ctctl/ctctl

#!/usr/bin/env bash

# Print current environnement
current_env(){
  if [ -z "${CTCTL_DOMAIN}" ]; then
    echo "Unknown container domain"
    kill -INT $$
  fi
  echo "Cluster: ${CTCTL_DOMAIN}"
  if [ -z "${CTCTL_ENV}" ]; then
    echo "Unknown container environment"
    kill -INT $$
  fi
  echo "Namespace: ${CTCTL_ENV}"
}

check_env() {
  if [ -n "${CTCTL_DOMAIN}" -a -n "${CTCTL_ENV}" ]; then
    echo 1
  else
    echo 0
  fi
}

# Run a shell in a container
# TODO : to implement
enter_ct(){
  echo "Select the job"
  select J in $(get_job_list); do
    if [ "${REPLY}" -ge 1 ] && [ "${REPLY}" -le $(get_job_list | wc -w) ]; then
      JOB=${J}
      break
    else
      echo "Invalid selection"
    fi
  done
}


load_config(){
  if [ -n "${CTCTL_DOMAIN}" -a -n "${CTCTL_ENV}" ]; then
    # Load env configuration
    if [ -e ~/.ctctl/${TARGET_DOM}/${CTCTL_ENV}.conf ]; then
      set -o allexport
      source ~/.ctctl/${TARGET_DOM}/${CTCTL_ENV}.conf
      set +o allexport
    fi

    # Load post login configuration
    if [ -e ~/.ctctl/${CTCTL_DOMAIN}/ctctl.local.conf ]; then
      set -o allexport
      source ~/.ctctl/${CTCTL_DOMAIN}/ctctl.local.conf
      set +o allexport
    fi

  fi
}

# Switch to a target environment (either from no current, or from another current env)
switch_env(){
  TARGET_DOM=$1
  TARGET_NAMESPACE=$2

  if [ -z "${TARGET_DOM}" ]; then
    echo "Select the container install you want to work on"
    TARGET_DOM=$(ls_env | ${FZF_CMD})
  fi

  if [ ! -e ~/.ctctl/${TARGET_DOM}/ctctl.conf ]; then
    echo "Env ${TARGET_DOM} doesn't exist"
    kill -INT $$
  fi

  # Clean any variable
  for VAR in $(printenv | perl -ne  '/^((CTCTL|CONSUL|VAULT|NOMAD)_[^=]+)=.*/ && print "$1\n"'); do
    unset ${VAR}
  done

  export CTCTL_DOMAIN=${TARGET_DOM}

  # Load default config
  set -o allexport
  source ~/.ctctl/${CTCTL_DOMAIN}/ctctl.conf
  set +o allexport

  # Load pre login env configuration
  if [ -e ~/.ctctl/${CTCTL_DOMAIN}/pre-login.conf ]; then
    set -o allexport
    source ~/.ctctl/${CTCTL_DOMAIN}/pre-login.conf
    set +o allexport
  fi

  # Authenticate
  auth_env

  local NAMESPACE_LIST=$(ls_namespace)

  if [ -z "${TARGET_NAMESPACE}" ]; then
    if [ $(echo -n ${NAMESPACE_LIST} | wc -w) -eq 1 ]; then
      TARGET_NAMESPACE=${NAMESPACE_LIST}
    else
      echo "Select the namespace you are working in"
      TARGET_NAMESPACE=$(ls_namespace | ${FZF_CMD})
    fi
  fi
  export CTCTL_ENV=${TARGET_NAMESPACE}

  # TODO : decide if we keep NOMAD_VAR_env
  export NOMAD_VAR_env=${TARGET_NAMESPACE}
  export NOMAD_NAMESPACE=${TARGET_NAMESPACE}
}

# Check if we have a valid token for vault
check_vault_token(){
  vault read auth/token/lookup-self > /dev/null 2>&1
  if [ $? == 0 ]; then
    echo 1
  else
    echo 0
  fi
}

# Check if we have a valid token for consul
check_consul_token(){
  CONSUL_TOKEN_VALID=0
  if [ -n "${CONSUL_HTTP_TOKEN}" ]; then
    consul acl token read -self > /dev/null 2>&1
    if [ $? == 0 ]; then
      echo 1
    else
      echo 0
    fi
  else
    echo 0
  fi
}

# Check if we have a valid token for nomad
check_nomad_token(){
  if [ -n "${NOMAD_TOKEN}" ]; then
    nomad acl token self > /dev/null 2>&1
    if [ $? == 0 ]; then
      echo 1
    else
      echo 0
    fi
  else
    echo 0
  fi
}

# Auth on vault, consul and nomad on the current env
auth_env(){
  if [ -z "${CTCTL_DOMAIN}" ]; then
    echo "Unknown environment"
    kill -INT $$
  fi
  if [ "$(check_vault_token)" != "1" ]; then
    echo "You're not connected on vault. Please enter your account password"
    export VAULT_TOKEN=$(vault login -field=token ${VAULT_AUTH_CONFIG:--method=ldap username=${CTCTL_USER:-$(whoami | sed -r 's/\@.*//')}} || kill -INT $$)
    echo "Logged on vault successfuly"
  else
    echo "Your vault token is valid"
    vault token renew > /dev/null 2>&1
  fi
  VAULT_TOKEN_INFO=$(vault read -format=json auth/token/lookup-self)
  # TODO make the role selection more generic
  if [ "$(echo $VAULT_TOKEN_INFO | jq '.data.policies | any(. == "admin-policy" or .== "admin")')" == "true" ]; then
    NOMAD_ROLE=admin
    CONSUL_ROLE=admin
  else
    NOMAD_ROLE=user
    CONSUL_ROLE=user
  fi

  # Root CA
  vault read -field certificate pki/root/cert/ca > ~/.ctctl/${CTCTL_DOMAIN}/root_ca.crt

  # Consul certificate
  # Get/renew cert if required.
  # Note 1: as the template is using pkiCert, the cert won't be renewed, unless necessary
  # Note 2: don't pass CONSUL_CLIENT_CERT CONSUL_CLIENT_KEY and CONSUL_CACERT because they would prevent consul-template from starting
  # to get/renew the cert if they are absent, or expired
  env -u CONSUL_CLIENT_CERT \
      -u CONSUL_CLIENT_KEY \
      -u CONSUL_CACERT \
    consul-template -config ~/.ctctl/${CTCTL_DOMAIN}/consul/consul-template.hcl -once

  # Get/renew cert for Nomad now
  consul-template -config ~/.ctctl/${CTCTL_DOMAIN}/nomad/consul-template.hcl -once

  # Check if we have a valid nomad token already
  if [ "$(check_nomad_token)" != "1" ]; then
    echo "Fecthing a Nomad token from vault"
    NOMAD_CREDS=$(vault read -format=json nomad/creds/${NOMAD_ROLE})
    export NOMAD_TOKEN=$(echo -n ${NOMAD_CREDS} | jq -r .data.secret_id)
    export NOMAD_LEASE=$(echo -n ${NOMAD_CREDS} | jq -r .lease_id)
    unset NOMAD_CREDS
  else
    echo "Nomad token is valid, renewing lease"
    vault lease renew ${NOMAD_LEASE} >/dev/null
  fi
  # Check if we have a valid consul token already
  if [ "$(check_consul_token)" != "1" ]; then
    echo "Fetching a Consul token from vault"
    CONSUL_CREDS=$(vault read -format=json consul/creds/${CONSUL_ROLE})
    export CONSUL_HTTP_TOKEN=$(echo -n ${CONSUL_CREDS} | jq -r .data.token)
    export CONSUL_LEASE=$(echo -n ${CONSUL_CREDS} | jq -r .lease_id)
    unset CONSUL_CREDS
  else
    echo "Consul token is valid, renewing lease"
    vault lease renew ${CONSUL_LEASE} >/dev/null
  fi

  load_config
}

renew_leases(){
  # Renew vault token
  [ -n "${VAULT_TOKEN}" ] && vault token renew >/dev/null
  [ -n "${NOMAD_LEASE}" ] && vault lease renew ${NOMAD_LEASE} >/dev/null
  [ -n "${CONSUL_LEASE}" ] && vault lease renew ${CONSUL_LEASE} > /dev/null
}

# Logout from the current env
logout_env(){
  if [ -z "${CTCTL_DOMAIN}" ]; then
    echo "Unknown environment"
    kill -INT $$
  fi
  echo "Disconecting from ${CTCTL_DOMAIN} environment"
  vault token revoke -self
  for VAR in $(printenv | perl -ne '/^((CTCTL|CONSUL|VAULT|NOMAD|LOKI)_[^=]+)=.*/ && print "$1\n"'); do
    unset $VAR
  done
  rm -f ~/.vault-token
}

# List available env
ls_env(){
  find ~/.ctctl/ -name ctctl.conf | xargs dirname | xargs basename -a
}

# List available namespaces
ls_namespace(){
  nomad namespace list -json | jq -r ".[] | .Name"
}

# List buildable Docker images
ls_build_docker_images(){
  (for JOB in $(find . -maxdepth 1 \( -name \*.nomad -o -name \*.nomad.hcl \)); do
       nomad run -output $JOB | jq '.Job.TaskGroups' | jq '.[] | .Tasks' | jq -r '.[] | .Config.image' 2>/dev/null
     done) | grep -E "${DOCKER_BUILD_REPO_REGEX:-docker-repo.ehtrace.com}" | sort -u
}

# Load policies for vault, Consul and Nomad
load_policies(){
  if [ "$(check_env)" = "0" ]; then
    echo "Not currently in a valid env. Run ctctl (with no argument) and select your env first"
    kill -INT $$
  fi
  if [ -d "./vault/policies" ]; then
    if [ "$(check_vault_token)" != "1" ]; then
      echo "No valid vault token. You have to authenticate first"
      kill -INT $$
    fi
    for PFILE in $(ls ./vault/policies/*.hcl 2>/dev/null); do
      PNAME=$(basename ${PFILE} .hcl)$(get_conf "env_suffix")
      echo "Loading vault policy ${PNA}"
      replace_conf_var ${PFILE} | vault policy write ${PNAME} -
    done
  fi
  if [ -d "./consul/policies" ]; then
    if [ "$(check_consul_token)" != "1" ]; then
      echo "No valid consul token. You have to authenticate first"
      kill -INT $$
    fi
    CONSUL_CUR_POLICIES=$(consul acl policy list -format=json)
    for PFILE in $(ls ./consul/policies/*.hcl 2>/dev/null); do
      PNAME=$(basename ${PFILE} .hcl)$(get_conf "env_suffix")
      # Consul do not use the same command to create a new policy and to update an existing one
      # so we need to detect if the policy already exists
      if [ "$(echo ${CONSUL_CUR_POLICIES} | jq -r '.[] | select(.Name=='\"${PNAME}\"') | .Name')" == "${PNAME}" ]; then
        echo "Updating consul policy ${PNAME}"
        replace_conf_var ${PFILE} | consul acl policy update -name=${PNAME} -rules=-
      else
        echo "Adding new consul policy ${PNAME}"
        replace_conf_var ${PFILE} | consul acl policy create -name=${PNAME} -rules=-
      fi
    done
  fi
  if [ -d "./nomad/policies" ]; then
    if [ "$(check_nomad_token)" != "1" ]; then
      echo "No valid nomad token. You have to authenticate first"
      kill -INT $$
    fi
    for PFILE in $(ls ./nomad/policies/*.hcl 2>/dev/null); do
      PNAME=$(basename ${PFILE} .hcl)$(get_conf "env_suffix")
      echo "Loading Nomad policy ${PNAME}"
      replace_conf_var ${PFILE} | nomad acl policy apply ${PNAME} -
    done
  fi
}

# Load consul services
load_consul_services(){
  if [ -d "./consul/services" ]; then
    if [ "$(check_consul_token)" != "1" ]; then
      echo "No valid consul token. You have to authenticate first"
      kill -INT $$
    fi
    for FILE in $(ls ./consul/services/*.hcl 2>/dev/null); do
      echo "Registering service from ${FILE}"
      TEMP=$(mktemp).hcl
      replace_conf_var ${FILE} > ${TEMP}
      consul services register ${TEMP}
      rm -f ${TEMP}
    done
  fi
}

# Load consul config
load_consul_conf(){
  if [ -d "./consul/config" ]; then
    if [ "$(check_consul_token)" != "1" ]; then
      echo "No valid consul token. You have to authenticate first"
      kill -INT $$
    fi
    # Note : service-defaults should be loaded before the others
    # but that should be the case
    for FILE in $(ls ./consul/config/*.hcl 2>/dev/null); do
      echo "Loading consul conf from ${FILE}"
      TEMP=$(mktemp)
      replace_conf_var ${FILE} > ${TEMP}
      consul config write ${TEMP}
      rm -f ${TEMP}
    done

    # Support storing consul config in subdir eg consul/config/service-defaults/foo.hcl
    # Or you can even omit service and use consul/config/defaults/bar.hcl, consul/config/intentions/bar.hcl
    for KIND in service-defaults defaults service-intentions intentions service-router router service-resolver resolver; do
      if [ -d ./consul/config/${KIND} ]; then
        for FILE in $(ls ./consul/config/${KIND}/*.hcl 2>/dev/null); do
          echo "Loading consul conf from ${FILE}"
          TEMP=$(mktemp)
          replace_conf_var ${FILE} > ${TEMP}
          consul config write ${TEMP}
          rm -f ${TEMP}
        done
      fi
    done
  fi
}

# Build all images for the current project
build_required_images(){
  for DOCKER_IMAGE in $(ls_build_docker_images); do
    if ! docker manifest inspect ${DOCKER_IMAGE} > /dev/null 2>&1; then
      build_image ${DOCKER_IMAGE}
    else
      echo "Image ${DOCKER_IMAGE} already available"
    fi
  done
}

# Build selected images
build_selected_images(){
  local NO_CACHE=$1
  for DOCKER_IMAGE in $(ls_build_docker_images | ${FZF_CMD} -m); do
    build_image "${DOCKER_IMAGE}" ${NO_CACHE}
  done
}

# Build a single image
build_image(){
  local DOCKER_IMAGE=$1
  local NO_CACHE=$2
  export DOCKER_BUILDKIT=1

  echo "Building image ${DOCKER_IMAGE}"
  # Extract the basename of the image, removing the repo and the tag
  local DIR=$(echo -n ${DOCKER_IMAGE} | sed -E 's/.+\/([^\/]+):.*/\1/')
  export DOCKER_IMAGE=${DOCKER_IMAGE}
  local LATEST=$(echo ${DOCKER_IMAGE} | sed 's/:.*/:latest/')
  if [ "${NO_CACHE}" != "" ]; then
    NO_CACHE="--no-cache"
   else
    NO_CACHE=""
  fi
  docker build ${NO_CACHE} -t ${DOCKER_IMAGE} -t ${LATEST} ${CTCTL_DOCKER_BUILD_OPTS:-} --progress=plain images/$DIR &&
    docker push ${DOCKER_IMAGE}
    docker push ${LATEST}
  unset DOCKER_BUILDKIT
}

# Run all executable in the prep.d directory
handle_prep_scripts(){
  if [ -d prep.d ]; then
    for H in $(find prep.d -type f -o -type l | sort); do
      if [ -x "${H}" ]; then
        echo "Running script ${H}"
        $H $1
      else
        echo "Skiping $H (not executable)"
      fi
    done
  fi
}

# Render templates using levant
render_templates(){
  MERGED_CONF=$(mktemp tmp.XXXXXXXX.yml)
  get_merged_conf > ${MERGED_CONF}

  handle_prep_scripts ${MERGED_CONF}

  for TEMPLATE in $(find . -type f -name \*.tpl ! -path "*templates/*"); do
    DIR=$(dirname ${TEMPLATE})
    FILE=$(basename ${TEMPLATE} .tpl)
    DEST=${DIR}/${FILE}
    echo "Rendering ${TEMPLATE} into ${DEST}"
    levant render -var-file ${MERGED_CONF} -log-level=WARN ${TEMPLATE} > ${DEST}
    nomad fmt ${DEST}
  done

  rm -f ${MERGED_CONF}
}

# Print Consul and Nomad tokens (not vault, for security reasons)
print_tokens(){
  if [ "$(check_nomad_token)" == "1" ]; then
    echo "Nomad token: ${NOMAD_TOKEN}"
  else
    echo "No valid Nomad token, you should auth with ctctl auth"
  fi
  if [ "$(check_consul_token)" == "1" ]; then
    echo "Consul token: ${CONSUL_HTTP_TOKEN}"
  else
    echo "No valid Consul token, you should auth with ctctl auth"
  fi
}

# Handle CSI volumes definition
handle_volumes(){
  if [ -d ./volumes ]; then
    for FILE in $(ls ./volumes/*.hcl 2>/dev/null); do
      echo "Handling volume definition ${FILE}"
      # Linstor volumes are just registered, while the other are created
      if [[ "$FILE" =~ ^linstor-.* ]]; then
        replace_conf_var ${FILE} | nomad volume register -
      else
        replace_conf_var ${FILE} | nomad volume create -
      fi
    done
  fi
}

# Follow current jobs logs
job_logs(){
  # Remove the first arg passed to ctctl, which is logs
  shift
  local SELECTOR
  local LOGCLI_CMD

  if [ -z "${LOKI_ADDR}" ]; then
    echo "You need to configure loki first (LOKI_ADDR, LOKI_USERNAME and LOKI_PASSWORD or LOKI_PWD_CMD)"
    kill -INT $$
  fi

  if [ -n "${LOKI_PWD_CMD}" ]; then
    export LOKI_PASSWORD=$(${LOKI_PWD_CMD})
  fi

  LOGCLI_CMD="logcli query --include-label=job --include-label=group --include-label=task"

  echo -n "$*" | grep -qP '\{.+\}' >/dev/null 2>&1
  # If a logcli filter was given, use it. Else, build one for jobs in the current dir
  if [ $? == 0 ]; then
    echo "Running ${LOGCLI_CMD} $@"
    ${LOGCLI_CMD} $@
  else
    # Exclude connect-proxy logs as it's often not wanted
    SELECTOR='{job=~"'$(get_job_list | sed 's/\s/|/g')'", task!~"connect-proxy-.+|tls-proxy|metrics-proxy"}'
    echo "Running ${LOGCLI_CMD} $@ ${SELECTOR}"
    ${LOGCLI_CMD} $@ "${SELECTOR}"
  fi
  unset LOKI_PASSWORD
}

### Helpers ###

# Merge the configuration files for the current env and return the result (as string)
get_merged_conf() {
  CONF_FILES=""
  if [ -e "./vars/defaults.yml" ]; then
    CONF_FILES="./vars/defaults.yml"
  fi
  if [ -e "jobs/common/vars/${CTCTL_ENV}.yml" ]; then
    CONF_FILES="${CONF_FILES} jobs/common/vars/${CTCTL_ENV}.yml"
  fi
  if [ -e "../common/vars/${CTCTL_ENV}.yml" ]; then
    CONF_FILES="${CONF_FILES} ../common/vars/${CTCTL_ENV}.yml"
  fi
  if [ -e "./vars/${CTCTL_ENV}.yml" ]; then
    CONF_FILES="${CONF_FILES} ./vars/${CTCTL_ENV}.yml"
  fi
  if [ "${CONF_FILES}" != "" ]; then
    echo "---"
    yq ea '. as $item ireduce ({}; . * $item | ... comments="")' $CONF_FILES
  else
    echo -n ""
  fi
}

# Replace ${local.conf.foo} or ${foo} with the value of foo from the various merged configuration files
# This is used to have policies (vault, consul, nomad) and config (consul intentions etc.) with variables
replace_conf_var() {
  MERGED_CONF=$(mktemp)
  get_merged_conf > $MERGED_CONF
  RES=$(cat $1 | \
    # Replace ${local.conf.foo} or ${foo} with the value of foo from the various merged configuration files \
    # This is used to have policies (vault, consul, nomad) and config (consul intentions etc.) with variables \
    perl -pe 'sub replace($) { my $val = shift; chomp(my $res = qx(yq .$val '$MERGED_CONF')); return $res; }; s!\$\{(local\.conf\.)?([^\}]+)\}! replace($2) !ge' | \
    # Replace $(foo) with the output of foo command, mainly used to fetch secrets from vault \
    perl -pe 'sub replace($) { my $val = shift; chomp(my $res = qx($val)); return $res; }; s!\$\(([^\)]+)\)! replace($1) !ge'
       )
  rm -f $MERGED_CONF
  echo "${RES}"
}

# Get a value from the conf
get_conf(){
  get_merged_conf | yq ".$1"
}

FZF_CMD=${CTCTL_FZF_CMD:-fzf --height=~10% --cycle --bind 'space:toggle' --marker='*'}

# Return a space separated list of jobs the current dir
get_job_list(){
  local JOBS=""
  for JOBFILE in $(find . -maxdepth 1 \( -name \*.nomad -o -name \*.nomad.hcl \)); do
    JOBS="${JOBS} $(nomad run -output ${JOBFILE} | jq -r '.Job.Name')"
  done
  echo $JOBS
}

case $1 in
  current)
    current_env
    renew_leases
    ;;
  auth)
    auth_env
    ;;
  disconnect)
    logout_env
    ;;
  ls|list)
    ls_env
    renew_leases
    ;;
  prep)
    render_templates
    load_policies
    load_consul_services
    load_consul_conf
    build_required_images
    renew_leases
    ;;
  volumes)
    handle_volumes
    renew_leases
    ;;
  build)
    build_selected_images
    renew_leases
    ;;
  build-no-cache)
    build_selected_images "no-cache"
    renew_leases
    ;;
  tokens)
    print_tokens
    renew_leases
    ;;
  logs)
    renew_leases
    job_logs "$@"
    ;;
  conf)
    renew_leases
    get_merged_conf
    ;;
  sh)
    enter_ct
    ;;
  switch)
    shift
    switch_env "$@"
    ;;
  *)
    switch_env "$@"
    ;;
esac