fs/skeleton/bin/babysit

#!/bin/sh
RETRY_TIMEOUT=$1

if [ -e /proc/$$/oom_score_adj ] ; then
  OOM_FILE=oom_score_adj
  echo -500 >/proc/$$/$OOM_FILE
else
  OOM_FILE=oom_adj
  echo -17 >/proc/$$/$OOM_FILE
fi

shift
if [ -z "$RETRY_TIMEOUT" -o -z "$*" ]; then
  echo "Usage: $0 <retry_timeout> <program> [args...]" >&2
  exit 1
fi

progname() {
  local first="$1"
  while [ $# -gt 0 ]; do
    case $1 in
      alivemonitor) shift 5 ;;
      babysit) shift 2 ;;
      startpid) shift 2 ;;
      deathrattle) shift 2 ;;
      *) echo "$1"; return ;;
    esac
  done
  echo "$first"
}

run() {
  "$@" &
  local pid=$!
  echo 0 >/proc/$pid/$OOM_FILE
  wait $pid
}

error() {
  echo "Error: '$name' exited with code $RV.  Retry in $RETRY_TIMEOUT secs."
}

# Run until the program manages to exit successfully.
name=$(progname "$@")
TMP_LOG=/tmp/babysit_errors_$(basename "$name")
until run "$@"; do
  RV=$?
  if [ "$RV" = 143 ]; then
    # SIGTERM means someone asked for it to die.
    # We can't treat SIGKILL the same way because it might have been
    # caused by the kernel OOM killer, in which case we definitely want to
    # restart the task.
    echo "SIGTERM: '$name' was killed explicitly.  Not retrying." >&2
    exit $RV
  fi
  error >&2
  error > $TMP_LOG
  sleep $RETRY_TIMEOUT || sleep 1  # just in case $RETRY_TIMEOUT is invalid
done
echo "'$name' exited successfully.  Not restarting." >&2
exit 0