BUGFIX: Change ownership to executing USER and group to atlab. Also automatically gzip the log output

This is a continuing revision for !17 (merged) and !18 (closed). Changes from !18 (closed) have been added here for convenience in resolving merge conflicts.

Major changes:

• Set user:group ownership to $USER:atlab.
• Set permissions to rws for the containing directory and r-- for the log itself
• 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:

• 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 #26 (closed)
• Fixes #17 (closed)

src/run-policy/run-mmpol.sh

@@ -97,34 +97,134 @@ tmpglobal=$outdir/slurm-tmp-${SLURM_JOBID}
 tmpscratch=$outdir/slurm-tmp-${SLURM_JOBID}
 mkdir -p $tmpglobal
 
-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 $policy_file`
+policy=$(basename ${policy_file})
 filetag="${policy}_slurm-${SLURM_JOBID}_${DATESTR}"
 
 cmd="mmapplypolicy ${device} -I defer \
-  -P $policy_file \
-  -g $tmpglobal \
-  -s $tmpscratch \
+  -P ${policy_file} \
+  -g ${tmpglobal} \
+  -s ${tmpscratch} \
   -f ${outdir}/list-${SLURM_JOBID} \
   -M FILEPATH=${device} \
   -M JOBID=${SLURM_JOBID} \
-  -M LIST_OUTPUT_FILE=${output_prefix} \
+  -M LIST_OUTPUT_FILE=${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}
 
-# run policy command
-$cmd
+    log_name="${output_log_prefix}_${filetag}"
+    log_dir="${outdir}/${log_name}"
 
-# tag output file with run metadata
-outfile=`ls -t $tmpglobal | head -1`
-if [[ "$outfile" != "" ]]
-then
-   mv -n $tmpglobal/$outfile $tmpglobal/../${output_log_prefix}_$filetag
+    mkdir -p ${log_dir}/raw
+    chmod 1770 ${log_dir}
+
+    # tag output file with run metadata
+    raw_log_file=$(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 550 ${log_dir}/raw
+    fi
+
+    chown -R ${USER}:atlab ${log_dir}
+    
+    rmdir ${tmpglobal}
 fi
-rmdir $tmpglobal
+