tools/salt-install/provision.sh

   1 #!/bin/bash
   2
   3 # Copyright (C) The Arvados Authors. All rights reserved.
   4 #
   5 # SPDX-License-Identifier: CC-BY-SA-3.0
   6
   7 # If you want to test arvados in a single host, you can run this script, which
   8 # will install it using salt masterless
   9 # This script is run by the Vagrant file when you run it with
  10 #
  11 # vagrant up
  12
  13 set -eu
  14 set -o pipefail
  15
  16 # capture the directory that the script is running from
  17 SCRIPT_DIR="$( cd "$( dirname "${BASH_SOURCE[0]}" )" >/dev/null 2>&1 && pwd )"
  18
  19 usage() {
  20   echo >&2
  21   echo >&2 "Usage: ${0} [-h] [-h]"
  22   echo >&2
  23   echo >&2 "${0} options:"
  24   echo >&2 "  -d, --debug                                 Run salt installation in debug mode"
  25   echo >&2 "  -c <local.params>, --config <local.params>  Path to the local.params config file"
  26   echo >&2 "  -t, --test                                  Test cluster by running \`arvados-client diagnostics\`"
  27   echo >&2 "                                              and a simple workflow"
  28   echo >&2 "  -r, --roles                                 List of Arvados roles to apply to the host, comma separated"
  29   echo >&2 "                                              Possible values are:"
  30   echo >&2 "                                                balancer"
  31   echo >&2 "                                                controller"
  32   echo >&2 "                                                dispatcher"
  33   echo >&2 "                                                keepproxy"
  34   echo >&2 "                                                keepbalance"
  35   echo >&2 "                                                keepstore"
  36   echo >&2 "                                                keepweb"
  37   echo >&2 "                                                monitoring"
  38   echo >&2 "                                                shell"
  39   echo >&2 "                                                webshell"
  40   echo >&2 "                                                websocket"
  41   echo >&2 "                                                workbench"
  42   echo >&2 "                                                workbench2"
  43   echo >&2 "                                              Defaults to applying them all"
  44   echo >&2 "  -h, --help                                  Display this help and exit"
  45   echo >&2 "  --dump-config <dest_dir>                    Dumps the pillars and states to a directory"
  46   echo >&2 "                                              This parameter does not perform any installation at all. It's"
  47   echo >&2 "                                              intended to give you a parsed set of configuration files so"
  48   echo >&2 "                                              you can inspect them or use them in you Saltstack infrastructure."
  49   echo >&2 "                                              It"
  50   echo >&2 "                                                - parses the pillar and states templates,"
  51   echo >&2 "                                                - downloads the helper formulas with their desired versions,"
  52   echo >&2 "                                                - prepares the 'top.sls' files both for pillars and states"
  53   echo >&2 "                                                  for the selected role(s)"
  54   echo >&2 "                                                - writes the resulting files into <dest_dir>"
  55   echo >&2 "  -v, --vagrant                               Run in vagrant and use the /vagrant shared dir"
  56   echo >&2 "  --development                               Run in dev mode, using snakeoil certs"
  57   echo >&2
  58 }
  59
  60 arguments() {
  61   # NOTE: This requires GNU getopt (part of the util-linux package on Debian-based distros).
  62   if ! which getopt > /dev/null; then
  63     echo >&2 "GNU getopt is required to run this script. Please install it and re-reun it"
  64     exit 1
  65   fi
  66
  67   TEMP=$(getopt -o c:dhp:r:tv \
  68     --long config:,debug,development,dump-config:,help,roles:,test,vagrant \
  69     -n "${0}" -- "${@}")
  70
  71   if [ ${?} != 0 ];
  72     then echo "Please check the parameters you entered and re-run again"
  73     exit 1
  74   fi
  75   # Note the quotes around `$TEMP': they are essential!
  76   eval set -- "$TEMP"
  77
  78   while [ ${#} -ge 1 ]; do
  79     case ${1} in
  80       -c | --config)
  81         CONFIG_FILE=${2}
  82         shift 2
  83         ;;
  84       -d | --debug)
  85         LOG_LEVEL="debug"
  86         shift
  87         set -x
  88         ;;
  89       --dump-config)
  90         if [[ ${2} = /* ]]; then
  91           DUMP_SALT_CONFIG_DIR=${2}
  92         else
  93           DUMP_SALT_CONFIG_DIR=${PWD}/${2}
  94         fi
  95         ## states
  96         S_DIR="${DUMP_SALT_CONFIG_DIR}/salt"
  97         ## formulas
  98         F_DIR="${DUMP_SALT_CONFIG_DIR}/formulas"
  99         ## pillars
 100         P_DIR="${DUMP_SALT_CONFIG_DIR}/pillars"
 101         ## tests
 102         T_DIR="${DUMP_SALT_CONFIG_DIR}/tests"
 103         DUMP_CONFIG="yes"
 104         shift 2
 105         ;;
 106       --development)
 107         DEV_MODE="yes"
 108         shift 1
 109         ;;
 110       -r | --roles)
 111         for i in ${2//,/ }
 112           do
 113             # Verify the role exists
 114             if [[ ! "database,balancer,controller,keepstore,websocket,keepweb,workbench2,webshell,keepbalance,keepproxy,shell,workbench,dispatcher,monitoring" == *"$i"* ]]; then
 115               echo "The role '${i}' is not a valid role"
 116               usage
 117               exit 1
 118             fi
 119             ROLES="${ROLES:-} ${i}"
 120           done
 121           shift 2
 122         ;;
 123       -t | --test)
 124         TEST="yes"
 125         shift
 126         ;;
 127       -v | --vagrant)
 128         VAGRANT="yes"
 129         shift
 130         ;;
 131       --)
 132         shift
 133         break
 134         ;;
 135       *)
 136         usage
 137         exit 1
 138         ;;
 139     esac
 140   done
 141 }
 142
 143 copy_custom_cert() {
 144   cert_dir=${1}
 145   cert_name=${2}
 146
 147   mkdir -p --mode=0700 /srv/salt/certs
 148
 149   if [ -f ${cert_dir}/${cert_name}.crt ]; then
 150     install --mode=0600 ${cert_dir}/${cert_name}.crt /srv/salt/certs/arvados-${cert_name}.pem
 151   else
 152     echo "${cert_dir}/${cert_name}.crt does not exist. Exiting"
 153     exit 1
 154   fi
 155   if [ -f ${cert_dir}/${cert_name}.key ]; then
 156     install --mode=0600 ${cert_dir}/${cert_name}.key /srv/salt/certs/arvados-${cert_name}.key
 157   else
 158     echo "${cert_dir}/${cert_name}.key does not exist. Exiting"
 159     exit 1
 160   fi
 161 }
 162
 163 apply_var_substitutions() {
 164   local SRCFILE=$1
 165   local DSTFILE=$2
 166   sed "s#__ANONYMOUS_USER_TOKEN__#${ANONYMOUS_USER_TOKEN}#g;
 167        s#__BLOB_SIGNING_KEY__#${BLOB_SIGNING_KEY}#g;
 168        s#__CONTROLLER_EXT_SSL_PORT__#${CONTROLLER_EXT_SSL_PORT}#g;
 169        s#__CLUSTER__#${CLUSTER}#g;
 170        s#__DOMAIN__#${DOMAIN}#g;
 171        s#__HOSTNAME_EXT__#${HOSTNAME_EXT}#g;
 172        s#__IP_INT__#${IP_INT}#g;
 173        s#__INITIAL_USER_EMAIL__#${INITIAL_USER_EMAIL}#g;
 174        s#__INITIAL_USER_PASSWORD__#${INITIAL_USER_PASSWORD}#g;
 175        s#__INITIAL_USER__#${INITIAL_USER}#g;
 176        s#__LE_AWS_REGION__#${LE_AWS_REGION:-}#g;
 177        s#__LE_AWS_SECRET_ACCESS_KEY__#${LE_AWS_SECRET_ACCESS_KEY:-}#g;
 178        s#__LE_AWS_ACCESS_KEY_ID__#${LE_AWS_ACCESS_KEY_ID:-}#g;
 179        s#__DATABASE_NAME__#${DATABASE_NAME}#g;
 180        s#__DATABASE_USER__#${DATABASE_USER}#g;
 181        s#__DATABASE_PASSWORD__#${DATABASE_PASSWORD}#g;
 182        s#__DATABASE_INT_IP__#${DATABASE_INT_IP:-}#g;
 183        s#__DATABASE_EXTERNAL_SERVICE_HOST_OR_IP__#${DATABASE_EXTERNAL_SERVICE_HOST_OR_IP:-}#g;
 184        s#__DATABASE_POSTGRESQL_VERSION__#${DATABASE_POSTGRESQL_VERSION}#g;
 185        s#__KEEPWEB_EXT_SSL_PORT__#${KEEPWEB_EXT_SSL_PORT}#g;
 186        s#__KEEP_EXT_SSL_PORT__#${KEEP_EXT_SSL_PORT}#g;
 187        s#__MANAGEMENT_TOKEN__#${MANAGEMENT_TOKEN}#g;
 188        s#__RELEASE__#${RELEASE}#g;
 189        s#__SYSTEM_ROOT_TOKEN__#${SYSTEM_ROOT_TOKEN}#g;
 190        s#__VERSION__#${VERSION}#g;
 191        s#__WEBSHELL_EXT_SSL_PORT__#${WEBSHELL_EXT_SSL_PORT}#g;
 192        s#__WEBSOCKET_EXT_SSL_PORT__#${WEBSOCKET_EXT_SSL_PORT}#g;
 193        s#__WORKBENCH1_EXT_SSL_PORT__#${WORKBENCH1_EXT_SSL_PORT}#g;
 194        s#__WORKBENCH2_EXT_SSL_PORT__#${WORKBENCH2_EXT_SSL_PORT}#g;
 195        s#__CLUSTER_INT_CIDR__#${CLUSTER_INT_CIDR}#g;
 196        s#__CONTROLLER_INT_IP__#${CONTROLLER_INT_IP}#g;
 197        s#__WEBSOCKET_INT_IP__#${WEBSOCKET_INT_IP}#g;
 198        s#__KEEP_INT_IP__#${KEEP_INT_IP}#g;
 199        s#__KEEPSTORE0_INT_IP__#${KEEPSTORE0_INT_IP}#g;
 200        s#__KEEPWEB_INT_IP__#${KEEPWEB_INT_IP}#g;
 201        s#__WEBSHELL_INT_IP__#${WEBSHELL_INT_IP}#g;
 202        s#__SHELL_INT_IP__#${SHELL_INT_IP}#g;
 203        s#__WORKBENCH1_INT_IP__#${WORKBENCH1_INT_IP}#g;
 204        s#__WORKBENCH2_INT_IP__#${WORKBENCH2_INT_IP}#g;
 205        s#__SSL_KEY_ENCRYPTED__#${SSL_KEY_ENCRYPTED}#g;
 206        s#__SSL_KEY_AWS_REGION__#${SSL_KEY_AWS_REGION:-}#g;
 207        s#__SSL_KEY_AWS_SECRET_NAME__#${SSL_KEY_AWS_SECRET_NAME}#g;
 208        s#__CONTROLLER_MAX_WORKERS__#${CONTROLLER_MAX_WORKERS:-}#g;
 209        s#__CONTROLLER_MAX_QUEUED_REQUESTS__#${CONTROLLER_MAX_QUEUED_REQUESTS:-128}#g;
 210        s#__CONTROLLER_MAX_GATEWAY_TUNNELS__#${CONTROLLER_MAX_GATEWAY_TUNNELS:-1000}#g;
 211        s#__MONITORING_USERNAME__#${MONITORING_USERNAME}#g;
 212        s#__MONITORING_EMAIL__#${MONITORING_EMAIL}#g;
 213        s#__MONITORING_PASSWORD__#${MONITORING_PASSWORD}#g;
 214        s#__DISPATCHER_SSH_PRIVKEY__#${DISPATCHER_SSH_PRIVKEY//$'\n'/\\n}#g;
 215        s#__ENABLE_BALANCER__#${ENABLE_BALANCER}#g;
 216        s#__DISABLED_CONTROLLER__#${DISABLED_CONTROLLER}#g;
 217        s#__BALANCER_NODENAME__#${ROLE2NODES['balancer']:-}#g;
 218        s#__PROMETHEUS_NODENAME__#${ROLE2NODES['monitoring']:-}#g;
 219        s#__PROMETHEUS_DATA_RETENTION_TIME__#${PROMETHEUS_DATA_RETENTION_TIME:-15d}#g;
 220        s#__LOKI_AWS_S3_ACCESS_KEY_ID__#${LOKI_AWS_S3_ACCESS_KEY_ID:-}#g;
 221        s#__LOKI_AWS_S3_SECRET_ACCESS_KEY__#${LOKI_AWS_S3_SECRET_ACCESS_KEY:-}#g;
 222        s#__LOKI_AWS_S3_BUCKET__#${LOKI_AWS_S3_BUCKET:-}#g;
 223        s#__LOKI_LOG_RETENTION_TIME__#${LOKI_LOG_RETENTION_TIME:-15d}#g;
 224        s#__LOKI_AWS_REGION__#${LOKI_AWS_REGION:-}#g;
 225        s#__CONTROLLER_NODES__#${ROLE2NODES['controller']:-}#g;
 226        s#__NODELIST__#${NODELIST}#g;
 227        s#__DISPATCHER_INT_IP__#${DISPATCHER_INT_IP}#g;
 228        s#__KEEPBALANCE_INT_IP__#${KEEPBALANCE_INT_IP}#g;
 229        s#__COMPUTE_AMI__#${COMPUTE_AMI:-}#g;
 230        s#__COMPUTE_SG__#${COMPUTE_SG:-}#g;
 231        s#__COMPUTE_SUBNET__#${COMPUTE_SUBNET:-}#g;
 232        s#__COMPUTE_AWS_REGION__#${COMPUTE_AWS_REGION:-}#g;
 233        s#__COMPUTE_USER__#${COMPUTE_USER:-}#g;
 234        s#__KEEP_AWS_S3_BUCKET__#${KEEP_AWS_S3_BUCKET:-}#g;
 235        s#__KEEP_AWS_IAM_ROLE__#${KEEP_AWS_IAM_ROLE:-}#g;
 236        s#__KEEP_AWS_REGION__#${KEEP_AWS_REGION:-}#g" \
 237   "${SRCFILE}" > "${DSTFILE}"
 238 }
 239
 240 DEV_MODE="no"
 241 CONFIG_FILE="${SCRIPT_DIR}/local.params"
 242 CONFIG_DIR="local_config_dir"
 243 DUMP_CONFIG="no"
 244 LOG_LEVEL="info"
 245 CONTROLLER_EXT_SSL_PORT=443
 246 TESTS_DIR="tests"
 247
 248 NGINX_INSTALL_SOURCE="install_from_repo"
 249
 250 CLUSTER=""
 251 DOMAIN=""
 252
 253 # Hostnames/IPs used for single-host deploys
 254 IP_INT="127.0.1.1"
 255
 256 # Initial user setup
 257 INITIAL_USER=""
 258 INITIAL_USER_EMAIL=""
 259 INITIAL_USER_PASSWORD=""
 260
 261 CONTROLLER_EXT_SSL_PORT=8000
 262 KEEP_EXT_SSL_PORT=25101
 263 # Both for collections and downloads
 264 KEEPWEB_EXT_SSL_PORT=9002
 265 WEBSHELL_EXT_SSL_PORT=4202
 266 WEBSOCKET_EXT_SSL_PORT=8002
 267 WORKBENCH1_EXT_SSL_PORT=443
 268 WORKBENCH2_EXT_SSL_PORT=3001
 269
 270 SSL_MODE="self-signed"
 271 USE_LETSENCRYPT_ROUTE53="no"
 272 CUSTOM_CERTS_DIR="${SCRIPT_DIR}/local_config_dir/certs"
 273
 274 GRAFANA_DASHBOARDS_DIR="${SCRIPT_DIR}/local_config_dir/dashboards"
 275
 276 ## These are ARVADOS-related parameters
 277 # For a stable release, change RELEASE "production" and VERSION to the
 278 # package version (including the iteration, e.g. X.Y.Z-1) of the
 279 # release.
 280 # The "local.params.example.*" files already set "RELEASE=production"
 281 # to deploy  production-ready packages
 282 RELEASE="development"
 283 VERSION="latest"
 284
 285 # These are arvados-formula-related parameters
 286 # An arvados-formula tag. For a stable release, this should be a
 287 # branch name (e.g. X.Y-dev) or tag for the release.
 288 # ARVADOS_TAG="2.2.0"
 289 # BRANCH="main"
 290
 291 # We pin the salt version to avoid potential incompatibilities when a new
 292 # stable version is released.
 293 SALT_VERSION="3006"
 294
 295 # Other formula versions we depend on
 296 POSTGRES_TAG="a809e03bad115bbdf24ad347e2dc9a52e144c31f"
 297 POSTGRES_URL="https://github.com/arvados/postgres-formula.git"
 298 NGINX_TAG="v2.8.1"
 299 DOCKER_TAG="v2.4.2"
 300 LOCALE_TAG="v0.3.5"
 301 LETSENCRYPT_TAG="v3.2.0"
 302 LOGROTATE_TAG="v0.14.0"
 303 PROMETHEUS_TAG="v5.6.5"
 304 GRAFANA_TAG="v3.1.3"
 305
 306 # Salt's dir
 307 DUMP_SALT_CONFIG_DIR=""
 308 ## states
 309 S_DIR="/srv/salt"
 310 STATES_TOP=${S_DIR}/top.sls
 311 ## formulas
 312 F_DIR="/srv/formulas"
 313 ## pillars
 314 P_DIR="/srv/pillars"
 315 PILLARS_TOP=${P_DIR}/top.sls
 316 ## tests
 317 T_DIR="/tmp/cluster_tests"
 318
 319 arguments ${@}
 320
 321 declare -A NODES
 322 declare -A ROLE2NODES
 323 declare NODELIST
 324
 325 source common.sh
 326
 327 if [ ! -d ${CONFIG_DIR} ]; then
 328   echo >&2 "You don't seem to have a config directory with pillars and states."
 329   echo >&2 "Please create a '${CONFIG_DIR}' directory (as configured in your '${CONFIG_FILE}'). Please see"
 330   echo >&2 "  * https://doc.arvados.org/install/salt-single-host.html#single_host, or"
 331   echo >&2 "  * https://doc.arvados.org/install/salt-multi-host.html#multi_host_multi_hostnames"
 332   exit 1
 333 fi
 334
 335 if grep -rni 'fixme' ${CONFIG_FILE}.secrets ${CONFIG_FILE} ${CONFIG_DIR} ; then
 336   echo >&2 "The config files has some parameters that need to be modified."
 337   echo >&2 "Please, fix them and re-run the provision script."
 338   exit 1
 339 fi
 340
 341 if ! grep -qE '^[[:alnum:]]{5}$' <<<${CLUSTER} ; then
 342   echo >&2 "ERROR: <CLUSTER> must be exactly 5 lowercase alphanumeric characters long"
 343   echo >&2 "Fix the cluster name in the 'local.params' file and re-run the provision script"
 344   exit 1
 345 fi
 346
 347 # Only used in single_host/single_name deploys
 348 if [ ! -z "${HOSTNAME_EXT:-}" ] ; then
 349   # We need to add some extra control vars to manage a single certificate vs. multiple
 350   USE_SINGLE_HOSTNAME="yes"
 351   # Make sure that the value configured as IP_INT is a real IP on the system.
 352   # If we don't error out early here when there is a mismatch, the formula will
 353   # fail with hard to interpret nginx errors later on.
 354   ip addr list |grep "${IP_INT}/" >/dev/null
 355   if [[ $? -ne 0 ]]; then
 356     echo "Unable to find the IP_INT address '${IP_INT}' on the system, please correct the value in local.params. Exiting..."
 357     exit 1
 358   fi
 359 else
 360   USE_SINGLE_HOSTNAME="no"
 361   # We set this variable, anyway, so sed lines do not fail and we don't need to add more
 362   # conditionals
 363   HOSTNAME_EXT="${DOMAIN}"
 364 fi
 365
 366 if [ "${DUMP_CONFIG}" = "yes" ]; then
 367   echo "The provision installer will just dump a config under ${DUMP_SALT_CONFIG_DIR} and exit"
 368 else
 369   # Read the variables of /etc/os-release but prefix their names with `_OS_`
 370   # to avoid name conflicts.
 371   eval "$(awk '(/^[A-Z_]+=/) { print "_OS_" $0 }' /etc/os-release)"
 372   echo "Detected distro families: ${_OS_ID:-} ${_OS_ID_LIKE:-}"
 373
 374   # Several of our formulas use the cron module, which requires the crontab
 375   # command. We install systemd-cron to ensure we have that.
 376   # The rest of these packages are required by the rest of the script.
 377   for OS_ID in ${_OS_ID:-} ${_OS_ID_LIKE:-}; do
 378     case "$OS_ID" in
 379       rhel)
 380         echo "WARNING! Disabling SELinux, see https://dev.arvados.org/issues/18019"
 381         sed -i 's/SELINUX=enforcing/SELINUX=permissive/g' /etc/sysconfig/selinux
 382         setenforce permissive
 383         yum install -y curl git jq systemd-cron
 384         if command -v salt-call >/dev/null; then
 385             echo "Salt already installed"
 386             break
 387         fi
 388         curl -L https://bootstrap.saltstack.com -o /tmp/bootstrap_salt.sh
 389         sh /tmp/bootstrap_salt.sh -XdfP -x python3 stable ${SALT_VERSION}
 390         break
 391         ;;
 392       debian)
 393         DEBIAN_FRONTEND=noninteractive apt-get -o DPkg::Lock::Timeout=120 update
 394         # This list includes our own dependencies, plus depdencies necessary
 395         # to retrieve the Salt apt repository.
 396         DEBIAN_FRONTEND=noninteractive apt-get install -y \
 397                                        apt-transport-https ca-certificates curl git gnupg jq systemd-cron
 398         if command -v salt-call >/dev/null; then
 399             echo "Salt already installed"
 400             break
 401         fi
 402         salt_apt_url="https://repo.saltproject.io/salt/py3/$_OS_ID/$_OS_VERSION_ID/$(dpkg --print-architecture)"
 403         salt_apt_key=SALT-PROJECT-GPG-PUBKEY-2023.gpg
 404         install -d -m 755 /etc/apt/keyrings
 405         curl -fsSL -o "/etc/apt/keyrings/$salt_apt_key" "$salt_apt_url/$salt_apt_key"
 406         chmod go+r "/etc/apt/keyrings/$salt_apt_key"
 407         install -b -m 644 /dev/stdin "/etc/apt/sources.list.d/salt$SALT_VERSION.sources" <<EOFSOURCES
 408 Types: deb
 409 URIs: $salt_apt_url/$SALT_VERSION
 410 Suites: $_OS_VERSION_CODENAME
 411 Components: main
 412 Signed-by: /etc/apt/keyrings/$salt_apt_key
 413 EOFSOURCES
 414         DEBIAN_FRONTEND=noninteractive apt-get -o DPkg::Lock::Timeout=120 update
 415         DEBIAN_FRONTEND=noninteractive apt-get install -y salt-minion
 416         break
 417         ;;
 418     esac
 419   done
 420
 421   # Set salt to masterless mode
 422   systemctl disable --now salt-minion.service
 423   cat > /etc/salt/minion << EOFSM
 424 failhard: "True"
 425
 426 file_client: local
 427 file_roots:
 428   base:
 429     - ${S_DIR}
 430     - ${F_DIR}/*
 431
 432 pillar_roots:
 433   base:
 434     - ${P_DIR}
 435 EOFSM
 436 fi
 437
 438 mkdir -p ${S_DIR} ${F_DIR} ${P_DIR} ${T_DIR}
 439
 440 # Get the formula and dependencies
 441 cd ${F_DIR} || exit 1
 442 echo "Cloning formulas"
 443 test -d docker && ( cd docker && git fetch ) \
 444   || git clone --quiet https://github.com/saltstack-formulas/docker-formula.git ${F_DIR}/docker
 445 ( cd docker && git checkout --quiet tags/"${DOCKER_TAG}" )
 446
 447 echo "...locale"
 448 test -d locale && ( cd locale && git fetch ) \
 449   || git clone --quiet https://github.com/saltstack-formulas/locale-formula.git ${F_DIR}/locale
 450 ( cd locale && git checkout --quiet tags/"${LOCALE_TAG}" )
 451
 452 echo "...nginx"
 453 test -d nginx && ( cd nginx && git fetch ) \
 454   || git clone --quiet https://github.com/saltstack-formulas/nginx-formula.git ${F_DIR}/nginx
 455 ( cd nginx && git checkout --quiet tags/"${NGINX_TAG}" )
 456
 457 echo "...postgres"
 458 test -d postgres && ( cd postgres && git fetch ) \
 459   || git clone --quiet ${POSTGRES_URL} ${F_DIR}/postgres
 460 ( cd postgres && git checkout --quiet "${POSTGRES_TAG}" )
 461
 462 echo "...prometheus"
 463 test -d prometheus && ( cd prometheus && git fetch ) \
 464   || git clone --quiet https://github.com/saltstack-formulas/prometheus-formula.git ${F_DIR}/prometheus
 465 ( cd prometheus && git checkout --quiet tags/"${PROMETHEUS_TAG}" )
 466
 467 echo "...grafana"
 468 test -d grafana && ( cd grafana && git fetch ) \
 469   || git clone --quiet https://github.com/saltstack-formulas/grafana-formula.git ${F_DIR}/grafana
 470 ( cd grafana && git checkout --quiet "${GRAFANA_TAG}" )
 471
 472 echo "...letsencrypt"
 473 test -d letsencrypt && ( cd letsencrypt && git fetch ) \
 474   || git clone --quiet https://github.com/saltstack-formulas/letsencrypt-formula.git ${F_DIR}/letsencrypt
 475 ( cd letsencrypt && git checkout --quiet tags/"${LETSENCRYPT_TAG}" )
 476
 477 echo "...logrotate"
 478 test -d logrotate && ( cd logrotate && git fetch ) \
 479   || git clone --quiet https://github.com/saltstack-formulas/logrotate-formula.git ${F_DIR}/logrotate
 480 ( cd logrotate && git checkout --quiet tags/"${LOGROTATE_TAG}" )
 481
 482 echo "...arvados"
 483 test -d arvados || git clone --quiet https://git.arvados.org/arvados-formula.git ${F_DIR}/arvados
 484
 485 # If we want to try a specific branch of the formula
 486 if [[ ! -z "${BRANCH:-}" && "x${BRANCH}" != "xmain" ]]; then
 487   ( cd ${F_DIR}/arvados && git fetch && git checkout --quiet "${BRANCH}" || git checkout --quiet -t origin/"${BRANCH}" -b "${BRANCH}" )
 488 elif [ "x${ARVADOS_TAG:-}" != "x" ]; then
 489   ( cd ${F_DIR}/arvados && git checkout --quiet tags/"${ARVADOS_TAG}" -b "${ARVADOS_TAG}" )
 490 fi
 491
 492 if [ "x${VAGRANT:-}" = "xyes" ]; then
 493   EXTRA_STATES_DIR="/home/vagrant/${CONFIG_DIR}/states"
 494   SOURCE_PILLARS_DIR="/home/vagrant/${CONFIG_DIR}/pillars"
 495   SOURCE_TOFS_DIR="/home/vagrant/${CONFIG_DIR}/tofs"
 496   SOURCE_TESTS_DIR="/home/vagrant/${TESTS_DIR}"
 497 else
 498   EXTRA_STATES_DIR="${SCRIPT_DIR}/${CONFIG_DIR}/states"
 499   SOURCE_PILLARS_DIR="${SCRIPT_DIR}/${CONFIG_DIR}/pillars"
 500   SOURCE_TOFS_DIR="${SCRIPT_DIR}/${CONFIG_DIR}/tofs"
 501   SOURCE_TESTS_DIR="${SCRIPT_DIR}/${TESTS_DIR}"
 502 fi
 503
 504 SOURCE_STATES_DIR="${EXTRA_STATES_DIR}"
 505
 506 echo "Writing pillars and states"
 507
 508 # Replace variables (cluster,  domain, etc) in the pillars, states and tests
 509 # to ease deployment for newcomers
 510 if [ ! -d "${SOURCE_PILLARS_DIR}" ]; then
 511   echo "${SOURCE_PILLARS_DIR} does not exist or is not a directory. Exiting."
 512   exit 1
 513 fi
 514 for f in $(ls "${SOURCE_PILLARS_DIR}"/*); do
 515   apply_var_substitutions "${f}" "${P_DIR}"/$(basename "${f}")
 516 done
 517
 518 if [ ! -d "${SOURCE_TESTS_DIR}" ]; then
 519   echo "WARNING: The tests directory was not copied to \"${SOURCE_TESTS_DIR}\"."
 520   if [ "x${TEST:-}" = "xyes" ]; then
 521     echo "WARNING: Disabling tests for this installation."
 522   fi
 523   TEST="no"
 524 else
 525   mkdir -p ${T_DIR}
 526   # Replace cluster and domain name in the test files
 527   for f in $(ls "${SOURCE_TESTS_DIR}"/*); do
 528     FILTERS="s#__CLUSTER__#${CLUSTER}#g;
 529          s#__CONTROLLER_EXT_SSL_PORT__#${CONTROLLER_EXT_SSL_PORT}#g;
 530          s#__DOMAIN__#${DOMAIN}#g;
 531          s#__IP_INT__#${IP_INT}#g;
 532          s#__INITIAL_USER_EMAIL__#${INITIAL_USER_EMAIL}#g;
 533          s#__INITIAL_USER_PASSWORD__#${INITIAL_USER_PASSWORD}#g
 534          s#__INITIAL_USER__#${INITIAL_USER}#g;
 535          s#__DATABASE_PASSWORD__#${DATABASE_PASSWORD}#g;
 536          s#__SYSTEM_ROOT_TOKEN__#${SYSTEM_ROOT_TOKEN}#g"
 537     if [ "$USE_SINGLE_HOSTNAME" = "yes" ]; then
 538       FILTERS="s#__CLUSTER__.__DOMAIN__#${HOSTNAME_EXT}#g;
 539          $FILTERS"
 540     fi
 541     sed "$FILTERS" \
 542       "${f}" > ${T_DIR}/$(basename "${f}")
 543   done
 544   chmod 755 ${T_DIR}/run-test.sh
 545 fi
 546
 547 # Replace helper state files that differ from the formula's examples
 548 if [ -d "${SOURCE_STATES_DIR}" ]; then
 549   mkdir -p "${F_DIR}"/extra/extra
 550   rm -rf "${F_DIR}"/extra/extra/*
 551
 552   for f in $(ls "${SOURCE_STATES_DIR}"/*); do
 553     apply_var_substitutions "${f}" "${F_DIR}/extra/extra"/$(basename "${f}")
 554   done
 555 fi
 556
 557 # Now, we build the SALT states/pillars trees
 558 # As we need to separate both states and pillars in case we want specific
 559 # roles, we iterate on both at the same time
 560
 561 # Formula template overrides (TOFS)
 562 # See: https://template-formula.readthedocs.io/en/latest/TOFS_pattern.html#template-override
 563 if [ -d ${SOURCE_TOFS_DIR} ]; then
 564   find ${SOURCE_TOFS_DIR} -mindepth 1 -maxdepth 1 -type d -exec cp -r "{}" ${S_DIR} \;
 565 fi
 566
 567 # States
 568 cat > ${STATES_TOP} << EOFTSLS
 569 base:
 570   '*':
 571     - locale
 572 EOFTSLS
 573
 574 # Pillars
 575 cat > ${PILLARS_TOP} << EOFPSLS
 576 base:
 577   '*':
 578     - locale
 579     - arvados
 580 EOFPSLS
 581
 582 # States, extra states
 583 if [ -d "${F_DIR}"/extra/extra ]; then
 584   SKIP_SNAKE_OIL="snakeoil_certs"
 585
 586   if [[ "$DEV_MODE" = "yes" || "${SSL_MODE}" == "self-signed" ]] ; then
 587     # In dev mode, we create some snake oil certs that we'll
 588     # use as CUSTOM_CERTS, so we don't skip the states file.
 589     # Same when using self-signed certificates.
 590     SKIP_SNAKE_OIL="dont_add_snakeoil_certs"
 591   fi
 592   for f in $(ls "${F_DIR}"/extra/extra/*.sls | egrep -v "${SKIP_SNAKE_OIL}|shell_"); do
 593   echo "    - extra.$(basename ${f} | sed 's/.sls$//g')" >> ${STATES_TOP}
 594   done
 595   # Use byo or self-signed certificates
 596   if [ "${SSL_MODE}" != "lets-encrypt" ]; then
 597     mkdir -p "${F_DIR}"/extra/extra/files
 598   fi
 599 fi
 600
 601 # If we want specific roles for a node, just add the desired states
 602 # and its dependencies
 603 if [ -z "${ROLES:-}" ]; then
 604   # States
 605   echo "    - nginx.passenger" >> ${STATES_TOP}
 606   if [ "${SSL_MODE}" = "lets-encrypt" ]; then
 607     if [ "${USE_LETSENCRYPT_ROUTE53}" = "yes" ]; then
 608       grep -q "aws_credentials" ${STATES_TOP} || echo "    - extra.aws_credentials" >> ${STATES_TOP}
 609     fi
 610     grep -q "letsencrypt" ${STATES_TOP} || echo "    - letsencrypt" >> ${STATES_TOP}
 611   else
 612     mkdir -p --mode=0700 /srv/salt/certs
 613     if [ "${SSL_MODE}" = "bring-your-own" ]; then
 614       # Copy certs to formula extra/files
 615       install --mode=0600 ${CUSTOM_CERTS_DIR}/* /srv/salt/certs/
 616       # We add the custom_certs state
 617       grep -q "custom_certs" ${STATES_TOP} || echo "    - extra.custom_certs" >> ${STATES_TOP}
 618       if [ "${SSL_KEY_ENCRYPTED}" = "yes" ]; then
 619         grep -q "ssl_key_encrypted" ${STATES_TOP} || echo "    - extra.ssl_key_encrypted" >> ${STATES_TOP}
 620       fi
 621     fi
 622     # In self-signed mode, the certificate files will be created and put in the
 623     # destination directory by the snakeoil_certs.sls state file
 624   fi
 625
 626   echo "    - postgres" >> ${STATES_TOP}
 627   echo "    - logrotate" >> ${STATES_TOP}
 628   echo "    - docker.software" >> ${STATES_TOP}
 629   echo "    - arvados.repo" >> ${STATES_TOP}
 630   echo "    - arvados.config" >> ${STATES_TOP}
 631   echo "    - arvados.ruby" >> ${STATES_TOP}
 632   echo "    - arvados.api" >> ${STATES_TOP}
 633   echo "    - arvados.controller" >> ${STATES_TOP}
 634   echo "    - arvados.keepstore" >> ${STATES_TOP}
 635   echo "    - arvados.websocket" >> ${STATES_TOP}
 636   echo "    - arvados.keepweb" >> ${STATES_TOP}
 637   echo "    - arvados.workbench2" >> ${STATES_TOP}
 638   echo "    - arvados.keepproxy" >> ${STATES_TOP}
 639   echo "    - arvados.shell" >> ${STATES_TOP}
 640   echo "    - arvados.dispatcher" >> ${STATES_TOP}
 641   echo "    - extra.shell_sudo_passwordless" >> ${STATES_TOP}
 642   echo "    - extra.shell_cron_add_login_sync" >> ${STATES_TOP}
 643   echo "    - extra.passenger_rvm" >> ${STATES_TOP}
 644   echo "    - extra.workbench1_uninstall" >> ${STATES_TOP}
 645
 646   # Pillars
 647   echo "    - docker" >> ${PILLARS_TOP}
 648   echo "    - nginx_api_configuration" >> ${PILLARS_TOP}
 649   echo "    - logrotate" >> ${PILLARS_TOP}
 650   echo "    - logrotate_api" >> ${PILLARS_TOP}
 651   echo "    - nginx_controller_configuration" >> ${PILLARS_TOP}
 652   echo "    - nginx_keepproxy_configuration" >> ${PILLARS_TOP}
 653   echo "    - nginx_keepweb_configuration" >> ${PILLARS_TOP}
 654   echo "    - nginx_passenger" >> ${PILLARS_TOP}
 655   echo "    - nginx_websocket_configuration" >> ${PILLARS_TOP}
 656   echo "    - nginx_webshell_configuration" >> ${PILLARS_TOP}
 657   echo "    - nginx_workbench2_configuration" >> ${PILLARS_TOP}
 658   echo "    - nginx_workbench_configuration" >> ${PILLARS_TOP}
 659   echo "    - logrotate_wb1" >> ${PILLARS_TOP}
 660   echo "    - postgresql" >> ${PILLARS_TOP}
 661
 662   # We need to tweak the Nginx's pillar depending whether we want plan nginx or nginx+passenger
 663   NGINX_INSTALL_SOURCE="install_from_phusionpassenger"
 664   sed -i "s/__NGINX_INSTALL_SOURCE__/${NGINX_INSTALL_SOURCE}/g" ${P_DIR}/nginx_passenger.sls
 665
 666   if [ "${SSL_MODE}" = "lets-encrypt" ]; then
 667     if [ "${USE_LETSENCRYPT_ROUTE53}" = "yes" ]; then
 668       grep -q "aws_credentials" ${PILLARS_TOP} || echo "    - aws_credentials" >> ${PILLARS_TOP}
 669     fi
 670     grep -q "letsencrypt" ${PILLARS_TOP} || echo "    - letsencrypt" >> ${PILLARS_TOP}
 671
 672     hosts=("controller" "websocket" "workbench" "workbench2" "webshell" "keepproxy")
 673     if [ ${USE_SINGLE_HOSTNAME} = "no" ]; then
 674       hosts+=("download" "collections")
 675     else
 676       hosts+=("keepweb")
 677     fi
 678
 679     for c in "${hosts[@]}"; do
 680       # Are we in a single-host-single-hostname env?
 681       if [ "${USE_SINGLE_HOSTNAME}" = "yes" ]; then
 682         # Are we in a single-host-single-hostname env?
 683         CERT_NAME=${HOSTNAME_EXT}
 684       else
 685         # We are in a multiple-hostnames env
 686         CERT_NAME=${c}.${DOMAIN}
 687       fi
 688
 689       # As the pillar differs whether we use LE or custom certs, we need to do a final edition on them
 690       sed -i "s/__CERT_REQUIRES__/cmd: create-initial-cert-${CERT_NAME}*/g;
 691               s#__CERT_PEM__#/etc/letsencrypt/live/${CERT_NAME}/fullchain.pem#g;
 692               s#__CERT_KEY__#/etc/letsencrypt/live/${CERT_NAME}/privkey.pem#g" \
 693       ${P_DIR}/nginx_${c}_configuration.sls
 694     done
 695   else
 696     # Use custom certs (either dev mode or prod)
 697     grep -q "extra_custom_certs" ${PILLARS_TOP} || echo "    - extra_custom_certs" >> ${PILLARS_TOP}
 698     # And add the certs in the custom_certs pillar
 699     echo "extra_custom_certs_dir: /srv/salt/certs" > ${P_DIR}/extra_custom_certs.sls
 700     echo "extra_custom_certs:" >> ${P_DIR}/extra_custom_certs.sls
 701
 702     for c in controller websocket workbench workbench2 webshell keepweb keepproxy; do
 703       # Are we in a single-host-single-hostname env?
 704       if [ "${USE_SINGLE_HOSTNAME}" = "yes" ]; then
 705         # Are we in a single-host-single-hostname env?
 706         CERT_NAME=${HOSTNAME_EXT}
 707       else
 708         # We are in a multiple-hostnames env
 709         CERT_NAME=${c}
 710       fi
 711
 712       if [[ "$SSL_MODE" == "bring-your-own" ]]; then
 713         copy_custom_cert ${CUSTOM_CERTS_DIR} ${CERT_NAME}
 714       fi
 715
 716       grep -q ${CERT_NAME} ${P_DIR}/extra_custom_certs.sls || echo "  - ${CERT_NAME}" >> ${P_DIR}/extra_custom_certs.sls
 717
 718       # As the pillar differs whether we use LE or custom certs, we need to do a final edition on them
 719       sed -i "s/__CERT_REQUIRES__/file: extra_custom_certs_${CERT_NAME}_cert_file_copy/g;
 720               s#__CERT_PEM__#/etc/nginx/ssl/arvados-${CERT_NAME}.pem#g;
 721               s#__CERT_KEY__#/etc/nginx/ssl/arvados-${CERT_NAME}.key#g" \
 722       ${P_DIR}/nginx_${c}_configuration.sls
 723     done
 724   fi
 725 else
 726   # If we add individual roles, make sure we add the repo first
 727   echo "    - arvados.repo" >> ${STATES_TOP}
 728   # We add the extra_custom_certs state
 729   grep -q "extra.custom_certs"    ${STATES_TOP} || echo "    - extra.custom_certs" >> ${STATES_TOP}
 730   if [ "${SSL_KEY_ENCRYPTED}" = "yes" ]; then
 731     grep -q "ssl_key_encrypted" ${STATES_TOP} || echo "    - extra.ssl_key_encrypted" >> ${STATES_TOP}
 732   fi
 733
 734   # And we add the basic part for the certs pillar
 735   if [ "${SSL_MODE}" != "lets-encrypt" ]; then
 736     # And add the certs in the custom_certs pillar
 737     echo "extra_custom_certs_dir: /srv/salt/certs" > ${P_DIR}/extra_custom_certs.sls
 738     echo "extra_custom_certs:" >> ${P_DIR}/extra_custom_certs.sls
 739     grep -q "extra_custom_certs" ${PILLARS_TOP} || echo "    - extra_custom_certs" >> ${PILLARS_TOP}
 740   fi
 741
 742   # Prometheus state on all nodes due to the node exporter below
 743   grep -q "\- prometheus$" ${STATES_TOP} || echo "    - prometheus" >> ${STATES_TOP}
 744   # Prometheus node exporter pillar
 745   grep -q "prometheus_node_exporter" ${PILLARS_TOP} || echo "    - prometheus_node_exporter" >> ${PILLARS_TOP}
 746   # Grafana Alloy OpenTelemetry client state & pillar on all nodes
 747   grep -q "\- grafana$" ${STATES_TOP} || echo "    - grafana" >> ${STATES_TOP}
 748   grep -q "extra.alloy_install" ${STATES_TOP} || echo "    - extra.alloy_install" >> ${STATES_TOP}
 749   grep -q "alloy" ${PILLARS_TOP} || echo "    - alloy" >> ${PILLARS_TOP}
 750
 751   for R in ${ROLES:-}; do
 752     case "${R}" in
 753       "database")
 754         # Skip if using an external service
 755         if [[ "${DATABASE_EXTERNAL_SERVICE_HOST_OR_IP:-}" != "" ]]; then
 756           continue
 757         fi
 758         # States
 759         grep -q "\- postgres$" ${STATES_TOP} || echo "    - postgres" >> ${STATES_TOP}
 760         grep -q "extra.prometheus_pg_exporter" ${STATES_TOP} || echo "    - extra.prometheus_pg_exporter" >> ${STATES_TOP}
 761         # Pillars
 762         grep -q "postgresql" ${PILLARS_TOP} || echo "    - postgresql" >> ${PILLARS_TOP}
 763         grep -q "prometheus_pg_exporter" ${PILLARS_TOP} || echo "    - prometheus_pg_exporter" >> ${PILLARS_TOP}
 764       ;;
 765       "monitoring")
 766         ### Support files ###
 767         GRAFANA_DASHBOARDS_DEST_DIR=/srv/salt/dashboards
 768         mkdir -p "${GRAFANA_DASHBOARDS_DEST_DIR}"
 769         rm -f "${GRAFANA_DASHBOARDS_DEST_DIR}"/*
 770         # "ArvadosPromDataSource" is the hardcoded UID for Prometheus' datasource
 771         # in Grafana.
 772         # "ArvadosLokiDataSource" if Loki's UID in Grafana
 773         for f in $(ls "${GRAFANA_DASHBOARDS_DIR}"/*.json); do
 774           sed "s#__TLS_EXPIRATION_YELLOW__#${TLS_EXPIRATION_YELLOW}#g;
 775                s#__TLS_EXPIRATION_GREEN__#${TLS_EXPIRATION_GREEN}#g;
 776                s#\${DS_PROMETHEUS}#ArvadosPromDataSource#g;
 777                s#\${DS_LOKI}#ArvadosLokiDataSource#g" \
 778           "${f}" > "${GRAFANA_DASHBOARDS_DEST_DIR}"/$(basename "${f}")
 779         done
 780
 781         ### States ###
 782         grep -q "\- nginx$" ${STATES_TOP} || echo "    - nginx" >> ${STATES_TOP}
 783         grep -q "extra.nginx_prometheus_configuration" ${STATES_TOP} || echo "    - extra.nginx_prometheus_configuration" >> ${STATES_TOP}
 784
 785         grep -q "\- grafana$" ${STATES_TOP} || echo "    - grafana" >> ${STATES_TOP}
 786         grep -q "extra.grafana_datasource" ${STATES_TOP} || echo "    - extra.grafana_datasource" >> ${STATES_TOP}
 787         grep -q "extra.grafana_dashboards" ${STATES_TOP} || echo "    - extra.grafana_dashboards" >> ${STATES_TOP}
 788         grep -q "extra.grafana_admin_user" ${STATES_TOP} || echo "    - extra.grafana_admin_user" >> ${STATES_TOP}
 789
 790         grep -q "extra.loki_install" ${STATES_TOP} || echo "    - extra.loki_install" >> ${STATES_TOP}
 791
 792         if [ "${SSL_MODE}" = "lets-encrypt" ]; then
 793           grep -q "letsencrypt"     ${STATES_TOP} || echo "    - letsencrypt" >> ${STATES_TOP}
 794           if [ "x${USE_LETSENCRYPT_ROUTE53:-}" = "xyes" ]; then
 795             grep -q "aws_credentials" ${STATES_TOP} || echo "    - aws_credentials" >> ${STATES_TOP}
 796           fi
 797         elif [ "${SSL_MODE}" = "bring-your-own" ]; then
 798           for SVC in grafana prometheus loki; do
 799             copy_custom_cert ${CUSTOM_CERTS_DIR} ${SVC}
 800           done
 801         fi
 802         ### Pillars ###
 803         grep -q "prometheus_server" ${PILLARS_TOP} || echo "    - prometheus_server" >> ${PILLARS_TOP}
 804         grep -q "grafana" ${PILLARS_TOP} || echo "    - grafana" >> ${PILLARS_TOP}
 805         grep -q "loki" ${PILLARS_TOP} || echo "    - loki" >> ${PILLARS_TOP}
 806         for SVC in grafana prometheus loki; do
 807           grep -q "nginx_${SVC}_configuration" ${PILLARS_TOP} || echo "    - nginx_${SVC}_configuration" >> ${PILLARS_TOP}
 808         done
 809         grep -q "nginx_snippets" ${PILLARS_TOP} || echo "    - nginx_snippets" >> ${PILLARS_TOP}
 810         if [ "${SSL_MODE}" = "lets-encrypt" ]; then
 811           grep -q "letsencrypt"     ${PILLARS_TOP} || echo "    - letsencrypt" >> ${PILLARS_TOP}
 812           for SVC in grafana prometheus loki; do
 813             grep -q "letsencrypt_${SVC}_configuration" ${PILLARS_TOP} || echo "    - letsencrypt_${SVC}_configuration" >> ${PILLARS_TOP}
 814             sed -i "s/__CERT_REQUIRES__/cmd: create-initial-cert-${SVC}.${DOMAIN}*/g;
 815                     s#__CERT_PEM__#/etc/letsencrypt/live/${SVC}.${DOMAIN}/fullchain.pem#g;
 816                     s#__CERT_KEY__#/etc/letsencrypt/live/${SVC}.${DOMAIN}/privkey.pem#g" \
 817             ${P_DIR}/nginx_${SVC}_configuration.sls
 818           done
 819           if [ "${USE_LETSENCRYPT_ROUTE53}" = "yes" ]; then
 820             grep -q "aws_credentials" ${PILLARS_TOP} || echo "    - aws_credentials" >> ${PILLARS_TOP}
 821           fi
 822         elif [ "${SSL_MODE}" = "bring-your-own" ]; then
 823           grep -q "ssl_key_encrypted" ${PILLARS_TOP} || echo "    - ssl_key_encrypted" >> ${PILLARS_TOP}
 824           for SVC in grafana prometheus loki; do
 825             sed -i "s/__CERT_REQUIRES__/file: extra_custom_certs_${SVC}_cert_file_copy/g;
 826                     s#__CERT_PEM__#/etc/nginx/ssl/arvados-${SVC}.pem#g;
 827                     s#__CERT_KEY__#/etc/nginx/ssl/arvados-${SVC}.key#g" \
 828               ${P_DIR}/nginx_${SVC}_configuration.sls
 829             grep -q ${SVC} ${P_DIR}/extra_custom_certs.sls || echo "  - ${SVC}" >> ${P_DIR}/extra_custom_certs.sls
 830           done
 831         fi
 832       ;;
 833       "balancer")
 834         ### States ###
 835         grep -q "\- nginx$" ${STATES_TOP} || echo "    - nginx" >> ${STATES_TOP}
 836
 837         if [ "${SSL_MODE}" = "lets-encrypt" ]; then
 838           grep -q "letsencrypt"     ${STATES_TOP} || echo "    - letsencrypt" >> ${STATES_TOP}
 839           if [ "x${USE_LETSENCRYPT_ROUTE53:-}" = "xyes" ]; then
 840             grep -q "aws_credentials" ${STATES_TOP} || echo "    - aws_credentials" >> ${STATES_TOP}
 841           fi
 842         elif [ "${SSL_MODE}" = "bring-your-own" ]; then
 843           copy_custom_cert ${CUSTOM_CERTS_DIR} ${R}
 844         fi
 845
 846         ### Pillars ###
 847         grep -q "nginx_${R}_configuration" ${PILLARS_TOP} || echo "    - nginx_${R}_configuration" >> ${PILLARS_TOP}
 848
 849         if [ "${SSL_MODE}" = "lets-encrypt" ]; then
 850           grep -q "letsencrypt"     ${PILLARS_TOP} || echo "    - letsencrypt" >> ${PILLARS_TOP}
 851
 852           grep -q "letsencrypt_${R}_configuration" ${PILLARS_TOP} || echo "    - letsencrypt_${R}_configuration" >> ${PILLARS_TOP}
 853           sed -i "s/__CERT_REQUIRES__/cmd: create-initial-cert-${ROLE2NODES['balancer']}*/g;
 854                   s#__CERT_PEM__#/etc/letsencrypt/live/${ROLE2NODES['balancer']}/fullchain.pem#g;
 855                   s#__CERT_KEY__#/etc/letsencrypt/live/${ROLE2NODES['balancer']}/privkey.pem#g" \
 856           ${P_DIR}/nginx_${R}_configuration.sls
 857
 858           if [ "${USE_LETSENCRYPT_ROUTE53}" = "yes" ]; then
 859             grep -q "aws_credentials" ${PILLARS_TOP} || echo "    - aws_credentials" >> ${PILLARS_TOP}
 860           fi
 861         elif [ "${SSL_MODE}" = "bring-your-own" ]; then
 862           grep -q "ssl_key_encrypted" ${PILLARS_TOP} || echo "    - ssl_key_encrypted" >> ${PILLARS_TOP}
 863           sed -i "s/__CERT_REQUIRES__/file: extra_custom_certs_${R}_cert_file_copy/g;
 864                   s#__CERT_PEM__#/etc/nginx/ssl/arvados-${R}.pem#g;
 865                   s#__CERT_KEY__#/etc/nginx/ssl/arvados-${R}.key#g" \
 866             ${P_DIR}/nginx_${R}_configuration.sls
 867           grep -q "${R}" ${P_DIR}/extra_custom_certs.sls || echo "  - ${R}" >> ${P_DIR}/extra_custom_certs.sls
 868         fi
 869       ;;
 870       "controller")
 871         ### States ###
 872         grep -q "    - logrotate" ${STATES_TOP} || echo "    - logrotate" >> ${STATES_TOP}
 873         if grep -q "    - nginx.*$" ${STATES_TOP}; then
 874           sed -i s/"^    - nginx.*$"/"    - nginx.passenger"/g ${STATES_TOP}
 875         else
 876           echo "    - nginx.passenger" >> ${STATES_TOP}
 877         fi
 878         echo "    - extra.passenger_rvm" >> ${STATES_TOP}
 879         grep -q "^    - postgres\\.client$" ${STATES_TOP} || echo "    - postgres.client" >> ${STATES_TOP}
 880         if [[ "${DATABASE_EXTERNAL_SERVICE_HOST_OR_IP:-}" != "" ]]; then
 881           grep -q "    - extra.postgresql_external" ${STATES_TOP} || echo "    - extra.postgresql_external" >> ${STATES_TOP}
 882         fi
 883
 884         ### If we don't install and run LE before arvados-api-server, it fails and breaks everything
 885         ### after it. So we add this here as we are, after all, sharing the host for api and controller
 886         if [ "${ENABLE_BALANCER}" == "no" ]; then
 887           if [ "${SSL_MODE}" = "lets-encrypt" ]; then
 888             if [ "x${USE_LETSENCRYPT_ROUTE53:-}" = "xyes" ]; then
 889               grep -q "aws_credentials" ${STATES_TOP} || echo "    - aws_credentials" >> ${STATES_TOP}
 890             fi
 891             grep -q "letsencrypt"     ${STATES_TOP} || echo "    - letsencrypt" >> ${STATES_TOP}
 892           elif [ "${SSL_MODE}" = "bring-your-own" ]; then
 893             copy_custom_cert ${CUSTOM_CERTS_DIR} ${R}
 894             grep -q controller ${P_DIR}/extra_custom_certs.sls || echo "  - controller" >> ${P_DIR}/extra_custom_certs.sls
 895           fi
 896         fi
 897         grep -q "arvados.api" ${STATES_TOP} || echo "    - arvados.api" >> ${STATES_TOP}
 898         grep -q "arvados.controller" ${STATES_TOP} || echo "    - arvados.controller" >> ${STATES_TOP}
 899
 900         ### Pillars ###
 901         grep -q "logrotate" ${PILLARS_TOP}                || echo "    - logrotate" >> ${PILLARS_TOP}
 902         grep -q "logrotate_api" ${PILLARS_TOP}            || echo "    - logrotate_api" >> ${PILLARS_TOP}
 903         grep -q "aws_credentials" ${PILLARS_TOP}          || echo "    - aws_credentials" >> ${PILLARS_TOP}
 904         grep -q "postgresql" ${PILLARS_TOP}               || echo "    - postgresql" >> ${PILLARS_TOP}
 905         grep -q "nginx_passenger" ${PILLARS_TOP}          || echo "    - nginx_passenger" >> ${PILLARS_TOP}
 906         grep -q "nginx_snippets" ${PILLARS_TOP}           || echo "    - nginx_snippets" >> ${PILLARS_TOP}
 907         grep -q "nginx_api_configuration" ${PILLARS_TOP} || echo "    - nginx_api_configuration" >> ${PILLARS_TOP}
 908         grep -q "nginx_controller_configuration" ${PILLARS_TOP} || echo "    - nginx_controller_configuration" >> ${PILLARS_TOP}
 909
 910         if [[ "${DATABASE_EXTERNAL_SERVICE_HOST_OR_IP:-}" != "" ]]; then
 911           grep -q "    - postgresql_external" ${PILLARS_TOP} || echo "    - postgresql_external" >> ${PILLARS_TOP}
 912         fi
 913
 914         if [ "${ENABLE_BALANCER}" == "no" ]; then
 915           if [ "${SSL_MODE}" = "lets-encrypt" ]; then
 916             if [ "${USE_LETSENCRYPT_ROUTE53}" = "yes" ]; then
 917               grep -q "aws_credentials" ${PILLARS_TOP} || echo "    - aws_credentials" >> ${PILLARS_TOP}
 918             fi
 919
 920             grep -q "letsencrypt"     ${PILLARS_TOP} || echo "    - letsencrypt" >> ${PILLARS_TOP}
 921             grep -q "letsencrypt_${R}_configuration" ${PILLARS_TOP} || echo "    - letsencrypt_${R}_configuration" >> ${PILLARS_TOP}
 922             sed -i "s/__CERT_REQUIRES__/cmd: create-initial-cert-${R}.${DOMAIN}*/g;
 923                     s#__CERT_PEM__#/etc/letsencrypt/live/${R}.${DOMAIN}/fullchain.pem#g;
 924                     s#__CERT_KEY__#/etc/letsencrypt/live/${R}.${DOMAIN}/privkey.pem#g" \
 925             ${P_DIR}/nginx_${R}_configuration.sls
 926           else
 927             grep -q "ssl_key_encrypted" ${PILLARS_TOP} || echo "    - ssl_key_encrypted" >> ${PILLARS_TOP}
 928             sed -i "s/__CERT_REQUIRES__/file: extra_custom_certs_${R}_cert_file_copy/g;
 929                     s#__CERT_PEM__#/etc/nginx/ssl/arvados-${R}.pem#g;
 930                     s#__CERT_KEY__#/etc/nginx/ssl/arvados-${R}.key#g" \
 931             ${P_DIR}/nginx_${R}_configuration.sls
 932             grep -q ${R} ${P_DIR}/extra_custom_certs.sls || echo "  - ${R}" >> ${P_DIR}/extra_custom_certs.sls
 933           fi
 934         fi
 935         # We need to tweak the Nginx's pillar depending whether we want plain nginx or nginx+passenger
 936         NGINX_INSTALL_SOURCE="install_from_phusionpassenger"
 937         sed -i "s/__NGINX_INSTALL_SOURCE__/${NGINX_INSTALL_SOURCE}/g" ${P_DIR}/nginx_passenger.sls
 938       ;;
 939       "websocket" | "workbench" | "workbench2" | "webshell" | "keepweb" | "keepproxy")
 940         ### States ###
 941         grep -q "\- nginx$" ${STATES_TOP} || echo "    - nginx" >> ${STATES_TOP}
 942
 943         if [ "${SSL_MODE}" = "lets-encrypt" ]; then
 944           if [ "x${USE_LETSENCRYPT_ROUTE53:-}" = "xyes" ]; then
 945             grep -q "aws_credentials" ${STATES_TOP} || echo "    - aws_credentials" >> ${STATES_TOP}
 946           fi
 947           grep -q "letsencrypt"     ${STATES_TOP} || echo "    - letsencrypt" >> ${STATES_TOP}
 948         else
 949           # Use custom certs, special case for keepweb
 950           if [ ${R} = "keepweb" ]; then
 951             if [ "${SSL_MODE}" = "bring-your-own" ]; then
 952               copy_custom_cert ${CUSTOM_CERTS_DIR} download
 953               copy_custom_cert ${CUSTOM_CERTS_DIR} collections
 954             fi
 955           else
 956             if [ "${SSL_MODE}" = "bring-your-own" ]; then
 957               copy_custom_cert ${CUSTOM_CERTS_DIR} ${R}
 958             fi
 959           fi
 960         fi
 961
 962         # webshell role is just a nginx vhost, so it has no state
 963         # workbench role is deprecated since 2.7.0
 964         if [[ "${R}" != "webshell" && "${R}" != "workbench" ]]; then
 965           grep -q "arvados.${R}" ${STATES_TOP} || echo "    - arvados.${R}" >> ${STATES_TOP}
 966         fi
 967         # Make sure wb1's package get uninstalled
 968         if [[ "${R}" == "workbench" ]]; then
 969           grep -q "workbench1_uninstall" ${STATES_TOP} || echo "    - extra.workbench1_uninstall" >> ${STATES_TOP}
 970         fi
 971
 972         ### Pillars ###
 973         grep -q "nginx_${R}_configuration" ${PILLARS_TOP} || echo "    - nginx_${R}_configuration" >> ${PILLARS_TOP}
 974         grep -q "nginx_snippets" ${PILLARS_TOP} || echo "    - nginx_snippets" >> ${PILLARS_TOP}
 975         # Special case for keepweb
 976         if [ ${R} = "keepweb" ]; then
 977           grep -q "nginx_download_configuration" ${PILLARS_TOP} || echo "    - nginx_download_configuration" >> ${PILLARS_TOP}
 978           grep -q "nginx_collections_configuration" ${PILLARS_TOP} || echo "    - nginx_collections_configuration" >> ${PILLARS_TOP}
 979         fi
 980
 981         if [ "${SSL_MODE}" = "lets-encrypt" ]; then
 982           if [ "${USE_LETSENCRYPT_ROUTE53}" = "yes" ]; then
 983             grep -q "aws_credentials" ${PILLARS_TOP} || echo "    - aws_credentials" >> ${PILLARS_TOP}
 984           fi
 985           grep -q "letsencrypt"     ${PILLARS_TOP} || echo "    - letsencrypt" >> ${PILLARS_TOP}
 986           grep -q "letsencrypt_${R}_configuration" ${PILLARS_TOP} || echo "    - letsencrypt_${R}_configuration" >> ${PILLARS_TOP}
 987
 988           # As the pillar differ whether we use LE or custom certs, we need to do a final edition on them
 989           # Special case for keepweb
 990           if [ ${R} = "keepweb" ]; then
 991             for kwsub in download collections; do
 992               sed -i "s/__CERT_REQUIRES__/cmd: create-initial-cert-${kwsub}.${DOMAIN}*/g;
 993                       s#__CERT_PEM__#/etc/letsencrypt/live/${kwsub}.${DOMAIN}/fullchain.pem#g;
 994                       s#__CERT_KEY__#/etc/letsencrypt/live/${kwsub}.${DOMAIN}/privkey.pem#g" \
 995               ${P_DIR}/nginx_${kwsub}_configuration.sls
 996             done
 997           else
 998             sed -i "s/__CERT_REQUIRES__/cmd: create-initial-cert-${R}.${DOMAIN}*/g;
 999                     s#__CERT_PEM__#/etc/letsencrypt/live/${R}.${DOMAIN}/fullchain.pem#g;
1000                     s#__CERT_KEY__#/etc/letsencrypt/live/${R}.${DOMAIN}/privkey.pem#g" \
1001             ${P_DIR}/nginx_${R}_configuration.sls
1002           fi
1003         else
1004           grep -q "ssl_key_encrypted" ${PILLARS_TOP} || echo "    - ssl_key_encrypted" >> ${PILLARS_TOP}
1005           # As the pillar differ whether we use LE or custom certs, we need to do a final edition on them
1006           # Special case for keepweb
1007           if [ ${R} = "keepweb" ]; then
1008             for kwsub in download collections; do
1009               sed -i "s/__CERT_REQUIRES__/file: extra_custom_certs_${kwsub}_cert_file_copy/g;
1010                       s#__CERT_PEM__#/etc/nginx/ssl/arvados-${kwsub}.pem#g;
1011                       s#__CERT_KEY__#/etc/nginx/ssl/arvados-${kwsub}.key#g" \
1012               ${P_DIR}/nginx_${kwsub}_configuration.sls
1013               grep -q ${kwsub} ${P_DIR}/extra_custom_certs.sls || echo "  - ${kwsub}" >> ${P_DIR}/extra_custom_certs.sls
1014             done
1015           else
1016             sed -i "s/__CERT_REQUIRES__/file: extra_custom_certs_${R}_cert_file_copy/g;
1017                     s#__CERT_PEM__#/etc/nginx/ssl/arvados-${R}.pem#g;
1018                     s#__CERT_KEY__#/etc/nginx/ssl/arvados-${R}.key#g" \
1019             ${P_DIR}/nginx_${R}_configuration.sls
1020             grep -q ${R} ${P_DIR}/extra_custom_certs.sls || echo "  - ${R}" >> ${P_DIR}/extra_custom_certs.sls
1021           fi
1022         fi
1023         # We need to tweak the Nginx's pillar depending whether we want plain nginx or nginx+passenger
1024         sed -i "s/__NGINX_INSTALL_SOURCE__/${NGINX_INSTALL_SOURCE}/g" ${P_DIR}/nginx_passenger.sls
1025       ;;
1026       "shell")
1027         # States
1028         echo "    - extra.shell_sudo_passwordless" >> ${STATES_TOP}
1029         echo "    - extra.shell_cron_add_login_sync" >> ${STATES_TOP}
1030         grep -q "docker" ${STATES_TOP}       || echo "    - docker.software" >> ${STATES_TOP}
1031         grep -q "arvados.${R}" ${STATES_TOP} || echo "    - arvados.${R}" >> ${STATES_TOP}
1032         # Pillars
1033         grep -q "docker" ${PILLARS_TOP}       || echo "    - docker" >> ${PILLARS_TOP}
1034       ;;
1035       "dispatcher" | "keepbalance" | "keepstore")
1036         # States
1037         grep -q "arvados.${R}" ${STATES_TOP} || echo "    - arvados.${R}" >> ${STATES_TOP}
1038         # Pillars
1039         # ATM, no specific pillar needed
1040       ;;
1041       *)
1042         echo "Unknown role ${R}"
1043         exit 1
1044       ;;
1045     esac
1046   done
1047 fi
1048
1049 if [ "${DUMP_CONFIG}" = "yes" ]; then
1050   # We won't run the rest of the script because we're just dumping the config
1051   exit 0
1052 fi
1053
1054 # Now run the install
1055 salt-call --state-output=mixed --local state.apply -l ${LOG_LEVEL}
1056
1057 # Finally, make sure that /etc/hosts is not overwritten on reboot
1058 if [ -d /etc/cloud/cloud.cfg.d ]; then
1059   # TODO: will this work on CentOS?
1060   sed -i 's/^manage_etc_hosts: true/#manage_etc_hosts: true/g' /etc/cloud/cloud.cfg.d/*
1061 fi
1062
1063 # Leave a copy of the Arvados CA so the user can copy it where it's required
1064 if [ "${SSL_MODE}" = "self-signed" ]; then
1065   echo "Copying the Arvados CA certificate '${DOMAIN}-arvados-snakeoil-ca.crt' to the installer dir, so you can import it"
1066   if [ "x${VAGRANT:-}" = "xyes" ]; then
1067     cp /etc/ssl/certs/arvados-snakeoil-ca.pem /vagrant/${DOMAIN}-arvados-snakeoil-ca.pem
1068   else
1069     cp /etc/ssl/certs/arvados-snakeoil-ca.pem ${SCRIPT_DIR}/${DOMAIN}-arvados-snakeoil-ca.crt
1070   fi
1071 fi
1072
1073 if [ "x${VAGRANT:-}" = "xyes" ]; then
1074     # If running in a vagrant VM, also add default user to docker group
1075     echo "Adding the vagrant user to the docker group"
1076     usermod -a -G docker vagrant
1077 fi
1078
1079 # Test that the installation finished correctly
1080 if [ "x${TEST:-}" = "xyes" ]; then
1081   cd ${T_DIR}
1082   ./run-test.sh
1083 fi