include" + --sysconfdir="${EPREFIX}/etc/${PN}" + --with-hwloc="${EPREFIX}/usr" + --htmldir="${EPREFIX}/usr/share/doc/${PF}" + ) + use pam && myconf+=( --with-pam_dir=$(getpam_mod_dir) ) + use mysql || myconf+=( --without-mysql_config ) + econf "${myconf[@]}" \ + $(use_enable debug) \ + $(use_with lua) \ + $(use_enable pam) \ + $(use_enable X x11) \ + $(use_with munge) \ + $(use_with json) \ + $(use_with hdf5) \ + $(use_with nvml) \ + $(use_with ofed) \ + $(use_with ucx) \ + $(use_with yaml) \ + $(use_enable static-libs static) \ + $(use_enable slurmrestd) \ + $(use_enable multiple-slurmd) + + # --htmldir does not seems to propagate... Documentations are installed + # in /usr/share/doc/slurm-2.3.0/html + # instead of /usr/share/doc/slurm-2.3.0.2/html + sed \ + -e "s|htmldir = .*/html|htmldir = \${prefix}/share/doc/slurm-${PVR}/html|g" \ + -i doc/html/Makefile || die + if use perl ; then + # small hack to make it compile + mkdir -p "${S}/src/api/.libs" || die + mkdir -p "${S}/src/db_api/.libs" || die + touch "${S}/src/api/.libs/libslurm.so" || die + touch "${S}/src/db_api/.libs/libslurmdb.so" || die + cd "${LIBSLURM_PERL_S}" || die + S="${LIBSLURM_PERL_S}" SRC_PREP="no" perl-module_src_configure + cd "${LIBSLURMDB_PERL_S}" || die + S="${LIBSLURMDB_PERL_S}" SRC_PREP="no" perl-module_src_configure + cd "${S}" || die + rm -rf "${S}/src/api/.libs" "${S}/src/db_api/.libs" || die + fi +} + +src_compile() { + default + use pam && emake -C contribs/pam + if use perl ; then + cd "${LIBSLURM_PERL_S}" || die + S="${LIBSLURM_PERL_S}" perl-module_src_compile + cd "${LIBSLURMDB_PERL_S}" || die + S="${LIBSLURMDB_PERL_S}" perl-module_src_compile + cd "${S}" || die + fi + use torque && emake -C contribs/torque +} + +src_install() { + default + use pam && emake DESTDIR="${D}" -C contribs/pam install + if use perl; then + cd "${LIBSLURM_PERL_S}" || die + S="${LIBSLURM_PERL_S}" perl-module_src_install + cd "${LIBSLURMDB_PERL_S}" || die + S="${LIBSLURMDB_PERL_S}" perl-module_src_install + cd "${S}" || die + fi + if use torque; then + emake DESTDIR="${D}" -C contribs/torque + rm -f "${D}"/usr/bin/mpiexec || die + fi + use static-libs || find "${ED}" -name '*.la' -exec rm {} + + # install sample configs + keepdir /etc/slurm + insinto /etc/slurm + doins \ + etc/prolog.example \ + etc/cgroup.conf.example \ + etc/slurm.conf.example \ + etc/slurmdbd.conf.example + exeinto /etc/slurm + keepdir /etc/slurm/layouts.d + # install init.d files + newinitd "$(prefixify_ro "${FILESDIR}/slurmd.initd")" slurmd + newinitd "$(prefixify_ro "${FILESDIR}/slurmctld.initd")" slurmctld + newinitd "$(prefixify_ro "${FILESDIR}/slurmdbd.initd")" slurmdbd + # install conf.d files + newconfd "${FILESDIR}/slurm.confd" slurm + # install logrotate file + insinto /etc/logrotate.d + newins "${FILESDIR}/logrotate" slurm + # install bashcomp + newbashcomp contribs/slurm_completion_help/slurm_completion.sh scontrol + bashcomp_alias scontrol \ + sreport sacctmgr squeue scancel sshare sbcast sinfo \ + sprio sacct salloc sbatch srun sattach sdiag sstat + # install systemd files + newtmpfiles "${FILESDIR}/slurm.tmpfiles" slurm.conf + systemd_dounit etc/slurmd.service etc/slurmctld.service etc/slurmdbd.service + + paths=( + /var/${PN}/checkpoint + /var/${PN} + /var/spool/${PN}/slurmd + /var/spool/${PN} + /var/log/${PN} + /var/tmp/${PN}/${PN}d + /var/tmp/${PN} + /run/${PN} + ) + local folder_path + for folder_path in ${paths[@]}; do + keepdir ${folder_path} + fowners ${PN}:${PN} ${folder_path} + done +} + +pkg_preinst() { + if use munge; then + sed -i 's,\(SLURM_USE_MUNGE=\).*,\11,' "${D}"/etc/conf.d/slurm || die + fi +} + +pkg_postinst() { + tmpfiles_process slurm.conf + + elog "Please visit the file '/usr/share/doc/${P}/html/configurator.html" + elog "through a (javascript enabled) browser to create a configureation file." + elog "Copy that file to /etc/slurm/slurm.conf on all nodes (including the headnode) of your cluster." + echo + elog "For cgroup support, please see https://www.schedmd.com/slurmdocs/cgroup.conf.html"; + elog "Your kernel must be compiled with the wanted cgroup feature:" + elog " For the proctrack plugin:" + elog " freezer" + elog " For the task plugin:" + elog " cpuset, memory, devices" + elog " For the accounting plugin:" + elog " cpuacct, memory, blkio" + elog "Then, set these options in /etc/slurm/slurm.conf:" + elog " ProctrackType=proctrack/cgroup" + elog " TaskPlugin=task/cgroup" + einfo + ewarn "Paths were created for slurm. Please use these paths in /etc/slurm/slurm.conf:" + for folder_path in ${paths[@]}; do + ewarn " ${folder_path}" + done +}
