waitforjobs

#!/usr/bin/env bash
# reimplement waitforjobs as a shell script to avoid having to source stuff
# 20230119WF - init

# this script will be among those. %a is command with arguements
parent_processes(){ ps --no-headers -o "%a" --ppid $PPID; }
dt_msg(){ echo "$(date +%FT%H:%M): $*"; }
waitforjobs_usage(){
 cat <<HERE
USAGE:
  waitforjobs [-j numjobs] [-s sleeptimesecs] [-c "auto"] [-h|--help]"

SYNOPSIS:
  wait until fewer than MAXJOBS are running.
  check the number of jobs every SLEEPTIME

  useful when not using SLURM/TORQUE but just bash forks with '&'
   sleep 3 &
   sleep 3 &
   waitforjobs -j 1 -s 1
  will wait until all jobs are finished
   # 2023-03-15T13:45: sleep 1 on 2/1 procs: sleep 3; sleep 3;

OPTIONS:
  -j --maxjobs    set number of threads before waiting
  -s --sleeptime  set time for 'sleep' command
  -c/--config     config file that can be resourced to update jobs and sleep
  --nightjobs     set njobs to run after work hours
                  if "auto", uses /tmp/host-user-basename.jobcfg
  -h --help       this message

ENVIRONMENT:
  MAXJOBS      if -j not specified
  SLEEPTIME    if -s not specified

DEFAULTS:
  2 jobs and 60s sleeps

EXAMPLES
    waitforjobs -j 2 -s 60s

    waitforfjobs --maxjobs 2 --sleeptime 60s

    export SLEEPTIME=60s MAXJOBS=2
    waitforjobs

    MAXJOBS=2 waitforjobs

HERE
}
initial_config(){
  echo "maxjobs=${maxjobs:-2}"
  echo "nightjobs=${nightjobs:-2}"
  echo "sleeptime=${sleeptime:-60s}"
}
reload_config(){
   config="$1"
   if [[ $config = "auto" ]]; then
      parent_name=$(ps -cq $PPID -o command=|sed 's/[^a-zA-Z0-9@_-]//')
      config="${TMPDIR:-/tmp}/$HOSTNAME-$USER-$parent_name.jobcfg"
      [ ! -r "$config" ] &&
         initial_config > "$config" &&
         warn "# wrote waitforjobs config to $config"
   fi
   if [ ! -r "$config" ]; then
     echo "warn 'watiforjobs config $config does not exist!' && exit 2"
     return 2
   fi
   # only grab lines that look like things we care about
   # in case file has something accidentally malicious
   grep -P '^(maxjobs|sleeptime|nightjobs)=' "$config"
}
is_after_hours() {
   # after 6 or before 8
   [ $# -gt 0 ] && hour="${1}" || hour=$(date +%H)
   test "$hour" -gt 18 -o "$hour" -lt 8
}
waiting(){
   sleeptime=${SLEEPTIME:-60s}
   maxjobs=${MAXJOBS:-2}
   nightjobs=${maxjobs}
   config=""

   while [ $# -gt 0 ]; do
      case "$1" in
         -j|--maxjobs) maxjobs=$2; shift 2;;
         -s|--sleeptime) sleeptime=$2; shift 2;;
         -c|--config) config="$2"; shift 2;;
         --nightjobs) nightjobs="$2"; shift 2;;
         -h|--help) waitforjobs_usage; exit 0;;
         *) warn "'$1' is bad option! see 'waitforjobs --help'";  exit 1;;
      esac
   done

   while mapfile -t procs < <(parent_processes) &&
      # greater (not equal)  b/c expect parent process (waitforjobs itself) to be in the count
      # min $#procs is 1
      nproc=${#procs[@]}
      [ -n "$config" ] && eval "$(reload_config "$config")"

      # job limit based on time of day
      # nightjobs default to maxjobs when not expliclty set
      njoblimit="$maxjobs"
      is_after_hours && njoblimit="$nightjobs"

      [ "$nproc" -gt "$njoblimit" ]; do
         max_disp=$(("$nproc" - 1))
         # assume last is always parent
         dt_msg "sleep $sleeptime on ${nproc}/$max_disp procs: $(printf "%s; " "${procs[@]::$nproc-1}")"
      sleep "$sleeptime"
   done
}

eval "$(iffmain waiting)"