#!/bin/sh

if [ $# != 5 ]
then
	echo "usage: $0 <glexec> <proxy> <sandbox> <max-glexec-retries> <min-glexec-retry-delay>" >&2
	exit 1
fi

GLEXEC="$1"
PROXY="$2"
SANDBOX="$3"
MAX_GLEXEC_ERRORS="$4"
MIN_GLEXEC_RETRY_DELAY="$5"

# move the condor-owned sandbox aside to make way for the job's
#
mv "$SANDBOX" "$SANDBOX.condor"

# tar up the contents of the sandbox and pipe them over to
# a process running as the job owner via glexec
#
SH=`readlink -f /bin/sh`
GLEXEC_CLIENT_CERT="$SANDBOX.condor/$PROXY"
export GLEXEC_CLIENT_CERT
# prevent glexec from creating a proxy in /tmp that is not cleaned up
GLEXEC_TARGET_PROXY=/dev/null
export GLEXEC_TARGET_PROXY
glexec_errors=0
while [ 1 ]; do
  tar -C "$SANDBOX.condor" -c  . | \
      "$GLEXEC" "$SH" -c "install -m 755 -d \"$SANDBOX\" && tar -C \"$SANDBOX\" -x"
  glexec_rc=$?
  if [ $glexec_rc -eq 0 ]; then
    break
  elif [ $glexec_rc -ne 202 ] && [ $glexec_rc -ne 203 ]; then
    # Either a non-transient glexec error, or an error from the command
    # we are asking glexec to run.
    exit 2
  fi

  # This _could_ be a transient issue, such as a communication error with GUMS,
  # so retry up to some limit.
  glexec_errors=$(( $glexec_errors + 1 ))
  if [ $glexec_errors -gt $MAX_GLEXEC_ERRORS ]; then
    exit 3
  fi

  # truncated exponential backoff
  # sleep for X * min(100,1+random_number_in_range(0,glexec_errors-1))
  delay_rand=$(( (1 + 0x0$(hexdump -n 4 -e '"%4x\n"' /dev/urandom) % $glexec_errors) % 100 ))
  delay=$(( $MIN_GLEXEC_RETRY_DELAY * $delay_rand ))
  echo "Glexec exited with status $glexec_rc; retrying in $delay seconds." >&2
  sleep $delay
done

# now get rid of everything in the condor-owned sandbox except
# the proxy
#
for FILE in "$SANDBOX.condor"/* "$SANDBOX.condor"/.*
do
	if [ "$FILE" = "$SANDBOX.condor/." ]
	then
		continue
	fi
	if [ "$FILE" = "$SANDBOX.condor/.." ]
	then
		continue
	fi
	if [ "$FILE" = "$SANDBOX.condor/$PROXY" ]
	then
		continue
	fi
	rm -rf "$FILE"
done