stockholm/krebs/5pkgs/test/infest-cac-centos7/notes

# nix-shell -p gnumake jq openssh cac-api cac-panel sshpass
set -eufx

# 2 secrets are required:

krebs_cred=${krebs_cred-./cac.json}
retiolum_key=${retiolum_key-./retiolum.rsa_key.priv}

clear_defer(){
  echo "${trapstr:-exit}"
  trap - INT TERM EXIT KILL
}
defer(){
  if test -z "${debug:-}"; then
    trapstr="$1;${trapstr:-exit}"
    trap "$trapstr" INT TERM EXIT KILL
  fi
}

# Sanity
if test ! -r "$krebs_cred";then
  echo "\$krebs_cred=$krebs_cred must be readable"; exit 1
fi
if test ! -r "$retiolum_key";then
  echo "\$retiolum_key=$retiolum_key must be readable"; exit 1
fi

krebs_secrets=$(mktemp -d)
sec_file=$krebs_secrets/cac_config
krebs_ssh=$krebs_secrets/tempssh
export cac_resources_cache=$krebs_secrets/res_cache.json
export cac_servers_cache=$krebs_secrets/servers_cache.json
export cac_tasks_cache=$krebs_secrets/tasks_cache.json
export cac_templates_cache=$krebs_secrets/templates_cache.json
# we need to receive this key from buildmaster to speed up tinc bootstrap
defer "trap - INT TERM EXIT"
defer "rm -r $krebs_secrets"

cat > $sec_file <<EOF
cac_login="$(jq -r .email $krebs_cred)"
cac_key="$(cac-panel --config $krebs_cred settings | jq -r .apicode)"
EOF

export cac_secrets=$sec_file
cac-panel --config $krebs_cred add-api-ip

# test login:
cac-api update
cac-api servers

# preserve old trap
old_trapstr=$(clear_defer)
while true;do
  # Template 26: CentOS7
  # TODO: use cac-api templates to determine the real Centos7 template in case it changes
  out=$(cac-api build cpu=1 ram=512 storage=10 os=26 2>&1)
  if name=$(echo "$out" | jq -r .servername);then
    id=servername:$name
    echo "got a working machine, id=$id"
  else
    echo "Unable to build a virtual machine, retrying in 15 seconds" >&2
    echo "Output of build program: $out" >&2
    sleep 15
    continue
  fi

  clear_defer >/dev/null
  defer "cac-api delete $id"

  # TODO: timeout?

  wait_login_cac(){
    # we wait for 30 minutes
    for t in `seq 180`;do
      # now we have a working cac-api server
      if cac-api ssh $1 -o ConnectTimeout=10 \
                    cat /etc/redhat-release | \
                      grep CentOS ;then
        return 0
      fi
      sleep 10
    done
    return 1
  }
  # die on timeout
  if ! wait_login_cac $id;then
    echo "unable to boot a working system within time frame, retrying..." >&2
    echo "Cleaning up old image,last status: $(cac-api update;cac-api getserver $id | jq -r .status)"
    eval "$(clear_defer | sed 's/;exit//')"
    sleep 15
  else
    echo "got a working system" >&2
    break
  fi
done
clear_defer >/dev/null
defer "cac-api delete $id;$old_trapstr"

mkdir -p shared/2configs/temp
cac-api generatenetworking $id > \
  shared/2configs/temp/networking.nix
# new temporary ssh key we will use to log in after install
ssh-keygen -f $krebs_ssh -N ""
cp $retiolum_key $krebs_secrets/retiolum.rsa_key.priv
# we override the directories for secrets and stockholm
# additionally we set the ssh key we generated
ip=$(cac-api getserver $id | jq -r .ip)

cat > shared/2configs/temp/dirs.nix <<EOF
_: {
  krebs.build.source.dir = {
    secrets.path = "$krebs_secrets";
    stockholm.path = "$(pwd)";
  };
  users.extraUsers.root.openssh.authorizedKeys.keys = [
    "$(cat ${krebs_ssh}.pub)"
  ];
}
EOF

make install \
    LOGNAME=shared \
    SSHPASS="$(cac-api getserver $id | jq -r .rootpass)" \
    ssh='sshpass -e ssh -S none -o StrictHostKeyChecking=no -o UserKnownHostsFile=/dev/null' \
    system=test-centos7 \
    target=$ip

# TODO: generate secrets directory $krebs_secrets for nix import
cac-api powerop $id reset

wait_login(){
  # timeout
  for t in `seq 90`;do
    # now we have a working cac-api server
    if ssh -o StrictHostKeyChecking=no \
           -o UserKnownHostsFile=/dev/null \
           -i $krebs_ssh \
           -o ConnectTimeout=10 \
           -o BatchMode=yes \
           root@$1 nixos-version ;then
      return 0
    fi
    sleep 10
  done
  return 1
}
wait_login $ip