Merge branch '2023.06-software.eessi.io' of github.com:EESSI/software…

…-layer into riscv

.github/workflows/scripts/only_latest_easystacks.sh

@@ -0,0 +1,58 @@
+#!/bin/bash
+#
+# This script figures out the latest version of EasyBuild being used for the installation of easystack
+# files.
+#
+# This file is part of the EESSI software layer, see
+# https://github.com/EESSI/software-layer.git
+#
+# author: Alan O'Cais (CECAM)
+#
+# license: GPLv2
+#
+
+EESSI_VERSION=${EESSI_VERSION:-"2023.06"}
+
+directory="easystacks/software.eessi.io/${EESSI_VERSION}"
+# List of example filenames
+files=($(find "$directory" -name "*.yml" | grep -e '-eb-'))
+[ -n "$DEBUG" ] && echo "${files[@]}"
+
+versions=()
+# Loop over each filename
+for filename in "${files[@]}"; do
+    # Extract the semantic version using grep
+    version=$(echo "$filename" | grep -oP '(?<=eb-)\d+\.\d+\.\d+?(?=-)')
+
+    # Output the result
+    [ -n "$DEBUG" ] && echo "Filename: $filename"
+    [ -n "$DEBUG" ] && echo "Extracted version: $version"
+    [ -n "$DEBUG" ] && echo
+    versions+=("$version")
+done
+highest_version=$(printf "%s\n" "${versions[@]}" | sort -V | tail -n 1)
+
+[ -n "$DEBUG" ] && echo "Highest version: $highest_version"
+[ -n "$DEBUG" ] && echo
+[ -n "$DEBUG" ] && echo "Matching files:"
+all_latest_easystacks=($(find $directory -type f -name "*eb-$highest_version*.yml"))
+
+accel_latest_easystacks=()
+cpu_latest_easystacks=()
+
+# Loop through the array and split based on partial matching of string
+accel="/accel/"
+for item in "${all_latest_easystacks[@]}"; do
+  if [[ "$item" == *"$accel"* ]]; then
+    accel_latest_easystacks+=("$item")
+  else
+    cpu_latest_easystacks+=("$item")
+  fi
+done
+
+# Output the results
+if [ -n "$ACCEL_EASYSTACKS" ]; then
+    echo "${accel_latest_easystacks[@]}"
+else
+    echo "${cpu_latest_easystacks[@]}"
+fi

.github/workflows/test-software.eessi.io.yml

@@ -57,15 +57,20 @@ jobs:
               env | grep ^EESSI | sort
 
               # first check the CPU-only builds for this CPU target
-              echo "just run check_missing_installations.sh (should use easystacks/software.eessi.io/${{matrix.EESSI_VERSION}}/eessi-${{matrix.EESSI_VERSION}}-*.yml)"
-              for easystack_file in $(ls easystacks/software.eessi.io/${{matrix.EESSI_VERSION}}/eessi-${{matrix.EESSI_VERSION}}-eb-*.yml); do
+              echo "just run check_missing_installations.sh (should use easystacks/software.eessi.io/${{matrix.EESSI_VERSION}}/eessi-${{matrix.EESSI_VERSION}}-*.yml with latest EasyBuild release)"
+              for easystack_file in $(EESSI_VERSION=${{matrix.EESSI_VERSION}} .github/workflows/scripts/only_latest_easystacks.sh); do
                   if [ ${{matrix.EESSI_SOFTWARE_SUBDIR_OVERRIDE}} = "x86_64/amd/zen4" ]; then
                       if grep -q 2022b <<<"${easystack_file}"; then
-                          # skip the check of installed software on zen4 for foss/2022b builds
-                          continue
-                      elif grep -q CUDA <<<"${easystack_file}"; then
-                          # skip the check of install CUDA software in the CPU path for zen4
-                          continue
+                        # skip the check of installed software on zen4 for foss/2022b builds
+                        continue
+                      fi
+                      if [[ $easystack_file == *"rebuilds"* ]]; then
+                        # Also handle rebuilds, make a temporary EasyStack file where we clean out all 2022b stuff and use that
+                        new_easystack=$(mktemp pruned_easystackXXX --suffix=.yml)
+                        # first clean out the options then clean out the .eb name
+                        sed '/2022b\|12\.2\.0/,/\.eb/{/\.eb/!d}' "${easystack_file}" | sed '/2022b\|12\.2\.0/d' > $new_easystack
+                        diff --unified=0 "$easystack_file" "$new_easystack" || :
+                        easystack_file="$new_easystack"
                       fi
                   fi
                   echo "check missing installations for ${easystack_file}..."
@@ -82,7 +87,7 @@ jobs:
                   for accel in ${accelerators}; do
                       module use ${EESSI_SOFTWARE_PATH}/accel/${accel}/modules/all
                       echo "checking missing installations for accelerator ${accel} using modulepath: ${MODULEPATH}"
-                      for easystack_file in $(ls easystacks/software.eessi.io/${{matrix.EESSI_VERSION}}/accel/$(dirname ${accel})/eessi-${{matrix.EESSI_VERSION}}-eb-*.yml); do
+                      for easystack_file in $(EESSI_VERSION=${{matrix.EESSI_VERSION}} ACCEL_EASYSTACKS=1 .github/workflows/scripts/only_latest_easystacks.sh); do
                           echo "check missing installations for ${easystack_file}..."
                           ./check_missing_installations.sh ${easystack_file}
                           ec=$?

EESSI-extend-2023.06-easybuild.eb

@@ -88,6 +88,16 @@ if (os.getenv("EESSI_CVMFS_INSTALL") ~= nil) then
   end
   eessi_cvmfs_install = true
   easybuild_installpath = os.getenv("EESSI_SOFTWARE_PATH")
+  eessi_accelerator_target = os.getenv("EESSI_ACCELERATOR_TARGET")
+  if (eessi_accelerator_target ~= nil) then
+    cuda_compute_capability = string.match(eessi_accelerator_target, "^nvidia/cc([0-9][0-9])$")
+    if (cuda_compute_capability ~= nil) then
+      easybuild_installpath = pathJoin(easybuild_installpath, 'accel', eessi_accelerator_target)
+      easybuild_cuda_compute_capabilities = cuda_compute_capability:sub(1, 1) ..  "." .. cuda_compute_capability:sub(2, 2)
+    else
+      LmodError("Incorrect value for $EESSI_ACCELERATOR_TARGET: " .. eessi_accelerator_target)
+    end
+  end
 elseif (os.getenv("EESSI_SITE_INSTALL") ~= nil) then
   -- Make sure no other EESSI install environment variables are set
   if ((os.getenv("EESSI_PROJECT_INSTALL") ~= nil) or (os.getenv("EESSI_USER_INSTALL") ~= nil)) then
@@ -147,6 +157,11 @@ setenv ("EASYBUILD_UMASK", "022")
 -- Allow this module to be loaded when running EasyBuild
 setenv ("EASYBUILD_ALLOW_LOADED_MODULES", "EasyBuild,EESSI-extend")
 
+-- Set environment variables if building for CUDA compute capabilities
+if (easybuild_cuda_compute_capabilities ~= nil) then
+  setenv ("EASYBUILD_CUDA_COMPUTE_CAPABILITIES", easybuild_cuda_compute_capabilities)
+end
+
 -- Set all related environment variables if we have project or user installations (including extending MODULEPATH)
 if (user_modulepath ~= nil) then
   -- Use a more restrictive umask for this case

EESSI-install-software.sh

@@ -229,22 +229,34 @@ if [[ "${EESSI_CVMFS_REPO}" != /cvmfs/dev.eessi.io ]]; then
     ${TOPDIR}/install_scripts.sh --prefix ${EESSI_PREFIX}
 fi
 
+echo ">> Configuring EasyBuild..."
+
+# Make sure EESSI-extend is not loaded, and configure location variables for a
+#   CVMFS installation
+module unload EESSI-extend
+unset EESSI_USER_INSTALL
+unset EESSI_PROJECT_INSTALL
+unset EESSI_SITE_INSTALL
+export EESSI_CVMFS_INSTALL=1
+
+# We now run 'source load_eessi_extend_module.sh' to load or install and load the
+#   EESSI-extend module which sets up all build environment settings.
+# The script requires the EESSI_VERSION given as argument, a couple of
+#   environment variables set (TMPDIR, EB and EASYBUILD_INSTALLPATH) and the
+#   function check_exit_code defined.
+# NOTE 1, the script exits if those variables/functions are undefined.
+# NOTE 2, loading the EESSI-extend module may adjust the value of EASYBUILD_INSTALLPATH,
+#   e.g., to point to the installation directory for accelerators.
+# NOTE 3, we have to set a default for EASYBUILD_INSTALLPATH here in cases the
+#   EESSI-extend module itself needs to be installed.
+export EASYBUILD_INSTALLPATH=${EESSI_PREFIX}/software/${EESSI_OS_TYPE}/${EESSI_SOFTWARE_SUBDIR_OVERRIDE}
+source load_eessi_extend_module.sh ${EESSI_VERSION}
+
 # Install full CUDA SDK and cu* libraries in host_injections
 # Hardcode this for now, see if it works
 # TODO: We should make a nice yaml and loop over all CUDA versions in that yaml to figure out what to install
 # Allow skipping CUDA SDK install in e.g. CI environments
-# The install_cuda... script uses EasyBuild. So, we need to check if we have EB
-# or skip this step.
 echo "Going to install full CUDA SDK and cu* libraries under host_injections if necessary"
-module_avail_out=$TMPDIR/ml.out
-module avail 2>&1 | grep EasyBuild &> ${module_avail_out}
-if [[ $? -eq 0 ]]; then
-    echo_green ">> Found an EasyBuild module"
-else
-    echo_yellow ">> No EasyBuild module found: skipping step to install CUDA (see output in ${module_avail_out})"
-    export skip_cuda_install=True
-fi
-
 temp_install_storage=${TMPDIR}/temp_install_storage
 mkdir -p ${temp_install_storage}
 if [ -z "${skip_cuda_install}" ] || [ ! "${skip_cuda_install}" ]; then
@@ -253,7 +265,7 @@ if [ -z "${skip_cuda_install}" ] || [ ! "${skip_cuda_install}" ]; then
         --accept-cuda-eula \
         --accept-cudnn-eula
 else
-    echo "Skipping installation of CUDA SDK and cu* libraries in host_injections, since the --skip-cuda-install flag was passed OR no EasyBuild module was found"
+    echo "Skipping installation of CUDA SDK and cu* libraries in host_injections, since the --skip-cuda-install flag was passed"
 fi
 
 # Install NVIDIA drivers in host_injections (if they exist)
@@ -263,18 +275,6 @@ if command_exists "nvidia-smi"; then
 fi
 
 
-echo ">> Configuring EasyBuild..."
-
-# Make sure that we use the EESSI_CVMFS_INSTALL
-# Since the path is set when loading EESSI-extend, we reload it to make sure it works - even if it is already loaded
-# Note we need to do this after running install_cuda_and_libraries, since that does installations in the EESSI_SITE_INSTALL 
-unset EESSI_USER_INSTALL
-unset EESSI_PROJECT_INSTALL
-unset EESSI_SITE_INSTALL
-export EESSI_CVMFS_INSTALL=1
-module unload EESSI-extend
-module load EESSI-extend/${EESSI_VERSION}-easybuild
-
 if [ ! -z "${shared_fs_path}" ]; then
     shared_eb_sourcepath=${shared_fs_path}/easybuild/sources
     echo ">> Using ${shared_eb_sourcepath} as shared EasyBuild source path"

EESSI-remove-software.sh

@@ -129,6 +129,9 @@ if [ $EUID -eq 0 ]; then
                     echo_yellow "Removing ${app_dir} and ${app_module}..."
                     rm -rf ${app_dir}
                     rm -rf ${app_module}
+                    # recreate some directory to work around permission denied
+                    # issues when rebuilding the package
+                    mkdir -p ${app_dir}/easybuild
                 done
             else
                 fatal_error "Easystack file ${easystack_file} not found!"

bot/build.sh

@@ -208,6 +208,7 @@ else
     declare -a REMOVAL_STEP_ARGS=()
     REMOVAL_STEP_ARGS+=("--save" "${TARBALL_TMP_REMOVAL_STEP_DIR}")
     REMOVAL_STEP_ARGS+=("--storage" "${STORAGE}")
+
     # add fakeroot option in order to be able to remove software, see:
     # https://github.com/EESSI/software-layer/issues/312
     REMOVAL_STEP_ARGS+=("--fakeroot")

bot/check-build.sh

@@ -17,6 +17,7 @@
 #  - SUCCESS (all of)
 #    - working directory contains slurm-JOBID.out file
 #    - working directory contains eessi*tar.gz
+#    - no message FATAL
 #    - no message ERROR
 #    - no message FAILED
 #    - no message ' required modules missing:'
@@ -25,6 +26,7 @@
 #  - FAILED (one of ... implemented as NOT SUCCESS)
 #    - no slurm-JOBID.out file
 #    - no tarball
+#    - message with FATAL
 #    - message with ERROR
 #    - message with FAILED
 #    - message with ' required modules missing:'
@@ -105,6 +107,16 @@ else
     [[ ${VERBOSE} -ne 0 ]] && echo "   Slurm output file '"${job_out}"' NOT found"
 fi
 
+FATAL=-1
+if [[ ${SLURM_OUTPUT_FOUND} -eq 1 ]]; then
+  GP_fatal='FATAL: '
+  grep_out=$(grep -v "^>> searching for " ${job_dir}/${job_out} | grep "${GP_fatal}")
+  [[ $? -eq 0 ]] && FATAL=1 || FATAL=0
+  # have to be careful to not add searched for pattern into slurm out file
+  [[ ${VERBOSE} -ne 0 ]] && echo ">> searching for '"${GP_fatal}"'"
+  [[ ${VERBOSE} -ne 0 ]] && echo "${grep_out}"
+fi
+
 ERROR=-1
 if [[ ${SLURM_OUTPUT_FOUND} -eq 1 ]]; then
   GP_error='ERROR: '
@@ -163,6 +175,7 @@ fi
 
 [[ ${VERBOSE} -ne 0 ]] && echo "SUMMARY: ${job_dir}/${job_out}"
 [[ ${VERBOSE} -ne 0 ]] && echo "  <test name>: <actual result> (<expected result>)"
+[[ ${VERBOSE} -ne 0 ]] && echo "  FATAL......: $([[ $FATAL -eq 1 ]] && echo 'yes' || echo 'no') (no)"
 [[ ${VERBOSE} -ne 0 ]] && echo "  ERROR......: $([[ $ERROR -eq 1 ]] && echo 'yes' || echo 'no') (no)"
 [[ ${VERBOSE} -ne 0 ]] && echo "  FAILED.....: $([[ $FAILED -eq 1 ]] && echo 'yes' || echo 'no') (no)"
 [[ ${VERBOSE} -ne 0 ]] && echo "  REQ_MISSING: $([[ $MISSING -eq 1 ]] && echo 'yes' || echo 'no') (no)"
@@ -190,6 +203,7 @@ job_result_file=_bot_job${SLURM_JOB_ID}.result
 
 # Default reason:
 if [[ ${SLURM_OUTPUT_FOUND} -eq 1 ]] && \
+   [[ ${FATAL} -eq 0 ]] && \
    [[ ${ERROR} -eq 0 ]] && \
    [[ ${FAILED} -eq 0 ]] && \
    [[ ${MISSING} -eq 0 ]] && \
@@ -223,6 +237,7 @@ fi
 #     <dt>_Details_</dt>
 #     <dd>
 #       :white_check_mark: job output file <code>slurm-4682.out</code><br/>
+#       :white_check_mark: no message matching <code>FATAL: </code><br/>
 #       :white_check_mark: no message matching <code>ERROR: </code><br/>
 #       :white_check_mark: no message matching <code>FAILED: </code><br/>
 #       :white_check_mark: no message matching <code> required modules missing:</code><br/>
@@ -264,6 +279,7 @@ fi
 #     <dt>_Details_</dt>
 #     <dd>
 #       :white_check_mark: job output file <code>slurm-4682.out</code><br/>
+#       :x: no message matching <code>FATAL: </code><br/>
 #       :x: no message matching <code>ERROR: </code><br/>
 #       :white_check_mark: no message matching <code>FAILED: </code><br/>
 #       :x: no message matching <code> required modules missing:</code><br/>
@@ -381,6 +397,10 @@ success_msg="job output file <code>${job_out}</code>"
 failure_msg="no job output file <code>${job_out}</code>"
 comment_details_list=${comment_details_list}$(add_detail ${SLURM_OUTPUT_FOUND} 1 "${success_msg}" "${failure_msg}")
 
+success_msg="no message matching <code>${GP_fatal}</code>"
+failure_msg="found message matching <code>${GP_fatal}</code>"
+comment_details_list=${comment_details_list}$(add_detail ${FATAL} 0 "${success_msg}" "${failure_msg}")
+
 success_msg="no message matching <code>${GP_error}</code>"
 failure_msg="found message matching <code>${GP_error}</code>"
 comment_details_list=${comment_details_list}$(add_detail ${ERROR} 0 "${success_msg}" "${failure_msg}")

check_missing_installations.sh

@@ -25,17 +25,6 @@ easystack=$1
 
 LOCAL_TMPDIR=$(mktemp -d)
 
-# Clone the develop branch of EasyBuild and use that to search for easyconfigs
-
-if [[ -z ${EASYBUILD_ROBOT_PATHS} ]]; then
-    git clone -b develop https://github.com/easybuilders/easybuild-easyconfigs.git $LOCAL_TMPDIR/easyconfigs
-    export EASYBUILD_ROBOT_PATHS=$LOCAL_TMPDIR/easyconfigs/easybuild/easyconfigs
-fi
-
-# All PRs used in EESSI are supposed to be merged, so we can strip out all cases of from-pr
-tmp_easystack=${LOCAL_TMPDIR}/$(basename ${easystack})
-grep -v from-pr ${easystack} > ${tmp_easystack}
-
 source $TOPDIR/scripts/utils.sh
 
 source $TOPDIR/configure_easybuild
@@ -45,34 +34,11 @@ ${EB:-eb} --show-config
 
 echo ">> Checking for missing installations in ${EASYBUILD_INSTALLPATH}..."
 eb_missing_out=$LOCAL_TMPDIR/eb_missing.out
-${EB:-eb} --easystack ${tmp_easystack} --missing 2>&1 | tee ${eb_missing_out}
+${EB:-eb} --easystack ${easystack} --missing 2>&1 | tee ${eb_missing_out}
 exit_code=${PIPESTATUS[0]}
 
 ok_msg="Command 'eb --missing ...' succeeded, analysing output..."
 fail_msg="Command 'eb --missing ...' failed, check log '${eb_missing_out}'"
-if [ "$exit_code" -ne 0 ] && [ ! -z "$pr_exceptions" ]; then
-    # We might have failed due to unmerged PRs. Try to make exceptions for --from-pr added in this PR
-    # to software-layer, and see if then it passes. If so, we can report a more specific fail_msg
-    # Note that if no --from-pr's were used in this PR, $pr_exceptions will be empty and we might as
-    # well skip this check - unmerged PRs can not be the reason for the non-zero exit code in that scenario
-
-    # Let's use awk so we can allow for exceptions if we are given a PR diff file
-    awk_command="awk '\!/'from-pr'/ EXCEPTIONS' $easystack"
-    awk_command=${awk_command/\\/}  # Strip out the backslash we needed for !
-    eval ${awk_command/EXCEPTIONS/$pr_exceptions} > ${tmp_easystack}
-
-    msg=">> Checking for missing installations in ${EASYBUILD_INSTALLPATH},"
-    msg="${msg} allowing for --from-pr's that were added in this PR..."
-    echo ${msg}
-    eb_missing_out=$LOCAL_TMPDIR/eb_missing_with_from_pr.out
-    ${EB:-eb} --easystack ${tmp_easystack} --missing 2>&1 | tee ${eb_missing_out}
-    exit_code_with_from_pr=${PIPESTATUS[0]}
-
-    # If now we succeeded, the reason must be that we originally stripped the --from-pr's
-    if [ "$exit_code_with_from_pr" -eq 0 ]; then
-        fail_msg="$fail_msg (are you sure all PRs referenced have been merged in EasyBuild?)"
-    fi
-fi
 
 check_exit_code ${exit_code} "${ok_msg}" "${fail_msg}"
 

create_lmodsitepackage.py

@@ -114,7 +114,7 @@
     -- If we try to load CUDA itself, check if the full CUDA SDK was installed on the host in host_injections.
     -- This is required for end users to build additional CUDA software. If the full SDK isn't present, refuse
     -- to load the CUDA module and print an informative message on how to set up GPU support for EESSI
-    local refer_to_docs = "For more information on how to do this, see https://www.eessi.io/docs/gpu/.\\n"
+    local refer_to_docs = "For more information on how to do this, see https://www.eessi.io/docs/site_specific_config/gpu/.\\n"
     if simpleName == 'CUDA' then
         -- get the full host_injections path
         local hostInjections = string.gsub(os.getenv('EESSI_SOFTWARE_PATH') or "", 'versions', 'host_injections')

easystacks/software.eessi.io/2023.06/accel/nvidia/eessi-2023.06-eb-4.9.4-2023a-CUDA.yml

@@ -1,3 +1,7 @@
 easyconfigs:
   - CUDA-12.1.1.eb
   - cuDNN-8.9.2.26-CUDA-12.1.1.eb
+  - LightGBM-4.5.0-foss-2023a-CUDA-12.1.1.eb:
+      options:
+        # see https://github.com/easybuilders/easybuild-easyconfigs/pull/21699
+        from-commit: e3407bd127d248c08960f6b09c973da0fdecc2c3

easystacks/software.eessi.io/2023.06/accel/nvidia/eessi-2023.06-eb-4.9.4-2023b-CUDA.yml

@@ -0,0 +1,19 @@
+easyconfigs:
+  - CUDA-12.4.0.eb:
+      options:
+        accept-eula-for: CUDA
+  - UCX-CUDA-1.15.0-GCCcore-13.2.0-CUDA-12.4.0.eb
+  - UCC-CUDA-1.2.0-GCCcore-13.2.0-CUDA-12.4.0.eb:
+      options:
+        # see https://github.com/easybuilders/easybuild-easyconfigs/pull/21565
+        from-commit: 46141a3f40e699433fac03af2d3ed81bd5a62da7
+  - OSU-Micro-Benchmarks-7.5-gompi-2023b-CUDA-12.4.0.eb:
+      options:
+        # see https://github.com/easybuilders/easybuild-easyconfigs/pull/21926
+        from-commit: de79ec74eb076e1aceda5e21235a73c05ed6764c
+  - GROMACS-2024.4-foss-2023b-CUDA-12.4.0.eb:
+      options:
+        # see https://github.com/easybuilders/easybuild-easyconfigs/pull/21549
+        from-commit: 12b53341343967ce5a402fe8190a3c85bce7d49b
+        # see https://github.com/easybuilders/easybuild-easyblocks/pull/3523
+        include-easyblocks-from-commit: 90495ed23d26b3d5fd8162bf5d7b4c073a0682fe