Merge pull request #850 from casparvl/use_local_reframe_config_file

Replace the use of a ReFrame template config file for a manually created one

Author: laraPPr

bot/test.sh

@@ -172,6 +172,10 @@ EESSI_SOFTWARE_SUBDIR_OVERRIDE=${EESSI_SOFTWARE_SUBDIR_OVERRIDE:-${CPU_TARGET}}
 export EESSI_SOFTWARE_SUBDIR_OVERRIDE
 echo "bot/test.sh: EESSI_SOFTWARE_SUBDIR_OVERRIDE='${EESSI_SOFTWARE_SUBDIR_OVERRIDE}'"
 
+# determine accelerator target (if any) from .architecture in ${JOB_CFG_FILE}
+export EESSI_ACCELERATOR_TARGET=$(cfg_get_value "architecture" "accelerator")
+echo "bot/test.sh: EESSI_ACCELERATOR_TARGET='${EESSI_ACCELERATOR_TARGET}'"
+
 # get EESSI_OS_TYPE from .architecture.os_type in ${JOB_CFG_FILE} (default: linux)
 EESSI_OS_TYPE=$(cfg_get_value "architecture" "os_type")
 export EESSI_OS_TYPE=${EESSI_OS_TYPE:-linux}
@@ -219,6 +223,9 @@ declare -a TEST_SUITE_ARGS=()
 if [[ ${EESSI_SOFTWARE_SUBDIR_OVERRIDE} =~ .*/generic$ ]]; then
     TEST_SUITE_ARGS+=("--generic")
 fi
+if [[ ${SHARED_FS_PATH} ]]; then
+    TEST_SUITE_ARGS+=("--shared-fs-path" "${SHARED_FS_PATH}")
+fi
 # [[ ! -z ${BUILD_LOGS_DIR} ]] && TEST_SUITE_ARGS+=("--build-logs-dir" "${BUILD_LOGS_DIR}")
 # [[ ! -z ${SHARED_FS_PATH} ]] && TEST_SUITE_ARGS+=("--shared-fs-path" "${SHARED_FS_PATH}")
 

reframe_config_bot.py.tmpl

@@ -132,56 +132,42 @@ else
 fi
 
 # Configure ReFrame, see https://www.eessi.io/docs/test-suite/installation-configuration
-export RFM_CONFIG_FILES=$TOPDIR/reframe_config_bot.py
-export RFM_CONFIG_FILE_TEMPLATE=$TOPDIR/reframe_config_bot.py.tmpl
+# RFM_CONFIG_FILES _has_ to be set by the site hosting the bot, so that it knows where to find the ReFrame
+# config file that matches the bot config. See https://gitlab.com/eessi/support/-/issues/114#note_2293660921
+if [ -z "$RFM_CONFIG_FILES" ]; then
+    if [ -z "${shared_fs_path}" ]; then
+        fatal_error "Environment variable 'shared_fs_path' was expected, but was not set"
+    fi
+    # Try to find a config file at $shared_fs_path/reframe_config.py
+    export RFM_CONFIG_FILES="${shared_fs_path}/reframe_config.py"
+    if [ ! -f "${RFM_CONFIG_FILES}" ]; then
+        # If we haven't found the ReFrame config, print an informative error
+        err_msg="Please put a ReFrame configuration file in ${shared_fs_path}/reframe_config.py"
+        err_msg="${err_msg} or set RFM_CONFIG_FILES in the environment of this bot instance to point to a valid"
+        err_msg="${err_msg} ReFrame configuration file that matches the bot config."
+        err_msg="${err_msg} For more information, see https://gitlab.com/eessi/support/-/issues/114#note_2293660921"
+        fatal_error "${err_msg}"
+    fi
+fi
 export RFM_CHECK_SEARCH_PATH=$TESTSUITEPREFIX/eessi/testsuite/tests
 export RFM_CHECK_SEARCH_RECURSIVE=1
 export RFM_PREFIX=$PWD/reframe_runs
 
-echo "Configured reframe with the following environment variables:"
-env | grep "RFM_"
-
-# The /sys inside the container is not the same as the /sys of the host
-# We want to extract the memory limit from the cgroup on the host (which is typically set by SLURM).
-# Thus, bot/test.sh bind-mounts the host's /sys/fs/cgroup into /hostsys/fs/cgroup
-# and that's the prefix we use to extract the memory limit from
-cgroup_v1_mem_limit="/hostsys/fs/cgroup/memory/$(</proc/self/cpuset)/memory.limit_in_bytes"
-cgroup_v2_mem_limit="/hostsys/fs/cgroup/$(</proc/self/cpuset)/memory.max"
-if [ -f "$cgroup_v1_mem_limit" ]; then
-    echo "Getting memory limit from file $cgroup_v1_mem_limit"
-    cgroup_mem_bytes=$(cat "$cgroup_v1_mem_limit")
-elif [ -f "$cgroup_v2_mem_limit" ]; then
-    echo "Getting memory limit from file $cgroup_v2_mem_limit"
-    cgroup_mem_bytes=$(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
-else
-    fatal_error "Both files ${cgroup_v1_mem_limit} and ${cgroup_v2_mem_limit} couldn't be found. Failed to get the memory limit from the current cgroup"
+# Get the correct partition name
+REFRAME_PARTITION_NAME=${EESSI_SOFTWARE_SUBDIR//\//_}
+if [ ! -z "$EESSI_ACCELERATOR_TARGET" ]; then
+    REFRAME_PARTITION_NAME=${REFRAME_PARTITION_NAME}_${EESSI_ACCELERATOR_TARGET//\//_}
 fi
-if [[ $? -eq 0 ]]; then
-    # Convert to MiB
-    cgroup_mem_mib=$(("$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 "Constructed partition name based on EESSI_SOFTWARE_SUBDIR and EESSI_ACCELERATOR_TARGET: ${REFRAME_PARTITION_NAME}"
 
-cp ${RFM_CONFIG_FILE_TEMPLATE} ${RFM_CONFIG_FILES}
-echo "Replacing memory limit in the ReFrame config file with the detected CGROUP memory limit: ${cgroup_mem_mib} MiB"
-sed -i "s/__MEM_PER_NODE__/${cgroup_mem_mib}/g" $RFM_CONFIG_FILES
-RFM_PARTITION="${SLURM_JOB_PARTITION}"
-echo "Replacing partition name in the template ReFrame config file: ${RFM_PARTITION}"
-sed -i "s/__RFM_PARTITION__/${RFM_PARTITION}/g" $RFM_CONFIG_FILES
+# Set the reframe system name, including partition
+export RFM_SYSTEM="BotBuildTests:${REFRAME_PARTITION_NAME}"
+
+echo "Configured reframe with the following environment variables:"
+env | grep "RFM_"
 
 # Make debugging easier by printing the final config file:
-echo "Final config file (after replacements):"
+echo "ReFrame config file used:"
 cat "${RFM_CONFIG_FILES}"
 
 # Workaround for https://github.com/EESSI/software-layer/pull/467#issuecomment-1973341966
@@ -217,7 +203,18 @@ else
     fatal_error "Failed to extract names of tests to run: ${REFRAME_NAME_ARGS}"
     exit ${test_selection_exit_code}
 fi
-export REFRAME_ARGS="--tag CI --tag 1_node --nocolor ${REFRAME_NAME_ARGS}"
+# Allow people deploying the bot to overrwide this
+if [ -z "$REFRAME_SCALE_TAG" ]; then
+    REFRAME_SCALE_TAG="--tag 1_node"
+fi
+if [ -z "$REFRAME_CI_TAG" ]; then
+    REFRAME_CI_TAG="--tag CI"
+fi
+# Allow bot-deployers to add additional args through the environment
+if [ -z "$REFRAME_ADDITIONAL_ARGS" ]; then
+    REFRAME_ADDITIONAL_ARGS=""
+fi
+export REFRAME_ARGS="${REFRAME_CI_TAG} ${REFRAME_SCALE_TAG} ${REFRAME_ADDITIONAL_ARGS} --nocolor ${REFRAME_NAME_ARGS}"
 
 # List the tests we want to run
 echo "Listing tests: reframe ${REFRAME_ARGS} --list"