contrib/balance_pool

#!/bin/bash

cd $(dirname $(readlink -f $0))/..

DRYRUN=echo
NUM=
SPARES=suites
HELP=0
while getopts 'hn:xs:' OPT; do
  case $OPT in
    n) NUM=$OPTARG ;;
    x) DRYRUN= ;;
    s) SPARES=$OPTARG ;;
    \?|h) HELP=1 ;;
  esac
done
shift $(( OPTIND - 1 ))

if [ $HELP -ne 0 -o $# -ne 2 ]; then
  echo "usage: $(basename $0) [options] <board> <pool>" >&2
  echo "Options:" >&2
  echo "    -n #: adjust pool size to #" >&2
  echo "    -x: execute changes (default is to print changes only)" >&2
  echo "    -s <spares>: specify pool for spares (default: suites)" >&2
  exit 1
fi


# List atest information for all DUTs in a pool.
#
# Arguments:  The name of a pool
# Standard output: `atest host list` output for the given pool, with
#   the initial header line removed.
list_pool() {
  cli/atest host list -w cautotest -b pool:$1,board:$BOARD | sed 1d
}


# Find all DUTs in a pool
#
# Arguments:  The name of a pool
# Standard output: All DUTs in the pool with the user-specified
#   board.
list_duts() {
  list_pool $1 | awk '{ print $1 }'
}

# Find DUTs in a pool and not shared with any other pool.
#
# Arguments:  The name of a pool
# Standard output: All DUTs that are in that pool, and in no other.
list_unshared_duts() {
  list_pool $1 | awk '$0 !~ /pool:.*pool:/ {print $1}'
}


# Find all working DUTs in a list
#
# Arguments:  A list of DUTs
# Standard output:  The subset of given DUTs that are working
filter_working() {
  python site_utils/dut_status.py "$@" | awk '/OK/ {print $1}'
}


# Exchange a list of DUTs between two pools
#
# Arguments:
#   1: Pool to be augmented with DUTs
#   2: Pool from which to take the DUTs
#   remainder are DUTs to be exchanged
#
# For a dry run, this function prints the exchange commands rather
# than executing them.
exchange() {
  local to=$1
  local from=$2
  shift 2
  [ $# -gt 0 ] || return 0
  local list=$(echo "$@" | sed 's/ /,/g')
  $DRYRUN cli/atest label remove -m $list pool:$from
  $DRYRUN cli/atest label add -m $list pool:$to
}

BOARD=$1
POOL=$2

POOL_ALL=( $(list_duts $POOL) )
POOL_WORKING=( $(filter_working "${POOL_ALL[@]}") )
NWORKING=${#POOL_WORKING[@]}

if [ -z "$NUM" ]; then
  NUM=${#POOL_ALL[@]}
fi

REMOVE=()
REPLACEMENTS=()


# OK, command line parsing and setup are done.  We're operating in
# three steps here.
#
# Step 1. Find all the broken devices.  These will be the first we
# remove from $POOL.  Accumulate devices to be removed in the array
# `REMOVE`.
#
widx=0
tidx=0
while [ $tidx -lt ${#POOL_ALL[@]} ]; do
  if [ "${POOL_ALL[$tidx]}" != "${POOL_WORKING[$widx]}" ]; then
    REMOVE=( "${REMOVE[@]}" ${POOL_ALL[$tidx]} )
  else
    widx=$(( widx + 1 ))
  fi
  tidx=$(( tidx + 1 ))
done


# Step 2. Determine whether we have enough working devices, and what
# replacements, if any, we'll take.
#
if [ $NWORKING -gt $NUM ]; then
  # Step 2a.  More working devices than we need.  Add excess devices
  # into the `REMOVE` array.
  COUNT=$(( $NWORKING - $NUM - 1 ))
  for i in $(seq 0 $COUNT); do
    REMOVE=( "${REMOVE[@]}" ${POOL_WORKING[$i]} )
  done
elif [ $NWORKING -lt $NUM ]; then
  # Step 2b.  Not enough working devices; try to find spares.
  # Accumulate available spares into the `REPLACEMENTS` array.
  NEED=$(( NUM - NWORKING ))
  REPLACEMENTS=( $(filter_working $(list_unshared_duts $SPARES)) )
  if [ $NEED -gt ${#REPLACEMENTS[@]} ]; then
    # Step 2b-1.  Not enough working spares.  Print a warning, then
    # make sure we only remove as many broken devices as we have
    # spares to replace them.
    if [ ${#REPLACEMENTS[@]} -ne 0 ]; then
      SPARE_MSG="only ${#REPLACEMENTS[@]}"
    else
      SPARE_MSG="no"
    fi
    echo "WARN: Board $BOARD, $SPARE_MSG spares for $NEED needed DUTs" >&2
    while [ ${#REMOVE[@]} -gt ${#REPLACEMENTS[@]} ]; do
      unset REMOVE[$(( ${#REMOVE[@]} - 1 ))]
    done
  else
    # Step 2b-2.  We have enough working spares.  Remove excess spares.
    while [ ${#REPLACEMENTS[@]} -gt $NEED ]; do
      unset REPLACEMENTS[$(( ${#REPLACEMENTS[@]} - 1 ))]
    done
  fi
fi


# Step 3.  If it's a dry run, report some information.  Then,
# execute the pool exchanges.
#
# Make the dry run messages look like comments so that the output of
# this command is acceptable as input to the shell.
#
if [ -n "$DRYRUN" ]; then
  echo "# targeting $NUM working DUTs in $POOL; currently have $NWORKING"
  echo "# ${#REMOVE[@]} removals; ${#REPLACEMENTS[@]} spares"
  for h in "${REMOVE[@]}"; do
    echo "# remove $h"
  done
  for h in "${REPLACEMENTS[@]}"; do
    echo "# spare $h"
  done
fi

exchange $SPARES $POOL "${REMOVE[@]}"
exchange $POOL $SPARES "${REPLACEMENTS[@]}"