Merge pull request #1 from EESSI/2023.06-software.eessi.io

Pull Last merge from Upstream

Author: MaKaNu

easystacks/software.eessi.io/2023.06/a64fx/eessi-2023.06-eb-4.9.2-2023a.yml

@@ -4,3 +4,17 @@ easyconfigs:
   - SciPy-bundle-2023.07-gfbf-2023a.eb
   - ESPResSo-4.2.2-foss-2023a.eb
   - ParaView-5.11.2-foss-2023a.eb
+  - OpenFOAM-10-foss-2023a.eb:
+      options:
+        # see https://github.com/easybuilders/easybuild-easyconfigs/pull/20958
+        from-commit: dbadb2074464d816740ee0e95595c2cb31b6338f
+  - OpenFOAM-11-foss-2023a.eb:
+      options:
+        # see https://github.com/easybuilders/easybuild-easyconfigs/pull/20958
+        from-commit: dbadb2074464d816740ee0e95595c2cb31b6338f
+  - OpenFOAM-v2312-foss-2023a.eb:
+      options:
+        # https://github.com/easybuilders/easybuild-easyblocks/pull/3388
+        include-easyblocks-from-commit: c8256a36e7062bc09f5ce30552a9de9827054c9e
+        # https://github.com/easybuilders/easybuild-easyconfigs/pull/20841
+        from-commit: f0e91e6e430ebf902f7788ebb47f0203dee60649

easystacks/software.eessi.io/2023.06/eessi-2023.06-eb-4.9.2-2023b.yml

@@ -15,3 +15,7 @@ easyconfigs:
       options:
         # see https://github.com/easybuilders/easybuild-easyconfigs/pull/21034
         from-commit: 76e7fc6657bab64bfbec826540a3a8f0040258f2
+  - STAR-2.7.11b-GCC-13.2.0.eb:
+      options:
+        # see https://github.com/easybuilders/easybuild-easyconfigs/pull/21200
+        from-commit: 765ba900daf5953e306c4dad896febe52fdd6c00

easystacks/software.eessi.io/2023.06/zen4/eessi-2023.06-eb-4.9.2-001-system.yml

@@ -0,0 +1,2 @@
+easyconfigs:
+    - ReFrame-4.3.3.eb

eessi_container.sh

@@ -254,7 +254,7 @@ if [[ ${LIST_REPOS} -eq 1 ]]; then
             default_label=", default"
         else
             default_label=""
-	fi
+        fi
         echo "    ${cvmfs_repo} [CVMFS config repo${default_label}]"
     done
     for cfg_repo in "${!cfg_cvmfs_repos[@]}"
@@ -323,7 +323,7 @@ do
     if [[ ! -n "${eessi_cvmfs_repos[${cvmfs_repo_name}]}" ]] ; then
         [[ ${VERBOSE} -eq 1 ]] && echo "repo '${cvmfs_repo_name}' is not an EESSI CVMFS repository..."
         # cvmfs_repo_name is actually a repository ID, use that to obtain
-	#   the actual name from the EESSI_REPOS_CFG_FILE
+        #   the actual name from the EESSI_REPOS_CFG_FILE
         cfg_repo_id=${cvmfs_repo_name}
         cvmfs_repo_name=$(cfg_get_value ${cfg_repo_id} "repo_name")
     fi
@@ -595,11 +595,11 @@ do
     #   that the necessary information for accessing a CVMFS repository is made
     #   available inside the container
     if [[ -n "${cfg_cvmfs_repos[${cvmfs_repo_name}]}" ]] ; then
-	cfg_repo_id=${cvmfs_repo_name}
+        cfg_repo_id=${cvmfs_repo_name}
 
-	# obtain CVMFS repository name from section for the given ID
+        # obtain CVMFS repository name from section for the given ID
         cfg_repo_name=$(cfg_get_value ${cfg_repo_id} "repo_name")
-	# derive domain part from (cfg_)repo_name (everything after first '.')
+        # derive domain part from (cfg_)repo_name (everything after first '.')
         repo_name_domain=${repo_name#*.}
 
         # cfg_cvmfs_repos is populated through reading the file pointed to by
@@ -609,15 +609,15 @@ do
         # copy repos.cfg to job directory --> makes it easier to inspect the job
         cp -a ${EESSI_REPOS_CFG_FILE} ${EESSI_TMPDIR}/repos_cfg/.
 
-	# cfg file should include sections (one per CVMFS repository to be mounted)
-	#   with each section containing the settings:
-	#   - repo_name,
-	#   - repo_version,
-	#   - config_bundle, and
-	#   - a map { filepath_in_bundle -> container_filepath }
+        # cfg file should include sections (one per CVMFS repository to be mounted)
+        #   with each section containing the settings:
+        #   - repo_name,
+        #   - repo_version,
+        #   - config_bundle, and
+        #   - a map { filepath_in_bundle -> container_filepath }
         #
-	# The config_bundle includes the files which are mapped ('->') to a target
-	# location in container:
+        # The config_bundle includes the files which are mapped ('->') to a target
+        # location in container:
         # - default.local -> /etc/cvmfs/default.local
         #   contains CVMFS settings, e.g., CVMFS_HTTP_PROXY, CVMFS_QUOTA_LIMIT, ...
         # - ${repo_name_domain}.conf -> /etc/cvmfs/domain.d/${repo_name_domain}.conf
@@ -641,7 +641,7 @@ do
         # use information to set up dir ${EESSI_TMPDIR}/repos_cfg and define
         #   BIND mounts
         # check if config_bundle exists, if so, unpack it into
-	#   ${EESSI_TMPDIR}/repos_cfg; if it doesn't, exit with an error
+        #   ${EESSI_TMPDIR}/repos_cfg; if it doesn't, exit with an error
         # if config_bundle is relative path (no '/' at start) prepend it with
         #   EESSI_REPOS_CFG_DIR
         config_bundle_path=
@@ -726,7 +726,7 @@ do
     if [[ ${cfg_cvmfs_repos[${cvmfs_repo_name}]} ]]; then
         [[ ${VERBOSE} -eq 1 ]] && echo "repo '${cvmfs_repo_name}' is not an EESSI CVMFS repository..."
         # cvmfs_repo_name is actually a repository ID, use that to obtain
-	#   the actual name from the EESSI_REPOS_CFG_FILE
+        #   the actual name from the EESSI_REPOS_CFG_FILE
         cfg_repo_id=${cvmfs_repo_name}
         cvmfs_repo_name=$(cfg_get_value ${cfg_repo_id} "repo_name")
     fi
@@ -736,15 +736,52 @@ do
 
     # add fusemount options depending on requested access mode ('ro' - read-only; 'rw' - read & write)
     if [[ ${cvmfs_repo_access} == "ro" ]] ; then
-        export EESSI_READONLY="container:cvmfs2 ${cvmfs_repo_name} /cvmfs/${cvmfs_repo_name}"
+        # need to distinguish between basic "ro" access and "ro" after a "rw" session
+        if [[ -d ${EESSI_TMPDIR}/${cvmfs_repo_name}/overlay-upper ]]; then
+            # the overlay-upper directory is only created in a read-write-session, thus
+            # we are resuming from such a session here (otherwise there shouldn't be such
+            # directory yet as it is only created for read-write-sessions a bit further
+            # below); the overlay-upper directory can only exist because it is part of
+            # the ${RESUME} directory or tarball
+            # to be able to see the contents of the read-write session we have to mount
+            # the fuse-overlayfs (in read-only mode) on top of the CernVM-FS repository
+
+            echo "While processing '${cvmfs_repo_name}' to be mounted 'read-only' we detected an overlay-upper"
+            echo "  directory (${EESSI_TMPDIR}/${cvmfs_repo_name}/overlay-upper) likely from a previous"
+            echo "  session. Will use it as left-most directory in 'lowerdir' argument for fuse-overlayfs."
+
+            # make the target CernVM-FS repository available under /cvmfs_ro
+            export EESSI_READONLY="container:cvmfs2 ${cvmfs_repo_name} /cvmfs_ro/${cvmfs_repo_name}"
+
+            EESSI_FUSE_MOUNTS+=("--fusemount" "${EESSI_READONLY}")
+
+            # now, put the overlay-upper read-only on top of the repo and make it available under the usual prefix /cvmfs
+            EESSI_READONLY_OVERLAY="container:fuse-overlayfs"
+            # The contents of the previous session are available under
+            #   ${EESSI_TMPDIR} which is bind mounted to ${TMP_IN_CONTAINER}.
+            #   Hence, we have to use ${TMP_IN_CONTAINER}/${cvmfs_repo_name}/overlay-upper
+            # the left-most directory given for the lowerdir argument is put on top,
+            #   and with no upperdir=... the whole overlayfs is made available read-only
+            EESSI_READONLY_OVERLAY+=" -o lowerdir=${TMP_IN_CONTAINER}/${cvmfs_repo_name}/overlay-upper:/cvmfs_ro/${cvmfs_repo_name}"
+            EESSI_READONLY_OVERLAY+=" /cvmfs/${cvmfs_repo_name}"
+            export EESSI_READONLY_OVERLAY
+
+            EESSI_FUSE_MOUNTS+=("--fusemount" "${EESSI_READONLY_OVERLAY}")
+            export EESSI_FUSE_MOUNTS
+        else
+            # basic "ro" access that doesn't require any fuseoverlay-fs
+            echo "Mounting '${cvmfs_repo_name}' 'read-only' without fuse-overlayfs."
 
-        EESSI_FUSE_MOUNTS+=("--fusemount" "${EESSI_READONLY}")
-        export EESSI_FUSE_MOUNTS
+            export EESSI_READONLY="container:cvmfs2 ${cvmfs_repo_name} /cvmfs/${cvmfs_repo_name}"
+
+            EESSI_FUSE_MOUNTS+=("--fusemount" "${EESSI_READONLY}")
+            export EESSI_FUSE_MOUNTS
+        fi
     elif [[ ${cvmfs_repo_access} == "rw" ]] ; then
         # use repo-specific overlay directories
         mkdir -p ${EESSI_TMPDIR}/${cvmfs_repo_name}/overlay-upper
         mkdir -p ${EESSI_TMPDIR}/${cvmfs_repo_name}/overlay-work
-	[[ ${VERBOSE} -eq 1 ]] && echo -e "TMP directory contents:\n$(ls -l ${EESSI_TMPDIR})"
+        [[ ${VERBOSE} -eq 1 ]] && echo -e "TMP directory contents:\n$(ls -l ${EESSI_TMPDIR})"
 
         # set environment variables for fuse mounts in Singularity container
         export EESSI_READONLY="container:cvmfs2 ${cvmfs_repo_name} /cvmfs_ro/${cvmfs_repo_name}"
@@ -762,7 +799,7 @@ do
         export EESSI_FUSE_MOUNTS
     else
         echo -e "ERROR: access mode '${cvmfs_repo_access}' for CVMFS repository\n  '${cvmfs_repo_name}' is not known"
-	exit ${REPOSITORY_ERROR_EXITCODE}
+        exit ${REPOSITORY_ERROR_EXITCODE}
     fi
     # create repo_settings.sh file in ${EESSI_TMPDIR}/${cvmfs_repo_name} to store
     #   (intention is that the file could be just sourced to obtain the settings)

test_suite.sh

@@ -74,11 +74,17 @@ fi
 TMPDIR=$(mktemp -d)
 
 echo ">> Setting up environment..."
-module --force purge
-export EESSI_SOFTWARE_SUBDIR_OVERRIDE=$(python3 $TOPDIR/eessi_software_subdir.py $DETECTION_PARAMETERS)
+# For this call to be succesful, it needs to be able to import archspec (which is part of EESSI)
+# Thus, we execute it in a subshell where EESSI is already initialized (a bit like a bootstrap)
+export EESSI_SOFTWARE_SUBDIR_OVERRIDE=$(source $TOPDIR/init/bash > /dev/null 2>&1; python3 $TOPDIR/eessi_software_subdir.py $DETECTION_PARAMETERS)
+echo "EESSI_SOFTWARE_SUBDIR_OVERRIDE: $EESSI_SOFTWARE_SUBDIR_OVERRIDE"
 
 source $TOPDIR/init/bash
 
+# We have to ignore the LMOD cache, otherwise the software that is built in the build step cannot be found/loaded
+# Reason is that the LMOD cache is normally only updated on the Stratum 0, once everything is ingested
+export LMOD_IGNORE_CACHE=1
+
 # Load the ReFrame module
 # Currently, we load the default version. Maybe we should somehow make this configurable in the future?
 module load ReFrame
@@ -136,40 +142,66 @@ echo "Configured reframe with the following environment variables:"
 env | grep "RFM_"
 
 # Inject correct CPU/memory properties into the ReFrame config file
+echo "Collecting system-specific input for the ReFrame configuration file"
 cpuinfo=$(lscpu)
 if [[ "${cpuinfo}" =~ CPU\(s\):[^0-9]*([0-9]+) ]]; then
     cpu_count=${BASH_REMATCH[1]}
+    echo "Detected CPU count: ${cpu_count}"
 else
     fatal_error "Failed to get the number of CPUs for the current test hardware with lscpu."
 fi
 if [[ "${cpuinfo}" =~ Socket\(s\):[^0-9]*([0-9]+) ]]; then
     socket_count=${BASH_REMATCH[1]}
+    echo "Detected socket count: ${socket_count}"
 else
     fatal_error "Failed to get the number of sockets for the current test hardware with lscpu."
 fi
 if [[ "${cpuinfo}" =~ (Thread\(s\) per core:[^0-9]*([0-9]+)) ]]; then
     threads_per_core=${BASH_REMATCH[2]}
+    echo "Detected threads per core: ${threads_per_core}"
 else
     fatal_error "Failed to get the number of threads per core for the current test hardware with lscpu."
 fi
 if [[ "${cpuinfo}" =~ (Core\(s\) per socket:[^0-9]*([0-9]+)) ]]; then
     cores_per_socket=${BASH_REMATCH[2]}
+    echo "Detected cores per socket: ${cores_per_socket}"
 else
     fatal_error "Failed to get the number of cores per socket for the current test hardware with lscpu."
 fi
-cgroup_mem_bytes=$(cat /hostsys/fs/cgroup/memory/slurm/uid_${UID}/job_${SLURM_JOB_ID}/memory.limit_in_bytes)
+cgroup_v1_mem_limit="/sys/fs/cgroup/memory/$(</proc/self/cpuset)/memory.limit_in_bytes"
+cgroup_v2_mem_limit="/sys/fs/cgroup/$(</proc/self/cpuset)/memory.max"
+if [ -f "$cgroup_v1_mem_limit" ]; then
+    cgroup_mem_bytes=$(cat "$cgroup_v1_mem_limit")
+else
+    cgroup_mem_bytes=$(cat "$cgroup_v2_mem_limit")
+    if [ "$cgroup_mem_bytes" = 'max' ]; then
+        # In cgroupsv2, the memory.max file may contain 'max', meaning the group can use the full system memory
+        # Here, we get the system memory from /proc/meminfo. Units are supposedly always in kb, but lets match them too
+        cgroup_mem_kilobytes=$(grep -oP 'MemTotal:\s+\K\d+(?=\s+kB)' /proc/meminfo)
+        if [[ $? -ne 0 ]] || [[ -z "$cgroup_mem_kilobytes" ]]; then
+            fatal_error "Failed to get memory limit from /proc/meminfo"
+        fi
+        cgroup_mem_bytes=$(("$cgroup_mem_kilobytes"*1024))
+    fi
+fi
 if [[ $? -eq 0 ]]; then
     # Convert to MiB
-    cgroup_mem_mib=$((cgroup_mem_bytes/(1024*1024)))
+    cgroup_mem_mib=$(("$cgroup_mem_bytes"/(1024*1024)))
 else
     fatal_error "Failed to get the memory limit in bytes from the current cgroup"
 fi
+echo "Detected available memory: ${cgroup_mem_mib} MiB"
+
+echo "Replacing detected system information in template ReFrame config file..."
 cp ${RFM_CONFIG_FILE_TEMPLATE} ${RFM_CONFIG_FILES}
 sed -i "s/__NUM_CPUS__/${cpu_count}/g" $RFM_CONFIG_FILES
 sed -i "s/__NUM_SOCKETS__/${socket_count}/g" $RFM_CONFIG_FILES
 sed -i "s/__NUM_CPUS_PER_CORE__/${threads_per_core}/g" $RFM_CONFIG_FILES
 sed -i "s/__NUM_CPUS_PER_SOCKET__/${cores_per_socket}/g" $RFM_CONFIG_FILES
 sed -i "s/__MEM_PER_NODE__/${cgroup_mem_mib}/g" $RFM_CONFIG_FILES
+# Make debugging easier by printing the final config file:
+echo "Final config file (after replacements):"
+cat "${RFM_CONFIG_FILES}"
 
 # Workaround for https://github.com/EESSI/software-layer/pull/467#issuecomment-1973341966
 export PSM3_DEVICES='self,shm'  # this is enough, since we only run single node for now