Fix test_suite not being able to retrieve memory on systems that use cgroups v2

test_suite.sh

@@ -74,8 +74,10 @@ 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
 
@@ -136,40 +138,66 @@ 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