loop reboot cloud image

README

this is just some hacky user-datas that I put together to try to catch failure as in bug 1473527 (http://pad.lv/1473527) . Doing it helped me see ppc64el image bug 1474090 (http://pad.lv/1474090)

very un-finished.

things here:

a.) reboot-loop-userdata.txt
    this reboots in a loop up to NUM_BOOTS times
    youneed to replace NUM_BOOTS

b.) go-loop
    this downloads an image, boots it and runs with logging in a loop.

go-loop

#!/bin/bash

set -e
pkgs=""
which cloud-localds >/dev/null || pkgs="$pkgs cloud-image-utils"
which qemu-system-ppc64 >/dev/null || pkgs="$pkgs qemu-system-ppc64"
which qemu-img >/dev/null || pkgs="$pkgs qemu-utils"
[ -z "${pkgs# }" ] || { echo "need install : $pkgs"; exit 1; }

rel="wily"
burl="http://cloud-images.ubuntu.com/$rel/current/"
disk_url="${1:-${burl}/$rel-server-cloudimg-amd64-disk1.img}"
disk_img="$rel.img"
disk_raw="$rel.raw"
disk_boot="disk.img"
seed_img="seed.img"

if [ ! -f "$disk_img" ]; then
   wget "$disk_url" -O "$disk_img.tmp" && mv "$disk_img.tmp" "$disk_img"
fi

if [ ! -f "$disk_raw" ]; then
   echo "converting $disk_img to $disk_raw"
   qemu-img convert -O raw "$disk_img" "$disk_raw.tmp" &&
      mv "$disk_raw.tmp" "$disk_raw"
fi

if [ -f "$HOME/.ssh/id_rsa.pub" ]; then
   keyfile=$HOME/.ssh/id_rsa.pub
else
   [ -f ./test-id_rsa.pub ] || ssh-keygen -N '' -f ./test-id_rsa
   keyfile="$PWD/test-id_rsa.pub"
fi
pubkey=$(cat $keyfile)

cat > meta-data <<EOF
instance-id: $(uuidgen || echo i-abcdefg)
public-keys:
 - "$pubkey"
EOF

boot_system() {
    local disk="$1" seed="$2" name="$3"
    qemu-system-x86_64 -enable-kvm \
       -device virtio-net-pci,netdev=net00 -netdev type=user,id=net00 \
       -drive if=virtio,file=$disk -drive if=virtio,file=$seed \
       -nographic -m 1024 ${name:+-name "$name"}
}

KID=""
cleanup() {
    kill $KID
}
trap cleanup EXIT

myvmprefix="foo-"
while :; do
   i=$(($i+1))
   log=$(printf "log-%04d" "$i")
   #cloud-localds "$seed_img" user-data meta-data
   cloud-localds "$seed_img" reboot-loop-userdata.txt meta-data
   out=$(qemu-img create -f qcow2 -b "$disk_raw" "$disk_boot" 2>&1) || { echo "failed: $out"; exit 1; }
   echo "=== $(date -R) $log ==="
   sstart="$SECONDS"
   ln -sf "$log" current.log
   name=${myvmprefix}${log#log-}
   boot_system "$disk_boot" "$seed_img" "$name" > "$log" 2>&1 &
   KID=$!
   wait
   ret=$?
   echo "=== $(date -R) $log $ret [$(($SECONDS-$sstart))s]"
   [ $ret -eq 0 ] || exit $ret
done

reboot-loop-userdata.txt

#cloud-config
##
## This user-data will boot a system NUM_BOOTS times

sm_misc:
 - &enable_reboot_or_message |
   f=/usr/local/bin/reboot-or-message
   pbdir="/var/lib/cloud/scripts/per-boot/"
   [ -f "$f" -a -x "$pbdir/${f##*/}" ] && exit 0
   [ -d "$pbdir" ] || mkdir -p "$pbdir"
   [ -e "$pbdir/${f##*/}" ] || ln -sf "$f" "$pbdir/${f##*/}"
   [ -f "$f" ] || cat > "$f" <<"EOF"
   #!/bin/bash
   me="${0##*/}"
   datafile="/$me.mark"
   tsfile="/$me.tslog"
   LOG="/var/log/$me.log"
   MSG_LOG="$LOG"
   NUM_BOOTS=2
   DELAY=3s
   PATH=$PATH:/usr/local/bin
   set -f

   set_consoles() {
      local cmdline="" tok=""
      CONSOLES=""
      read cmdline </proc/cmdline
      for tok in $cmdline; do
         [ "${tok#console=}" = "$tok" ] && continue
         tok=${tok#console=}
         tok=${tok#/dev}
         tok=${tok%%,*}
         CONSOLES="$CONSOLES /dev/$tok"
      done
      CONSOLES=${CONSOLES# }
      [ -z "$CONSOLES" ] && CONSOLES="/dev/console"
   }

   msg() {
     local c="" now=""
     now="$(date -R)"
     echo "[stdout] $now" "$@"
     for c in ${CONSOLES} ${MSG_LOG}; do
        echo "[$c] $now" "$@" >> "$c"
     done
   }

   fatal_hang() {
     echo "$$" "$1"
     sleep 100h
     exit 99
   }

   do_reboot() {
      local id="$1" delay="${2:-3}" op="${3:-reboot}"
      exec >>"$LOG" 2>&1 </dev/null
      [ $? -eq 0 ] || fatal_hang "[$id] FAIL FAIL FAIL FAILED REDIRECT"
      MSG_LOG=""
      msg "====== [$id] $op after $delay ======"
      sleep $delay || fatal_hang "[$id] SLEEP FAILED? REALLY?"
      msg "====== [$id] slept $delay $op now ====="
      "$op"
      ret=$?
      msg "====== $op returned $ret ====="
      sync
      sleep 2m &&
         fatal_hang "[$id] WHY AM I STILL AROUND"
      exit
   }

   set_consoles
   if [ ! -f "$datafile" ]; then
      msg ====== first boot. =======
      echo "1" > "$datafile"
      : > "$tsfile"
      num=1
   else
      read num < "$datafile"
      num=$(($num+1))
      echo $num > "$datafile"
   fi
   read uptime idle < /proc/uptime
   now=$(date +%s)
   printf "%04d %s %s\n" "$num" "$uptime" "$now" >> "$tsfile"
   if [ "$num" -ge "$NUM_BOOTS" ]; then
      msg "success [$num boots]"
      do_reboot "finished" "$DELAY" poweroff & </dev/null >/dev/null 2>&1
      exit 0
   fi
   do_reboot "$num" $DELAY </dev/null >/dev/null 2>&1 &
   EOF
   chmod 755 "$f"

bootcmd:
 - [sh, -c, *enable_reboot_or_message ]

password: passw0rd
chpasswd: { expire: False }
ssh_pwauth: True