Release v0.2.0

The changes in this release cover general improvements to the scripts running the policy engine. Here is a full list of changes:

Major changes:

• Added getopt functionality allowing the user to specify the following variables in the function call:
  • outdir: parent directory for the log to be placed in
  • output-log-prefix: the prefix to add to the policy file name to specify extra metadata
  • policy-file: path to the policy file
  • device: path or name of the target of the policy
• Removed use of environment variables not explicitly set in the run-mmpol.sh script. These were also removed from the submit-pol-job script and replaced with the corresponding values to be explicitly passed
• Default policy file prefix is automatically created from the device path or name if not passed as an argument
• Set user:group ownership to $USER:atlab.
• Set permissions
  • log_dir: 2770 to cause subdirectories and files to be group writable and inherit the atlab group automatically
  • log_dir/raw: 1550 to set the sticky bit so the raw log cannot be deleted except by root and the raw directory cannot be changed
  • log_dir/raw/log.gz: 440
• Automatically gzip the log file
• Implements changes suggested in #17 (closed). run-mmpol now creates a single directory to store the raw policy output, the split gz files, and the parquet files plus anything else associated with the run. The directory name matches the name of the log file.

Minor changes:

• Changed some variable names to maintain consistency with related scripts
• Full device path is used in the prefix, changing / to -. All metadata entries in the file/directory names are separated by _ for clarity
• Added dry-run option to all 3 scripts for testing. Dry runs submit no jobs and make no changes to any directory structure.
• Changed methods for finding the raw log once it's created and moving/renaming it
• Removed default device from run-mmpol.sh. Now the device must be specified regardless of which script is run.

Syntax Changes:

• Encased all called variables in "{}" for clarity in variable substitution.
• Replaced ```` syntax with $() when performing command substitution

Issues Fixed:

• Fixes #30 (closed)
• Fixes #26 (closed)
• Fixes #25 (closed)
• Fixes #24 (closed)
• Fixes #17 (closed)

src/run-policy/run-mmpol.sh

@@ -4,43 +4,143 @@ set -euxo pipefail
 
 # run an mmapply policy across the cluster via slurm
 
-# gather info to map mmapplypolicy to runtime configuration
-# arguments passed via job env and runtime context
+############################################################
+# Default Values                                           #
+############################################################
 
-filesystem=${FILESYSTEM:-scratch}
-policyfile=$POLICYFILE
-tmpglobal=$DIR/slurm-tmp-${SLURM_JOBID}
-tmpscratch=$DIR/slurm-tmp-${SLURM_JOBID}
-mkdir -p $tmpglobal
+outdir="/data/rc/gpfs-policy/data"
+policy_file="./policy-def/list-path-external"
+output_log_prefix=""
+dry_run=""
 
-nodes=`scontrol show hostnames "${SLURM_JOB_NODELIST}" | tr '\n' ',' | sed -e 's/,$//'`
+############################################################
+# Help                                                     #
+############################################################
+
+usage()
+{
+>&2 cat << EOF
+Usage: $0 [ -h ] [ -o | --outdir ] [ -f | --output-prefix ] [ -P | --policy-file] device
+EOF
+exit 1
+}
+
+help()
+{
+>&2 cat << EOF
+Runs mmapplypolicy on the specified device/fileset. The policy file dictates the actions performed including list, delete, add, etc. This is most often called by the submit-pol-job wrapper instead of invoked directly
+
+Usage: $0 [ -h ] [ -o | --outdir ] [ -f | --output-prefix ] [ -P | --policy-file ] device
+
+options:
+    -h|--help           Print this Help.
+    --dry-run           Do not run the policy command or any command that modifies any files or directories such as 
+                            mkdir
+
+Required:
+    device              GPFS fileset/directory apply the policy to. Can be 
+                            specified as either the name of the fileset or the 
+                            full path to the directory 
+                            (Examples: scratch, /data/user/[username])
+
+Path:   
+    -o|--outdir         Parent directory to save policy output to 
+                            (default: /data/rc/gpfs-policy/data)
+    -f|--output-prefix  Prefix of the policy output file. Appended with a metadata string containing the policy name, 
+                            job ID, and date
+
+Policy Options:
+    -P|--policy-file    Path to policy file to apply to the given GPFS device
+EOF
+exit 0
+}
+
+args=$(getopt -a -o ho:f:P: --long help,outdir:,output-prefix:,policy-file:,dry-run -- "$@")
+
+if [[ $? -gt 0 ]]; then
+  usage
+fi
+
+eval set -- ${args}
+
+while :
+do
+  case $1 in
+    -h | --help)            help                    ;;
+    -o | --outdir)          outdir=$2               ; shift 2 ;;
+    -f | --output-prefix)   output_log_prefix=$2    ; shift 2 ;;
+    -P | --policy-file)     policy_file=$2          ; shift 2 ;;
+    --dry-run)              dry_run=true            ; shift 1 ;;
+    --) shift; break ;;
+    *) >&2 echo Unsupported option: $1
+       usage ;;
+  esac
+done
+
+if [[ $# -eq 0 ]]; then
+  usage
+fi
+
+device="$1"
+
+# Ensure device is specified
+if [[ -z "${device}" ]]; then
+    echo "Error: Specify either the name of a fileset or a directory path"
+    usage
+fi
+
+# set default output_log_prefix if not specified in the arguments
+if [[ -z "${output_log_prefix}" ]]; then
+    modified_device=$(echo "${device}" | sed -e 's|^/||' -e 's|/$||' -e 's|/|-|g')
+    output_log_prefix="list-policy_${modified_device}"
+fi
+
+# create temporary working directory for list aggregation
+tmpglobal="${outdir}/slurm-tmp-${SLURM_JOBID}"
+tmpscratch="${outdir}/slurm-tmp-${SLURM_JOBID}"
+
+nodes=$(scontrol show hostnames "${SLURM_JOB_NODELIST}" | tr '\n' ',' | sed -e 's/,$//')
 cores="${SLURM_CPUS_PER_TASK}"
 
-DATESTR=`date +'%Y-%m-%d-%H:%M:%S'`
+DATESTR=$(date +'%Y-%m-%dT%H:%M:%S')
 
-policy=`basename $policyfile`
+policy=$(basename ${policy_file})
 filetag="${policy}_slurm-${SLURM_JOBID}_${DATESTR}"
 
-cmd="mmapplypolicy ${filesystem} -I defer \
-  -P $policyfile \
-  -g $tmpglobal \
-  -s $tmpscratch \
-  -f ${DIR}/list-${SLURM_JOBID} \
-  -M FILEPATH=${filesystem} \
+cmd="mmapplypolicy ${device} -I defer \
+  -P ${policy_file} \
+  -g ${tmpglobal} \
+  -s ${tmpscratch} \
+  -f ${outdir}/list-${SLURM_JOBID} \
+  -M FILEPATH=${device} \
   -M JOBID=${SLURM_JOBID} \
-  -M LIST_OUTPUT_FILE=${OUTFILE:-/tmp/gpfs-list-policy}
+  -M LIST_OUTPUT_FILE=${output_log_prefix} \
   -N ${nodes} -n ${cores} -m ${cores}"
 
-# report final command in job log
-echo $cmd
+if [[ ! ${dry_run} ]]; then
+    mkdir -p ${tmpglobal}
+    
+    # run policy command
+    ${cmd}
+
+    log_name="${output_log_prefix}_${filetag}"
+    log_dir="${outdir}/${log_name}"
 
-# run policy command
-$cmd
+    mkdir -p ${log_dir}/raw
+    chmod 2770 ${log_dir}
 
-# tag output file with run metadata
-outfile=`ls -t $tmpglobal | head -1`
-if [[ "$outfile" != "" ]]
-then
-   mv -n $tmpglobal/$outfile $tmpglobal/../${outfile}_$filetag
+    # tag output file with run metadata
+    raw_log_file=$(find ${outdir} -maxdepth 1 -name "list-${SLURM_JOBID}*" -type f | head -1)
+    if [[ "$raw_log_file" != "" ]]; then
+        mv -n ${raw_log_file} ${log_dir}/raw/${log_name}
+        gzip ${log_dir}/raw/${log_name}
+
+        chmod 440 ${log_dir}/raw/${log_name}.gz
+        chmod 1550 ${log_dir}/raw
+    fi
+
+    chown -R ${USER}:atlab ${log_dir}
+    
+    rmdir ${tmpglobal}
 fi
-rmdir $tmpglobal
+