devstack/functions-common

2424 lines
71 KiB
Bash

#!/bin/bash
#
# functions-common - Common functions used by DevStack components
#
# The canonical copy of this file is maintained in the DevStack repo.
# All modifications should be made there and then sync'ed to other repos
# as required.
#
# This file is sorted alphabetically within the function groups.
#
# - Config Functions
# - Control Functions
# - Distro Functions
# - Git Functions
# - OpenStack Functions
# - Package Functions
# - Process Functions
# - Service Functions
# - System Functions
#
# The following variables are assumed to be defined by certain functions:
#
# - ``ENABLED_SERVICES``
# - ``ERROR_ON_CLONE``
# - ``FILES``
# - ``OFFLINE``
# - ``RECLONE``
# - ``REQUIREMENTS_DIR``
# - ``STACK_USER``
# - ``http_proxy``, ``https_proxy``, ``no_proxy``
#
# Save trace setting
_XTRACE_FUNCTIONS_COMMON=$(set +o | grep xtrace)
set +o xtrace
# ensure we don't re-source this in the same environment
[[ -z "$_DEVSTACK_FUNCTIONS_COMMON" ]] || return 0
declare -r -g _DEVSTACK_FUNCTIONS_COMMON=1
# Global Config Variables
declare -A -g GITREPO
declare -A -g GITBRANCH
declare -A -g GITDIR
KILL_PATH="$(which kill)"
# Save these variables to .stackenv
STACK_ENV_VARS="BASE_SQL_CONN DATA_DIR DEST ENABLED_SERVICES HOST_IP \
KEYSTONE_SERVICE_URI \
LOGFILE OS_CACERT SERVICE_HOST STACK_USER TLS_IP \
HOST_IPV6 SERVICE_IP_VERSION"
# Saves significant environment variables to .stackenv for later use
# Refers to a lot of globals, only TOP_DIR and STACK_ENV_VARS are required to
# function, the rest are simply saved and do not cause problems if they are undefined.
# save_stackenv [tag]
function save_stackenv {
local tag=${1:-""}
# Save some values we generated for later use
time_stamp=$(date "+$TIMESTAMP_FORMAT")
echo "# $time_stamp $tag" >$TOP_DIR/.stackenv
for i in $STACK_ENV_VARS; do
echo $i=${!i} >>$TOP_DIR/.stackenv
done
}
# Update/create user clouds.yaml file.
# clouds.yaml will have
# - A `devstack` entry for the `demo` user for the `demo` project.
# - A `devstack-admin` entry for the `admin` user for the `admin` project.
# write_clouds_yaml
function write_clouds_yaml {
# The location is a variable to allow for easier refactoring later to make it
# overridable. There is currently no usecase where doing so makes sense, so
# it's not currently configurable.
CLOUDS_YAML=/etc/openstack/clouds.yaml
sudo mkdir -p $(dirname $CLOUDS_YAML)
sudo chown -R $STACK_USER /etc/openstack
CA_CERT_ARG=''
if [ -f "$SSL_BUNDLE_FILE" ]; then
CA_CERT_ARG="--os-cacert $SSL_BUNDLE_FILE"
fi
# demo -> devstack
$PYTHON $TOP_DIR/tools/update_clouds_yaml.py \
--file $CLOUDS_YAML \
--os-cloud devstack \
--os-region-name $REGION_NAME \
$CA_CERT_ARG \
--os-auth-url $KEYSTONE_SERVICE_URI \
--os-username demo \
--os-password $ADMIN_PASSWORD \
--os-project-name demo
# alt_demo -> devstack-alt
$PYTHON $TOP_DIR/tools/update_clouds_yaml.py \
--file $CLOUDS_YAML \
--os-cloud devstack-alt \
--os-region-name $REGION_NAME \
$CA_CERT_ARG \
--os-auth-url $KEYSTONE_SERVICE_URI \
--os-username alt_demo \
--os-password $ADMIN_PASSWORD \
--os-project-name alt_demo
# admin -> devstack-admin
$PYTHON $TOP_DIR/tools/update_clouds_yaml.py \
--file $CLOUDS_YAML \
--os-cloud devstack-admin \
--os-region-name $REGION_NAME \
$CA_CERT_ARG \
--os-auth-url $KEYSTONE_SERVICE_URI \
--os-username admin \
--os-password $ADMIN_PASSWORD \
--os-project-name admin
# admin with a system-scoped token -> devstack-system
$PYTHON $TOP_DIR/tools/update_clouds_yaml.py \
--file $CLOUDS_YAML \
--os-cloud devstack-system-admin \
--os-region-name $REGION_NAME \
$CA_CERT_ARG \
--os-auth-url $KEYSTONE_SERVICE_URI \
--os-username admin \
--os-password $ADMIN_PASSWORD \
--os-system-scope all
cat >> $CLOUDS_YAML <<EOF
functional:
image_name: $DEFAULT_IMAGE_NAME
EOF
# CLean up any old clouds.yaml files we had laying around
rm -f $(eval echo ~"$STACK_USER")/.config/openstack/clouds.yaml
}
# trueorfalse <True|False> <VAR>
#
# Normalize config-value provided in variable VAR to either "True" or
# "False". If VAR is unset (i.e. $VAR evaluates as empty), the value
# of the second argument will be used as the default value.
#
# Accepts as False: 0 no No NO false False FALSE
# Accepts as True: 1 yes Yes YES true True TRUE
#
# usage:
# VAL=$(trueorfalse False VAL)
function trueorfalse {
local xtrace
xtrace=$(set +o | grep xtrace)
set +o xtrace
local default=$1
if [ -z $2 ]; then
die $LINENO "variable to normalize required"
fi
local testval=${!2:-}
case "$testval" in
"1" | [yY]es | "YES" | [tT]rue | "TRUE" ) echo "True" ;;
"0" | [nN]o | "NO" | [fF]alse | "FALSE" ) echo "False" ;;
* ) echo "$default" ;;
esac
$xtrace
}
function isset {
[[ -v "$1" ]]
}
# Control Functions
# =================
# Prints backtrace info
# filename:lineno:function
# backtrace level
function backtrace {
local level=$1
local deep
deep=$((${#BASH_SOURCE[@]} - 1))
echo "[Call Trace]"
while [ $level -le $deep ]; do
echo "${BASH_SOURCE[$deep]}:${BASH_LINENO[$deep-1]}:${FUNCNAME[$deep-1]}"
deep=$((deep - 1))
done
}
# Prints line number and "message" then exits
# die $LINENO "message"
function die {
local exitcode=$?
set +o xtrace
local line=$1; shift
if [ $exitcode == 0 ]; then
exitcode=1
fi
backtrace 2
err $line "$*"
# Give buffers a second to flush
sleep 1
exit $exitcode
}
# Checks an environment variable is not set or has length 0 OR if the
# exit code is non-zero and prints "message" and exits
# NOTE: env-var is the variable name without a '$'
# die_if_not_set $LINENO env-var "message"
function die_if_not_set {
local exitcode=$?
local xtrace
xtrace=$(set +o | grep xtrace)
set +o xtrace
local line=$1; shift
local evar=$1; shift
if ! is_set $evar || [ $exitcode != 0 ]; then
die $line "$*"
fi
$xtrace
}
function deprecated {
local text=$1
DEPRECATED_TEXT+="\n$text"
echo "WARNING: $text" >&2
}
# Prints line number and "message" in error format
# err $LINENO "message"
function err {
local exitcode=$?
local xtrace
xtrace=$(set +o | grep xtrace)
set +o xtrace
local msg="[ERROR] ${BASH_SOURCE[2]}:$1 $2"
echo "$msg" 1>&2;
if [[ -n ${LOGDIR} ]]; then
echo "$msg" >> "${LOGDIR}/error.log"
fi
$xtrace
return $exitcode
}
# Checks an environment variable is not set or has length 0 OR if the
# exit code is non-zero and prints "message"
# NOTE: env-var is the variable name without a '$'
# err_if_not_set $LINENO env-var "message"
function err_if_not_set {
local exitcode=$?
local xtrace
xtrace=$(set +o | grep xtrace)
set +o xtrace
local line=$1; shift
local evar=$1; shift
if ! is_set $evar || [ $exitcode != 0 ]; then
err $line "$*"
fi
$xtrace
return $exitcode
}
# Exit after outputting a message about the distribution not being supported.
# exit_distro_not_supported [optional-string-telling-what-is-missing]
function exit_distro_not_supported {
if [[ -z "$DISTRO" ]]; then
GetDistro
fi
if [ $# -gt 0 ]; then
die $LINENO "Support for $DISTRO is incomplete: no support for $@"
else
die $LINENO "Support for $DISTRO is incomplete."
fi
}
# Test if the named environment variable is set and not zero length
# is_set env-var
function is_set {
local var=\$"$1"
eval "[ -n \"$var\" ]" # For ex.: sh -c "[ -n \"$var\" ]" would be better, but several exercises depends on this
}
# Prints line number and "message" in warning format
# warn $LINENO "message"
function warn {
local exitcode=$?
local xtrace
xtrace=$(set +o | grep xtrace)
set +o xtrace
local msg="[WARNING] ${BASH_SOURCE[2]}:$1 $2"
echo "$msg"
$xtrace
return $exitcode
}
# Distro Functions
# ================
# Determine OS Vendor, Release and Update
#
# NOTE : For portability, you almost certainly do not want to use
# these variables directly! The "is_*" functions defined below this
# bundle up compatible platforms under larger umbrellas that we have
# determinted are compatible enough (e.g. is_ubuntu covers Ubuntu &
# Debian, is_fedora covers RPM-based distros). Higher-level functions
# such as "install_package" further abstract things in better ways.
#
# ``os_VENDOR`` - vendor name: ``Ubuntu``, ``Fedora``, etc
# ``os_RELEASE`` - major release: ``16.04`` (Ubuntu), ``23`` (Fedora)
# ``os_PACKAGE`` - package type: ``deb`` or ``rpm``
# ``os_CODENAME`` - vendor's codename for release: ``xenial``
declare -g os_VENDOR os_RELEASE os_PACKAGE os_CODENAME
# Make a *best effort* attempt to install lsb_release packages for the
# user if not available. Note can't use generic install_package*
# because they depend on this!
function _ensure_lsb_release {
if [[ -x $(command -v lsb_release 2>/dev/null) ]]; then
return
fi
if [[ -x $(command -v apt-get 2>/dev/null) ]]; then
sudo apt-get install -y lsb-release
elif [[ -x $(command -v zypper 2>/dev/null) ]]; then
sudo zypper -n install lsb-release
elif [[ -x $(command -v dnf 2>/dev/null) ]]; then
sudo dnf install -y redhat-lsb-core
else
die $LINENO "Unable to find or auto-install lsb_release"
fi
}
# GetOSVersion
# Set the following variables:
# - os_RELEASE
# - os_CODENAME
# - os_VENDOR
# - os_PACKAGE
function GetOSVersion {
# We only support distros that provide a sane lsb_release
_ensure_lsb_release
os_RELEASE=$(lsb_release -r -s)
os_CODENAME=$(lsb_release -c -s)
os_VENDOR=$(lsb_release -i -s)
if [[ $os_VENDOR =~ (Debian|Ubuntu|LinuxMint) ]]; then
os_PACKAGE="deb"
else
os_PACKAGE="rpm"
fi
typeset -xr os_VENDOR
typeset -xr os_RELEASE
typeset -xr os_PACKAGE
typeset -xr os_CODENAME
}
# Translate the OS version values into common nomenclature
# Sets global ``DISTRO`` from the ``os_*`` values
declare -g DISTRO
function GetDistro {
GetOSVersion
if [[ "$os_VENDOR" =~ (Ubuntu) || "$os_VENDOR" =~ (Debian) || \
"$os_VENDOR" =~ (LinuxMint) ]]; then
# 'Everyone' refers to Ubuntu / Debian / Mint releases by
# the code name adjective
DISTRO=$os_CODENAME
elif [[ "$os_VENDOR" =~ (Fedora) ]]; then
# For Fedora, just use 'f' and the release
DISTRO="f$os_RELEASE"
elif is_opensuse; then
DISTRO="opensuse-$os_RELEASE"
# Tumbleweed uses "n/a" as a codename, and the release is a datestring
# like 20180218, so not very useful. Leap however uses a release
# with a "dot", so for example 15.0
[ "$os_CODENAME" = "n/a" -a "$os_RELEASE" = "${os_RELEASE/\./}" ] && \
DISTRO="opensuse-tumbleweed"
elif is_suse_linux_enterprise; then
# just use major release
DISTRO="sle${os_RELEASE%.*}"
elif [[ "$os_VENDOR" =~ (Red.*Hat) || \
"$os_VENDOR" =~ (CentOS) || \
"$os_VENDOR" =~ (Scientific) || \
"$os_VENDOR" =~ (OracleServer) || \
"$os_VENDOR" =~ (Virtuozzo) ]]; then
# Drop the . release as we assume it's compatible
# XXX re-evaluate when we get RHEL10
DISTRO="rhel${os_RELEASE::1}"
elif [[ "$os_VENDOR" =~ (XenServer) ]]; then
DISTRO="xs${os_RELEASE%.*}"
else
# We can't make a good choice here. Setting a sensible DISTRO
# is part of the problem, but not the major issue -- we really
# only use DISTRO in the code as a fine-filter.
#
# The bigger problem is categorising the system into one of
# our two big categories as Ubuntu/Debian-ish or
# Fedora/CentOS-ish.
#
# The setting of os_PACKAGE above is only set to "deb" based
# on a hard-coded list of vendor names ... thus we will
# default to thinking unknown distros are RPM based
# (ie. is_ubuntu does not match). But the platform will then
# also not match in is_fedora, because that also has a list of
# names.
#
# So, if you are reading this, getting your distro supported
# is really about making sure it matches correctly in these
# functions. Then you can choose a sensible way to construct
# DISTRO based on your distros release approach.
die $LINENO "Unable to determine DISTRO, can not continue."
fi
typeset -xr DISTRO
}
# Utility function for checking machine architecture
# is_arch arch-type
function is_arch {
[[ "$(uname -m)" == "$1" ]]
}
# Determine if current distribution is an Oracle distribution
# is_oraclelinux
function is_oraclelinux {
if [[ -z "$os_VENDOR" ]]; then
GetOSVersion
fi
[ "$os_VENDOR" = "OracleServer" ]
}
# Determine if current distribution is a Fedora-based distribution
# (Fedora, RHEL, CentOS, etc).
# is_fedora
function is_fedora {
if [[ -z "$os_VENDOR" ]]; then
GetOSVersion
fi
[ "$os_VENDOR" = "Fedora" ] || [ "$os_VENDOR" = "Red Hat" ] || \
[ "$os_VENDOR" = "RedHatEnterpriseServer" ] || \
[ "$os_VENDOR" = "RedHatEnterprise" ] || \
[ "$os_VENDOR" = "CentOS" ] || [ "$os_VENDOR" = "OracleServer" ] || \
[ "$os_VENDOR" = "Virtuozzo" ]
}
# Determine if current distribution is a SUSE-based distribution
# (openSUSE, SLE).
# is_suse
function is_suse {
is_opensuse || is_suse_linux_enterprise
}
# Determine if current distribution is an openSUSE distribution
# is_opensuse
function is_opensuse {
if [[ -z "$os_VENDOR" ]]; then
GetOSVersion
fi
[[ "$os_VENDOR" =~ (openSUSE) ]]
}
# Determine if current distribution is a SUSE Linux Enterprise (SLE)
# distribution
# is_suse_linux_enterprise
function is_suse_linux_enterprise {
if [[ -z "$os_VENDOR" ]]; then
GetOSVersion
fi
[[ "$os_VENDOR" =~ (^SUSE) ]]
}
# Determine if current distribution is an Ubuntu-based distribution
# It will also detect non-Ubuntu but Debian-based distros
# is_ubuntu
function is_ubuntu {
if [[ -z "$os_PACKAGE" ]]; then
GetOSVersion
fi
[ "$os_PACKAGE" = "deb" ]
}
# Git Functions
# =============
# Returns openstack release name for a given branch name
# ``get_release_name_from_branch branch-name``
function get_release_name_from_branch {
local branch=$1
if [[ $branch =~ "stable/" || $branch =~ "proposed/" ]]; then
echo ${branch#*/}
else
echo "master"
fi
}
# git clone only if directory doesn't exist already. Since ``DEST`` might not
# be owned by the installation user, we create the directory and change the
# ownership to the proper user.
# Set global ``RECLONE=yes`` to simulate a clone when dest-dir exists
# Set global ``ERROR_ON_CLONE=True`` to abort execution with an error if the git repo
# does not exist (default is False, meaning the repo will be cloned).
# Uses globals ``ERROR_ON_CLONE``, ``OFFLINE``, ``RECLONE``
# git_clone remote dest-dir branch
function git_clone {
local git_remote=$1
local git_dest=$2
local git_ref=$3
local orig_dir
orig_dir=$(pwd)
local git_clone_flags=""
RECLONE=$(trueorfalse False RECLONE)
if [[ "${GIT_DEPTH}" -gt 0 ]]; then
git_clone_flags="$git_clone_flags --depth $GIT_DEPTH"
fi
if [[ "$OFFLINE" = "True" ]]; then
echo "Running in offline mode, clones already exist"
# print out the results so we know what change was used in the logs
cd $git_dest
git show --oneline | head -1
cd $orig_dir
return
fi
if echo $git_ref | egrep -q "^refs"; then
# If our branch name is a gerrit style refs/changes/...
if [[ ! -d $git_dest ]]; then
if [[ "$ERROR_ON_CLONE" = "True" ]]; then
echo "The $git_dest project was not found; if this is a gate job, add"
echo "the project to 'required-projects' in the job definition."
die $LINENO "Cloning not allowed in this configuration"
fi
git_timed clone $git_clone_flags $git_remote $git_dest
fi
cd $git_dest
git_timed fetch $git_remote $git_ref && git checkout FETCH_HEAD
else
# do a full clone only if the directory doesn't exist
if [[ ! -d $git_dest ]]; then
if [[ "$ERROR_ON_CLONE" = "True" ]]; then
echo "The $git_dest project was not found; if this is a gate job, add"
echo "the project to the \$PROJECTS variable in the job definition."
die $LINENO "Cloning not allowed in this configuration"
fi
# '--branch' can also take tags
git_timed clone $git_clone_flags $git_remote $git_dest --branch $git_ref
elif [[ "$RECLONE" = "True" ]]; then
# if it does exist then simulate what clone does if asked to RECLONE
cd $git_dest
# set the url to pull from and fetch
git remote set-url origin $git_remote
git_timed fetch origin
# remove the existing ignored files (like pyc) as they cause breakage
# (due to the py files having older timestamps than our pyc, so python
# thinks the pyc files are correct using them)
find $git_dest -name '*.pyc' -delete
# handle git_ref accordingly to type (tag, branch)
if [[ -n "`git show-ref refs/tags/$git_ref`" ]]; then
git_update_tag $git_ref
elif [[ -n "`git show-ref refs/heads/$git_ref`" ]]; then
git_update_branch $git_ref
elif [[ -n "`git show-ref refs/remotes/origin/$git_ref`" ]]; then
git_update_remote_branch $git_ref
else
die $LINENO "$git_ref is neither branch nor tag"
fi
fi
fi
# print out the results so we know what change was used in the logs
cd $git_dest
git show --oneline | head -1
cd $orig_dir
}
# A variation on git clone that lets us specify a project by it's
# actual name, like oslo.config. This is exceptionally useful in the
# library installation case
function git_clone_by_name {
local name=$1
local repo=${GITREPO[$name]}
local dir=${GITDIR[$name]}
local branch=${GITBRANCH[$name]}
git_clone $repo $dir $branch
}
# git can sometimes get itself infinitely stuck with transient network
# errors or other issues with the remote end. This wraps git in a
# timeout/retry loop and is intended to watch over non-local git
# processes that might hang. GIT_TIMEOUT, if set, is passed directly
# to timeout(1); otherwise the default value of 0 maintains the status
# quo of waiting forever.
# usage: git_timed <git-command>
function git_timed {
local count=0
local timeout=0
if [[ -n "${GIT_TIMEOUT}" ]]; then
timeout=${GIT_TIMEOUT}
fi
time_start "git_timed"
until timeout -s SIGINT ${timeout} git "$@"; do
# 124 is timeout(1)'s special return code when it reached the
# timeout; otherwise assume fatal failure
if [[ $? -ne 124 ]]; then
die $LINENO "git call failed: [git $@]"
fi
count=$(($count + 1))
warn $LINENO "timeout ${count} for git call: [git $@]"
if [ $count -eq 3 ]; then
die $LINENO "Maximum of 3 git retries reached"
fi
sleep 5
done
time_stop "git_timed"
}
# git update using reference as a branch.
# git_update_branch ref
function git_update_branch {
local git_branch=$1
git checkout -f origin/$git_branch
# a local branch might not exist
git branch -D $git_branch || true
git checkout -b $git_branch
}
# git update using reference as a branch.
# git_update_remote_branch ref
function git_update_remote_branch {
local git_branch=$1
git checkout -b $git_branch -t origin/$git_branch
}
# git update using reference as a tag. Be careful editing source at that repo
# as working copy will be in a detached mode
# git_update_tag ref
function git_update_tag {
local git_tag=$1
git tag -d $git_tag
# fetching given tag only
git_timed fetch origin tag $git_tag
git checkout -f $git_tag
}
# OpenStack Functions
# ===================
# Get the default value for HOST_IP
# get_default_host_ip fixed_range floating_range host_ip_iface host_ip
function get_default_host_ip {
local fixed_range=$1
local floating_range=$2
local host_ip_iface=$3
local host_ip=$4
local af=$5
# Search for an IP unless an explicit is set by ``HOST_IP`` environment variable
if [ -z "$host_ip" -o "$host_ip" == "dhcp" ]; then
host_ip=""
# Find the interface used for the default route
host_ip_iface=${host_ip_iface:-$(ip -f $af route | awk '/default/ {print $5}' | head -1)}
local host_ips
host_ips=$(LC_ALL=C ip -f $af addr show ${host_ip_iface} | sed /temporary/d |awk /$af'/ {split($2,parts,"/"); print parts[1]}')
local ip
for ip in $host_ips; do
# Attempt to filter out IP addresses that are part of the fixed and
# floating range. Note that this method only works if the ``netaddr``
# python library is installed. If it is not installed, an error
# will be printed and the first IP from the interface will be used.
# If that is not correct set ``HOST_IP`` in ``localrc`` to the correct
# address.
if [[ "$af" == "inet6" ]]; then
host_ip=$ip
break;
fi
if ! (address_in_net $ip $fixed_range || address_in_net $ip $floating_range); then
host_ip=$ip
break;
fi
done
fi
echo $host_ip
}
# Generates hex string from ``size`` byte of pseudo random data
# generate_hex_string size
function generate_hex_string {
local size=$1
hexdump -n "$size" -v -e '/1 "%02x"' /dev/urandom
}
# Grab a numbered field from python prettytable output
# Fields are numbered starting with 1
# Reverse syntax is supported: -1 is the last field, -2 is second to last, etc.
# get_field field-number
function get_field {
local data field
while read data; do
if [ "$1" -lt 0 ]; then
field="(\$(NF$1))"
else
field="\$$(($1 + 1))"
fi
echo "$data" | awk -F'[ \t]*\\|[ \t]*' "{print $field}"
done
}
# install default policy
# copy over a default policy.json and policy.d for projects
function install_default_policy {
local project=$1
local project_uc
project_uc=$(echo $1|tr a-z A-Z)
local conf_dir="${project_uc}_CONF_DIR"
# eval conf dir to get the variable
conf_dir="${!conf_dir}"
local project_dir="${project_uc}_DIR"
# eval project dir to get the variable
project_dir="${!project_dir}"
local sample_conf_dir="${project_dir}/etc/${project}"
local sample_policy_dir="${project_dir}/etc/${project}/policy.d"
# first copy any policy.json
cp -p $sample_conf_dir/policy.json $conf_dir
# then optionally copy over policy.d
if [[ -d $sample_policy_dir ]]; then
cp -r $sample_policy_dir $conf_dir/policy.d
fi
}
# Add a policy to a policy.json file
# Do nothing if the policy already exists
# ``policy_add policy_file policy_name policy_permissions``
function policy_add {
local policy_file=$1
local policy_name=$2
local policy_perm=$3
if grep -q ${policy_name} ${policy_file}; then
echo "Policy ${policy_name} already exists in ${policy_file}"
return
fi
# Add a terminating comma to policy lines without one
# Remove the closing '}' and all lines following to the end-of-file
local tmpfile
tmpfile=$(mktemp)
uniq ${policy_file} | sed -e '
s/]$/],/
/^[}]/,$d
' > ${tmpfile}
# Append policy and closing brace
echo " \"${policy_name}\": ${policy_perm}" >>${tmpfile}
echo "}" >>${tmpfile}
mv ${tmpfile} ${policy_file}
}
# Gets or creates a domain
# Usage: get_or_create_domain <name> <description>
function get_or_create_domain {
local domain_id
# Gets domain id
domain_id=$(
# Gets domain id
openstack domain show $1 \
-f value -c id 2>/dev/null ||
# Creates new domain
openstack domain create $1 \
--description "$2" \
-f value -c id
)
echo $domain_id
}
# Gets or creates group
# Usage: get_or_create_group <groupname> <domain> [<description>]
function get_or_create_group {
local desc="${3:-}"
local group_id
# Gets group id
group_id=$(
# Creates new group with --or-show
openstack group create $1 \
--domain $2 --description "$desc" --or-show \
-f value -c id
)
echo $group_id
}
# Gets or creates user
# Usage: get_or_create_user <username> <password> <domain> [<email>]
function get_or_create_user {
local user_id
if [[ ! -z "$4" ]]; then
local email="--email=$4"
else
local email=""
fi
# Gets user id
user_id=$(
# Creates new user with --or-show
openstack user create \
$1 \
--password "$2" \
--domain=$3 \
$email \
--or-show \
-f value -c id
)
echo $user_id
}
# Gets or creates project
# Usage: get_or_create_project <name> <domain>
function get_or_create_project {
local project_id
project_id=$(
# Creates new project with --or-show
openstack project create $1 \
--domain=$2 \
--or-show -f value -c id
)
echo $project_id
}
# Gets or creates role
# Usage: get_or_create_role <name>
function get_or_create_role {
local role_id
role_id=$(
# Creates role with --or-show
openstack role create $1 \
--or-show -f value -c id
)
echo $role_id
}
# Returns the domain parts of a function call if present
# Usage: _get_domain_args [<user_domain> <project_domain>]
function _get_domain_args {
local domain
domain=""
if [[ -n "$1" ]]; then
domain="$domain --user-domain $1"
fi
if [[ -n "$2" ]]; then
domain="$domain --project-domain $2"
fi
echo $domain
}
# Gets or adds user role to project
# Usage: get_or_add_user_project_role <role> <user> <project> [<user_domain> <project_domain>]
function get_or_add_user_project_role {
local user_role_id
domain_args=$(_get_domain_args $4 $5)
# Gets user role id
user_role_id=$(openstack role assignment list \
--role $1 \
--user $2 \
--project $3 \
$domain_args \
| grep '^|\s[a-f0-9]\+' | get_field 1)
if [[ -z "$user_role_id" ]]; then
# Adds role to user and get it
openstack role add $1 \
--user $2 \
--project $3 \
$domain_args
user_role_id=$(openstack role assignment list \
--role $1 \
--user $2 \
--project $3 \
$domain_args \
| grep '^|\s[a-f0-9]\+' | get_field 1)
fi
echo $user_role_id
}
# Gets or adds user role to domain
# Usage: get_or_add_user_domain_role <role> <user> <domain>
function get_or_add_user_domain_role {
local user_role_id
# Gets user role id
user_role_id=$(openstack role assignment list \
--role $1 \
--user $2 \
--domain $3 \
| grep '^|\s[a-f0-9]\+' | get_field 1)
if [[ -z "$user_role_id" ]]; then
# Adds role to user and get it
openstack role add $1 \
--user $2 \
--domain $3
user_role_id=$(openstack role assignment list \
--role $1 \
--user $2 \
--domain $3 \
| grep '^|\s[a-f0-9]\+' | get_field 1)
fi
echo $user_role_id
}
# Gets or adds group role to project
# Usage: get_or_add_group_project_role <role> <group> <project>
function get_or_add_group_project_role {
local group_role_id
# Gets group role id
group_role_id=$(openstack role assignment list \
--role $1 \
--group $2 \
--project $3 \
-f value)
if [[ -z "$group_role_id" ]]; then
# Adds role to group and get it
openstack role add $1 \
--group $2 \
--project $3
group_role_id=$(openstack role assignment list \
--role $1 \
--group $2 \
--project $3 \
-f value)
fi
echo $group_role_id
}
# Gets or creates service
# Usage: get_or_create_service <name> <type> <description>
function get_or_create_service {
local service_id
# Gets service id
service_id=$(
# Gets service id
openstack service show $2 -f value -c id 2>/dev/null ||
# Creates new service if not exists
openstack service create \
$2 \
--name $1 \
--description="$3" \
-f value -c id
)
echo $service_id
}
# Create an endpoint with a specific interface
# Usage: _get_or_create_endpoint_with_interface <service> <interface> <url> <region>
function _get_or_create_endpoint_with_interface {
local endpoint_id
endpoint_id=$(openstack endpoint list \
--service $1 \
--interface $2 \
--region $4 \
-c ID -f value)
if [[ -z "$endpoint_id" ]]; then
# Creates new endpoint
endpoint_id=$(openstack endpoint create \
$1 $2 $3 --region $4 -f value -c id)
fi
echo $endpoint_id
}
# Gets or creates endpoint
# Usage: get_or_create_endpoint <service> <region> <publicurl> [adminurl] [internalurl]
function get_or_create_endpoint {
# NOTE(jamielennnox): when converting to v3 endpoint creation we go from
# creating one endpoint with multiple urls to multiple endpoints each with
# a different interface. To maintain the existing function interface we
# create 3 endpoints and return the id of the public one. In reality
# returning the public id will not make a lot of difference as there are no
# scenarios currently that use the returned id. Ideally this behaviour
# should be pushed out to the service setups and let them create the
# endpoints they need.
local public_id
public_id=$(_get_or_create_endpoint_with_interface $1 public $3 $2)
# only create admin/internal urls if provided content for them
if [[ -n "$4" ]]; then
_get_or_create_endpoint_with_interface $1 admin $4 $2
fi
if [[ -n "$5" ]]; then
_get_or_create_endpoint_with_interface $1 internal $5 $2
fi
# return the public id to indicate success, and this is the endpoint most likely wanted
echo $public_id
}
# Get a URL from the identity service
# Usage: get_endpoint_url <service> <interface>
function get_endpoint_url {
echo $(openstack endpoint list \
--service $1 --interface $2 \
-c URL -f value)
}
# check if we are using ironic with hardware
# TODO(jroll) this is a kludge left behind when ripping ironic code
# out of tree, as it is used by nova and neutron.
# figure out a way to refactor nova/neutron code to eliminate this
function is_ironic_hardware {
is_service_enabled ironic && [[ "$IRONIC_IS_HARDWARE" == "True" ]] && return 0
return 1
}
# Package Functions
# =================
# _get_package_dir
function _get_package_dir {
local base_dir=$1
local pkg_dir
if [[ -z "$base_dir" ]]; then
base_dir=$FILES
fi
if is_ubuntu; then
pkg_dir=$base_dir/debs
elif is_fedora; then
pkg_dir=$base_dir/rpms
elif is_suse; then
pkg_dir=$base_dir/rpms-suse
else
exit_distro_not_supported "list of packages"
fi
echo "$pkg_dir"
}
# Wrapper for ``apt-get update`` to try multiple times on the update
# to address bad package mirrors (which happen all the time).
function apt_get_update {
# only do this once per run
if [[ "$REPOS_UPDATED" == "True" && "$RETRY_UPDATE" != "True" ]]; then
return
fi
# bail if we are offline
[[ "$OFFLINE" = "True" ]] && return
local sudo="sudo"
[[ "$(id -u)" = "0" ]] && sudo="env"
# time all the apt operations
time_start "apt-get-update"
local proxies="http_proxy=${http_proxy:-} https_proxy=${https_proxy:-} no_proxy=${no_proxy:-} "
local update_cmd="$sudo $proxies apt-get update"
if ! timeout 300 sh -c "while ! $update_cmd; do sleep 30; done"; then
die $LINENO "Failed to update apt repos, we're dead now"
fi
REPOS_UPDATED=True
# stop the clock
time_stop "apt-get-update"
}
# Wrapper for ``apt-get`` to set cache and proxy environment variables
# Uses globals ``OFFLINE``, ``*_proxy``
# apt_get operation package [package ...]
function apt_get {
local xtrace result
xtrace=$(set +o | grep xtrace)
set +o xtrace
[[ "$OFFLINE" = "True" || -z "$@" ]] && return
local sudo="sudo"
[[ "$(id -u)" = "0" ]] && sudo="env"
# time all the apt operations
time_start "apt-get"
$xtrace
$sudo DEBIAN_FRONTEND=noninteractive \
http_proxy=${http_proxy:-} https_proxy=${https_proxy:-} \
no_proxy=${no_proxy:-} \
apt-get --option "Dpkg::Options::=--force-confold" --assume-yes "$@" < /dev/null
result=$?
# stop the clock
time_stop "apt-get"
return $result
}
function _parse_package_files {
local files_to_parse=$@
if [[ -z "$DISTRO" ]]; then
GetDistro
fi
for fname in ${files_to_parse}; do
local OIFS line package distros distro
[[ -e $fname ]] || continue
OIFS=$IFS
IFS=$'\n'
for line in $(<${fname}); do
if [[ $line =~ "NOPRIME" ]]; then
continue
fi
# Assume we want this package; free-form
# comments allowed after a #
package=${line%%#*}
inst_pkg=1
# Look for # dist:xxx in comment
if [[ $line =~ (.*)#.*dist:([^ ]*) ]]; then
# We are using BASH regexp matching feature.
package=${BASH_REMATCH[1]}
distros=${BASH_REMATCH[2]}
# In bash ${VAR,,} will lowercase VAR
# Look for a match in the distro list
if [[ ! ${distros,,} =~ ${DISTRO,,} ]]; then
# If no match then skip this package
inst_pkg=0
fi
fi
# Look for # not:xxx in comment
if [[ $line =~ (.*)#.*not:([^ ]*) ]]; then
# We are using BASH regexp matching feature.
package=${BASH_REMATCH[1]}
distros=${BASH_REMATCH[2]}
# In bash ${VAR,,} will lowercase VAR
# Look for a match in the distro list
if [[ ${distros,,} =~ ${DISTRO,,} ]]; then
# If match then skip this package
inst_pkg=0
fi
fi
if [[ $inst_pkg = 1 ]]; then
echo $package
fi
done
IFS=$OIFS
done
}
# get_packages() collects a list of package names of any type from the
# prerequisite files in ``files/{debs|rpms}``. The list is intended
# to be passed to a package installer such as apt or yum.
#
# Only packages required for the services in 1st argument will be
# included. Two bits of metadata are recognized in the prerequisite files:
#
# - ``# NOPRIME`` defers installation to be performed later in `stack.sh`
# - ``# dist:DISTRO`` or ``dist:DISTRO1,DISTRO2`` limits the selection
# of the package to the distros listed. The distro names are case insensitive.
# - ``# not:DISTRO`` or ``not:DISTRO1,DISTRO2`` limits the selection
# of the package to the distros not listed. The distro names are case insensitive.
function get_packages {
local xtrace
xtrace=$(set +o | grep xtrace)
set +o xtrace
local services=$@
local package_dir
package_dir=$(_get_package_dir)
local file_to_parse=""
local service=""
if [ $# -ne 1 ]; then
die $LINENO "get_packages takes a single, comma-separated argument"
fi
if [[ -z "$package_dir" ]]; then
echo "No package directory supplied"
return 1
fi
for service in ${services//,/ }; do
# Allow individual services to specify dependencies
if [[ -e ${package_dir}/${service} ]]; then
file_to_parse="${file_to_parse} ${package_dir}/${service}"
fi
# NOTE(sdague) n-api needs glance for now because that's where
# glance client is
if [[ $service == n-api ]]; then
if [[ ! $file_to_parse =~ $package_dir/nova ]]; then
file_to_parse="${file_to_parse} ${package_dir}/nova"
fi
if [[ ! $file_to_parse =~ $package_dir/glance ]]; then
file_to_parse="${file_to_parse} ${package_dir}/glance"
fi
if [[ ! $file_to_parse =~ $package_dir/os-brick ]]; then
file_to_parse="${file_to_parse} ${package_dir}/os-brick"
fi
elif [[ $service == c-* ]]; then
if [[ ! $file_to_parse =~ $package_dir/cinder ]]; then
file_to_parse="${file_to_parse} ${package_dir}/cinder"
fi
if [[ ! $file_to_parse =~ $package_dir/os-brick ]]; then
file_to_parse="${file_to_parse} ${package_dir}/os-brick"
fi
elif [[ $service == s-* ]]; then
if [[ ! $file_to_parse =~ $package_dir/swift ]]; then
file_to_parse="${file_to_parse} ${package_dir}/swift"
fi
elif [[ $service == n-* ]]; then
if [[ ! $file_to_parse =~ $package_dir/nova ]]; then
file_to_parse="${file_to_parse} ${package_dir}/nova"
fi
if [[ ! $file_to_parse =~ $package_dir/os-brick ]]; then
file_to_parse="${file_to_parse} ${package_dir}/os-brick"
fi
elif [[ $service == g-* ]]; then
if [[ ! $file_to_parse =~ $package_dir/glance ]]; then
file_to_parse="${file_to_parse} ${package_dir}/glance"
fi
elif [[ $service == key* ]]; then
if [[ ! $file_to_parse =~ $package_dir/keystone ]]; then
file_to_parse="${file_to_parse} ${package_dir}/keystone"
fi
elif [[ $service == q-* || $service == neutron-* ]]; then
if [[ ! $file_to_parse =~ $package_dir/neutron-common ]]; then
file_to_parse="${file_to_parse} ${package_dir}/neutron-common"
fi
elif [[ $service == ir-* ]]; then
if [[ ! $file_to_parse =~ $package_dir/ironic ]]; then
file_to_parse="${file_to_parse} ${package_dir}/ironic"
fi
fi
done
echo "$(_parse_package_files $file_to_parse)"
$xtrace
}
# get_plugin_packages() collects a list of package names of any type from a
# plugin's prerequisite files in ``$PLUGIN/devstack/files/{debs|rpms}``. The
# list is intended to be passed to a package installer such as apt or yum.
#
# Only packages required for enabled and collected plugins will included.
#
# The same metadata used in the main DevStack prerequisite files may be used
# in these prerequisite files, see get_packages() for more info.
function get_plugin_packages {
local xtrace
xtrace=$(set +o | grep xtrace)
set +o xtrace
local files_to_parse=""
local package_dir=""
for plugin in ${DEVSTACK_PLUGINS//,/ }; do
package_dir="$(_get_package_dir ${GITDIR[$plugin]}/devstack/files)"
files_to_parse+=" $package_dir/$plugin"
done
echo "$(_parse_package_files $files_to_parse)"
$xtrace
}
# Search plugins for a bindep.txt file
#
# Uses globals ``BINDEP_CMD``, ``GITDIR``, ``DEVSTACK_PLUGINS``
#
# Note this is only valid after BINDEP_CMD is setup in stack.sh, and
# is thus not really intended to be called externally.
function _get_plugin_bindep_packages {
local xtrace
xtrace=$(set +o | grep xtrace)
set +o xtrace
local bindep_file
local packages
for plugin in ${DEVSTACK_PLUGINS//,/ }; do
bindep_file=${GITDIR[$plugin]}/devstack/files/bindep.txt
if [[ -f ${bindep_file} ]]; then
packages+=$($BINDEP_CMD -b --file ${bindep_file} || true)
fi
done
echo "${packages}"
$xtrace
}
# Distro-agnostic package installer
# Uses globals ``NO_UPDATE_REPOS``, ``REPOS_UPDATED``, ``RETRY_UPDATE``
# install_package package [package ...]
function update_package_repo {
NO_UPDATE_REPOS=${NO_UPDATE_REPOS:-False}
REPOS_UPDATED=${REPOS_UPDATED:-False}
RETRY_UPDATE=${RETRY_UPDATE:-False}
if [[ "$NO_UPDATE_REPOS" = "True" ]]; then
return 0
fi
if is_ubuntu; then
apt_get_update
fi
}
function real_install_package {
if is_ubuntu; then
apt_get install "$@"
elif is_fedora; then
yum_install "$@"
elif is_suse; then
zypper_install "$@"
else
exit_distro_not_supported "installing packages"
fi
}
# Distro-agnostic package installer
# install_package package [package ...]
function install_package {
update_package_repo
if ! real_install_package "$@"; then
RETRY_UPDATE=True update_package_repo && real_install_package "$@"
fi
}
# Distro-agnostic function to tell if a package is installed
# is_package_installed package [package ...]
function is_package_installed {
if [[ -z "$@" ]]; then
return 1
fi
if [[ -z "$os_PACKAGE" ]]; then
GetOSVersion
fi
if [[ "$os_PACKAGE" = "deb" ]]; then
dpkg -s "$@" > /dev/null 2> /dev/null
elif [[ "$os_PACKAGE" = "rpm" ]]; then
rpm --quiet -q "$@"
else
exit_distro_not_supported "finding if a package is installed"
fi
}
# Distro-agnostic package uninstaller
# uninstall_package package [package ...]
function uninstall_package {
if is_ubuntu; then
apt_get purge "$@"
elif is_fedora; then
sudo dnf remove -y "$@" ||:
elif is_suse; then
sudo zypper remove -y "$@" ||:
else
exit_distro_not_supported "uninstalling packages"
fi
}
# Wrapper for ``dnf`` to set proxy environment variables
# Uses globals ``OFFLINE``, ``*_proxy``
# The name is kept for backwards compatability with external
# callers, despite none of our supported platforms using yum
# any more.
# yum_install package [package ...]
function yum_install {
local result parse_yum_result
[[ "$OFFLINE" = "True" ]] && return
time_start "yum_install"
sudo_with_proxies dnf install -y "$@"
time_stop "yum_install"
}
# zypper wrapper to set arguments correctly
# Uses globals ``OFFLINE``, ``*_proxy``
# zypper_install package [package ...]
function zypper_install {
[[ "$OFFLINE" = "True" ]] && return
local sudo="sudo"
[[ "$(id -u)" = "0" ]] && sudo="env"
$sudo http_proxy="${http_proxy:-}" https_proxy="${https_proxy:-}" \
no_proxy="${no_proxy:-}" \
zypper --non-interactive install --auto-agree-with-licenses --no-recommends "$@"
}
# Run bindep and install packages it outputs
#
# Usage:
# install_bindep <path-to-bindep.txt> [profile,profile]
#
# Note unlike the bindep command itself, profile(s) specified should
# be a single, comma-separated string, no spaces.
function install_bindep {
local file=$1
local profiles=${2:-""}
local pkgs
if [[ ! -f $file ]]; then
warn $LINENO "Can not find bindep file: $file"
return
fi
# converting here makes it much easier to work with passing
# arguments
profiles=${profiles/,/ /}
# Note bindep returns 1 when packages need to be installed, so we
# have to ignore it's return for "-e"
pkgs=$($DEST/bindep-venv/bin/bindep -b --file $file $profiles || true)
if [[ -n "${pkgs}" ]]; then
install_package ${pkgs}
fi
}
function write_user_unit_file {
local service=$1
local command="$2"
local group=$3
local user=$4
local extra=""
if [[ -n "$group" ]]; then
extra="Group=$group"
fi
local unitfile="$SYSTEMD_DIR/$service"
mkdir -p $SYSTEMD_DIR
iniset -sudo $unitfile "Unit" "Description" "Devstack $service"
iniset -sudo $unitfile "Service" "User" "$user"
iniset -sudo $unitfile "Service" "ExecStart" "$command"
iniset -sudo $unitfile "Service" "KillMode" "process"
iniset -sudo $unitfile "Service" "TimeoutStopSec" "300"
iniset -sudo $unitfile "Service" "ExecReload" "$KILL_PATH -HUP \$MAINPID"
if [[ -n "$group" ]]; then
iniset -sudo $unitfile "Service" "Group" "$group"
fi
iniset -sudo $unitfile "Install" "WantedBy" "multi-user.target"
# changes to existing units sometimes need a refresh
$SYSTEMCTL daemon-reload
}
function write_uwsgi_user_unit_file {
local service=$1
local command="$2"
local group=$3
local user=$4
local unitfile="$SYSTEMD_DIR/$service"
mkdir -p $SYSTEMD_DIR
iniset -sudo $unitfile "Unit" "Description" "Devstack $service"
iniset -sudo $unitfile "Service" "SyslogIdentifier" "$service"
iniset -sudo $unitfile "Service" "User" "$user"
iniset -sudo $unitfile "Service" "ExecStart" "$command"
iniset -sudo $unitfile "Service" "ExecReload" "$KILL_PATH -HUP \$MAINPID"
iniset -sudo $unitfile "Service" "Type" "notify"
iniset -sudo $unitfile "Service" "KillMode" "process"
iniset -sudo $unitfile "Service" "Restart" "always"
iniset -sudo $unitfile "Service" "NotifyAccess" "all"
iniset -sudo $unitfile "Service" "RestartForceExitStatus" "100"
if [[ -n "$group" ]]; then
iniset -sudo $unitfile "Service" "Group" "$group"
fi
iniset -sudo $unitfile "Install" "WantedBy" "multi-user.target"
# changes to existing units sometimes need a refresh
$SYSTEMCTL daemon-reload
}
function _common_systemd_pitfalls {
local cmd=$1
# do some sanity checks on $cmd to see things we don't expect to work
if [[ "$cmd" =~ "sudo" ]]; then
read -r -d '' msg << EOF || true # read returns 1 for EOF, but it is ok here
You are trying to use run_process with sudo, this is not going to work under systemd.
If you need to run a service as a user other than \$STACK_USER call it with:
run_process \$name \$cmd \$group \$user
EOF
die $LINENO "$msg"
fi
if [[ ! "$cmd" =~ ^/ ]]; then
read -r -d '' msg << EOF || true # read returns 1 for EOF, but it is ok here
The cmd="$cmd" does not start with an absolute path. It will fail to
start under systemd.
Please update your run_process stanza to have an absolute path.
EOF
die $LINENO "$msg"
fi
}
# Helper function to build a basic unit file and run it under systemd.
function _run_under_systemd {
local service=$1
local command="$2"
local cmd=$command
# sanity check the command
_common_systemd_pitfalls "$cmd"
local systemd_service="devstack@$service.service"
local group=$3
local user=${4:-$STACK_USER}
if [[ "$command" =~ "uwsgi" ]] ; then
write_uwsgi_user_unit_file $systemd_service "$cmd" "$group" "$user"
else
write_user_unit_file $systemd_service "$cmd" "$group" "$user"
fi
$SYSTEMCTL enable $systemd_service
$SYSTEMCTL start $systemd_service
}
# Find out if a process exists by partial name.
# is_running name
function is_running {
local name=$1
ps auxw | grep -v grep | grep ${name} > /dev/null
local exitcode=$?
# some times I really hate bash reverse binary logic
return $exitcode
}
# Run a single service under screen or directly
# If the command includes shell metachatacters (;<>*) it must be run using a shell
# If an optional group is provided sg will be used to run the
# command as that group.
# run_process service "command-line" [group] [user]
function run_process {
local service=$1
local command="$2"
local group=$3
local user=$4
local name=$service
time_start "run_process"
if is_service_enabled $service; then
_run_under_systemd "$name" "$command" "$group" "$user"
fi
time_stop "run_process"
}
# Stop a service process
# If a PID is available use it, kill the whole process group via TERM
# If screen is being used kill the screen window; this will catch processes
# that did not leave a PID behind
# Uses globals ``SERVICE_DIR``
# stop_process service
function stop_process {
local service=$1
SERVICE_DIR=${SERVICE_DIR:-${DEST}/status}
if is_service_enabled $service; then
# Only do this for units which appear enabled, this also
# catches units that don't really exist for cases like
# keystone without a failure.
if $SYSTEMCTL is-enabled devstack@$service.service; then
$SYSTEMCTL stop devstack@$service.service
$SYSTEMCTL disable devstack@$service.service
fi
fi
}
# use systemctl to check service status
function service_check {
local service
for service in ${ENABLED_SERVICES//,/ }; do
# because some things got renamed like key => keystone
if $SYSTEMCTL is-enabled devstack@$service.service; then
# no-pager is needed because otherwise status dumps to a
# pager when in interactive mode, which will stop a manual
# devstack run.
$SYSTEMCTL status devstack@$service.service --no-pager
fi
done
}
function tail_log {
deprecated "With the removal of screen support, tail_log is deprecated and will be removed after Queens"
}
# Plugin Functions
# =================
DEVSTACK_PLUGINS=${DEVSTACK_PLUGINS:-""}
# enable_plugin <name> <url> [branch]
#
# ``name`` is an arbitrary name - (aka: glusterfs, nova-docker, zaqar)
# ``url`` is a git url
# ``branch`` is a gitref. If it's not set, defaults to master
function enable_plugin {
local name=$1
local url=$2
local branch=${3:-master}
if is_plugin_enabled $name; then
die $LINENO "Plugin attempted to be enabled twice: ${name} ${url} ${branch}"
fi
DEVSTACK_PLUGINS+=",$name"
GITREPO[$name]=$url
GITDIR[$name]=$DEST/$name
GITBRANCH[$name]=$branch
}
# is_plugin_enabled <name>
#
# Check if the plugin was enabled, e.g. using enable_plugin
#
# ``name`` The name with which the plugin was enabled
function is_plugin_enabled {
local name=$1
if [[ ",${DEVSTACK_PLUGINS}," =~ ",${name}," ]]; then
return 0
fi
return 1
}
# fetch_plugins
#
# clones all plugins
function fetch_plugins {
local plugins="${DEVSTACK_PLUGINS}"
local plugin
# short circuit if nothing to do
if [[ -z $plugins ]]; then
return
fi
echo "Fetching DevStack plugins"
for plugin in ${plugins//,/ }; do
git_clone_by_name $plugin
done
}
# load_plugin_settings
#
# Load settings from plugins in the order that they were registered
function load_plugin_settings {
local plugins="${DEVSTACK_PLUGINS}"
local plugin
# short circuit if nothing to do
if [[ -z $plugins ]]; then
return
fi
echo "Loading plugin settings"
for plugin in ${plugins//,/ }; do
local dir=${GITDIR[$plugin]}
# source any known settings
if [[ -f $dir/devstack/settings ]]; then
source $dir/devstack/settings
fi
done
}
# plugin_override_defaults
#
# Run an extremely early setting phase for plugins that allows default
# overriding of services.
function plugin_override_defaults {
local plugins="${DEVSTACK_PLUGINS}"
local plugin
# short circuit if nothing to do
if [[ -z $plugins ]]; then
return
fi
echo "Overriding Configuration Defaults"
for plugin in ${plugins//,/ }; do
local dir=${GITDIR[$plugin]}
# source any overrides
if [[ -f $dir/devstack/override-defaults ]]; then
# be really verbose that an override is happening, as it
# may not be obvious if things fail later.
echo "$plugin has overridden the following defaults"
cat $dir/devstack/override-defaults
source $dir/devstack/override-defaults
fi
done
}
# run_plugins
#
# Run the devstack/plugin.sh in all the plugin directories. These are
# run in registration order.
function run_plugins {
local mode=$1
local phase=$2
local plugins="${DEVSTACK_PLUGINS}"
local plugin
for plugin in ${plugins//,/ }; do
local dir=${GITDIR[$plugin]}
if [[ -f $dir/devstack/plugin.sh ]]; then
source $dir/devstack/plugin.sh $mode $phase
fi
done
}
function run_phase {
local mode=$1
local phase=$2
if [[ -d $TOP_DIR/extras.d ]]; then
local extra_plugin_file_name
for extra_plugin_file_name in $TOP_DIR/extras.d/*.sh; do
# NOTE(sdague): only process extras.d for the 3 explicitly
# white listed elements in tree. We want these to move out
# over time as well, but they are in tree, so we need to
# manage that.
local exceptions="80-tempest.sh"
local extra
extra=$(basename $extra_plugin_file_name)
if [[ ! ( $exceptions =~ "$extra" ) ]]; then
warn "use of extras.d is no longer supported"
warn "processing of project $extra is skipped"
else
[[ -r $extra_plugin_file_name ]] && source $extra_plugin_file_name $mode $phase
fi
done
fi
# the source phase corresponds to settings loading in plugins
if [[ "$mode" == "source" ]]; then
load_plugin_settings
verify_disabled_services
elif [[ "$mode" == "override_defaults" ]]; then
plugin_override_defaults
else
run_plugins $mode $phase
fi
}
# define_plugin <name>
#
# This function is a no-op. It allows a plugin to define its name So
# that other plugins may reference it by name. It should generally be
# the last component of the canonical git repo name. E.g.,
# openstack/devstack-foo should use "devstack-foo" as the name here.
#
# This function is currently a noop, but the value may still be used
# by external tools (as in plugin_requires) and may be used by
# devstack in the future.
#
# ``name`` is an arbitrary name - (aka: glusterfs, nova-docker, zaqar)
function define_plugin {
:
}
# plugin_requires <name> <other>
#
# This function is a no-op. It is currently used by external tools
# (such as the devstack module for Ansible) to automatically generate
# local.conf files. It is not currently used by devstack itself to
# resolve dependencies.
#
# ``name`` is an arbitrary name - (aka: glusterfs, nova-docker, zaqar)
# ``other`` is the name of another plugin
function plugin_requires {
:
}
# Service Functions
# =================
# remove extra commas from the input string (i.e. ``ENABLED_SERVICES``)
# _cleanup_service_list service-list
function _cleanup_service_list {
local xtrace
xtrace=$(set +o | grep xtrace)
set +o xtrace
echo "$1" | sed -e '
s/,,/,/g;
s/^,//;
s/,$//
'
$xtrace
}
# disable_all_services() removes all current services
# from ``ENABLED_SERVICES`` to reset the configuration
# before a minimal installation
# Uses global ``ENABLED_SERVICES``
# disable_all_services
function disable_all_services {
ENABLED_SERVICES=""
}
# Remove all services starting with '-'. For example, to install all default
# services except rabbit (rabbit) set in ``localrc``:
# ENABLED_SERVICES+=",-rabbit"
# Uses global ``ENABLED_SERVICES``
# disable_negated_services
function disable_negated_services {
local xtrace
xtrace=$(set +o | grep xtrace)
set +o xtrace
local to_remove=""
local remaining=""
local service
# build up list of services that should be removed; i.e. they
# begin with "-"
for service in ${ENABLED_SERVICES//,/ }; do
if [[ ${service} == -* ]]; then
to_remove+=",${service#-}"
else
remaining+=",${service}"
fi
done
# go through the service list. if this service appears in the "to
# be removed" list, drop it
ENABLED_SERVICES=$(remove_disabled_services "$remaining" "$to_remove")
$xtrace
}
# disable_service() prepares the services passed as argument to be
# removed from the ``ENABLED_SERVICES`` list, if they are present.
#
# For example:
# disable_service rabbit
#
# Uses global ``DISABLED_SERVICES``
# disable_service service [service ...]
function disable_service {
local xtrace
xtrace=$(set +o | grep xtrace)
set +o xtrace
local disabled_svcs="${DISABLED_SERVICES}"
local enabled_svcs=",${ENABLED_SERVICES},"
local service
for service in $@; do
disabled_svcs+=",$service"
if is_service_enabled $service; then
enabled_svcs=${enabled_svcs//,$service,/,}
fi
done
DISABLED_SERVICES=$(_cleanup_service_list "$disabled_svcs")
ENABLED_SERVICES=$(_cleanup_service_list "$enabled_svcs")
$xtrace
}
# enable_service() adds the services passed as argument to the
# ``ENABLED_SERVICES`` list, if they are not already present.
#
# For example:
# enable_service q-svc
#
# This function does not know about the special cases
# for nova, glance, and neutron built into is_service_enabled().
# Uses global ``ENABLED_SERVICES``
# enable_service service [service ...]
function enable_service {
local xtrace
xtrace=$(set +o | grep xtrace)
set +o xtrace
local tmpsvcs="${ENABLED_SERVICES}"
local service
for service in $@; do
if [[ ,${DISABLED_SERVICES}, =~ ,${service}, ]]; then
warn $LINENO "Attempt to enable_service ${service} when it has been disabled"
continue
fi
if ! is_service_enabled $service; then
tmpsvcs+=",$service"
fi
done
ENABLED_SERVICES=$(_cleanup_service_list "$tmpsvcs")
disable_negated_services
$xtrace
}
# is_service_enabled() checks if the service(s) specified as arguments are
# enabled by the user in ``ENABLED_SERVICES``.
#
# Multiple services specified as arguments are ``OR``'ed together; the test
# is a short-circuit boolean, i.e it returns on the first match.
#
# There are special cases for some 'catch-all' services::
# **nova** returns true if any service enabled start with **n-**
# **cinder** returns true if any service enabled start with **c-**
# **glance** returns true if any service enabled start with **g-**
# **neutron** returns true if any service enabled start with **q-**
# **swift** returns true if any service enabled start with **s-**
# **trove** returns true if any service enabled start with **tr-**
# For backward compatibility if we have **swift** in ENABLED_SERVICES all the
# **s-** services will be enabled. This will be deprecated in the future.
#
# Uses global ``ENABLED_SERVICES``
# is_service_enabled service [service ...]
function is_service_enabled {
local xtrace
xtrace=$(set +o | grep xtrace)
set +o xtrace
local enabled=1
local services=$@
local service
for service in ${services}; do
[[ ,${ENABLED_SERVICES}, =~ ,${service}, ]] && enabled=0
# Look for top-level 'enabled' function for this service
if type is_${service}_enabled >/dev/null 2>&1; then
# A function exists for this service, use it
is_${service}_enabled && enabled=0
fi
# TODO(dtroyer): Remove these legacy special-cases after the is_XXX_enabled()
# are implemented
[[ ${service} == n-cpu-* && ,${ENABLED_SERVICES} =~ ,"n-cpu" ]] && enabled=0
[[ ${service} == "nova" && ,${ENABLED_SERVICES} =~ ,"n-" ]] && enabled=0
[[ ${service} == "glance" && ,${ENABLED_SERVICES} =~ ,"g-" ]] && enabled=0
[[ ${service} == "neutron" && ,${ENABLED_SERVICES} =~ ,"q-" ]] && enabled=0
[[ ${service} == "trove" && ,${ENABLED_SERVICES} =~ ,"tr-" ]] && enabled=0
[[ ${service} == "swift" && ,${ENABLED_SERVICES} =~ ,"s-" ]] && enabled=0
[[ ${service} == s-* && ,${ENABLED_SERVICES} =~ ,"swift" ]] && enabled=0
done
$xtrace
return $enabled
}
# remove specified list from the input string
# remove_disabled_services service-list remove-list
function remove_disabled_services {
local xtrace
xtrace=$(set +o | grep xtrace)
set +o xtrace
local service_list=$1
local remove_list=$2
local service
local enabled=""
for service in ${service_list//,/ }; do
local remove
local add=1
for remove in ${remove_list//,/ }; do
if [[ ${remove} == ${service} ]]; then
add=0
break
fi
done
if [[ $add == 1 ]]; then
enabled="${enabled},$service"
fi
done
$xtrace
_cleanup_service_list "$enabled"
}
# Toggle enable/disable_service for services that must run exclusive of each other
# $1 The name of a variable containing a space-separated list of services
# $2 The name of a variable in which to store the enabled service's name
# $3 The name of the service to enable
function use_exclusive_service {
local options=${!1}
local selection=$3
local out=$2
[ -z $selection ] || [[ ! "$options" =~ "$selection" ]] && return 1
local opt
for opt in $options;do
[[ "$opt" = "$selection" ]] && enable_service $opt || disable_service $opt
done
eval "$out=$selection"
return 0
}
# Make sure that nothing has manipulated ENABLED_SERVICES in a way
# that conflicts with prior calls to disable_service.
# Uses global ``ENABLED_SERVICES``
function verify_disabled_services {
local service
for service in ${ENABLED_SERVICES//,/ }; do
if [[ ,${DISABLED_SERVICES}, =~ ,${service}, ]]; then
die $LINENO "ENABLED_SERVICES directly modified to overcome 'disable_service ${service}'"
fi
done
}
# System Functions
# ================
# Only run the command if the target file (the last arg) is not on an
# NFS filesystem.
function _safe_permission_operation {
local xtrace
xtrace=$(set +o | grep xtrace)
set +o xtrace
local args=( $@ )
local last
local sudo_cmd
local dir_to_check
let last="${#args[*]} - 1"
local dir_to_check=${args[$last]}
if [ ! -d "$dir_to_check" ]; then
dir_to_check=`dirname "$dir_to_check"`
fi
if is_nfs_directory "$dir_to_check" ; then
$xtrace
return 0
fi
sudo_cmd="sudo"
$xtrace
$sudo_cmd $@
}
# Exit 0 if address is in network or 1 if address is not in network
# ip-range is in CIDR notation: 1.2.3.4/20
# address_in_net ip-address ip-range
function address_in_net {
local ip=$1
local range=$2
local masklen=${range#*/}
local network
network=$(maskip ${range%/*} $(cidr2netmask $masklen))
local subnet
subnet=$(maskip $ip $(cidr2netmask $masklen))
[[ $network == $subnet ]]
}
# Add a user to a group.
# add_user_to_group user group
function add_user_to_group {
local user=$1
local group=$2
sudo usermod -a -G "$group" "$user"
}
# Convert CIDR notation to a IPv4 netmask
# cidr2netmask cidr-bits
function cidr2netmask {
local maskpat="255 255 255 255"
local maskdgt="254 252 248 240 224 192 128"
set -- ${maskpat:0:$(( ($1 / 8) * 4 ))}${maskdgt:$(( (7 - ($1 % 8)) * 4 )):3}
echo ${1-0}.${2-0}.${3-0}.${4-0}
}
# Check if this is a valid ipv4 address string
function is_ipv4_address {
local address=$1
local regex='([0-9]{1,3}\.){3}[0-9]{1,3}'
# TODO(clarkb) make this more robust
if [[ "$address" =~ $regex ]] ; then
return 0
else
return 1
fi
}
# Remove "[]" around urlquoted IPv6 addresses
function ipv6_unquote {
echo $1 | tr -d []
}
# Gracefully cp only if source file/dir exists
# cp_it source destination
function cp_it {
if [ -e $1 ] || [ -d $1 ]; then
cp -pRL $1 $2
fi
}
# HTTP and HTTPS proxy servers are supported via the usual environment variables [1]
# ``http_proxy``, ``https_proxy`` and ``no_proxy``. They can be set in
# ``localrc`` or on the command line if necessary::
#
# [1] http://www.w3.org/Daemon/User/Proxies/ProxyClients.html
#
# http_proxy=http://proxy.example.com:3128/ no_proxy=repo.example.net ./stack.sh
function export_proxy_variables {
if isset http_proxy ; then
export http_proxy=$http_proxy
fi
if isset https_proxy ; then
export https_proxy=$https_proxy
fi
if isset no_proxy ; then
export no_proxy=$no_proxy
fi
}
# Returns true if the directory is on a filesystem mounted via NFS.
function is_nfs_directory {
local mount_type
mount_type=`stat -f -L -c %T $1`
test "$mount_type" == "nfs"
}
# Return the network portion of the given IP address using netmask
# netmask is in the traditional dotted-quad format
# maskip ip-address netmask
function maskip {
local ip=$1
local mask=$2
local l="${ip%.*}"; local r="${ip#*.}"; local n="${mask%.*}"; local m="${mask#*.}"
local subnet
subnet=$((${ip%%.*}&${mask%%.*})).$((${r%%.*}&${m%%.*})).$((${l##*.}&${n##*.})).$((${ip##*.}&${mask##*.}))
echo $subnet
}
function is_provider_network {
if [ "$Q_USE_PROVIDER_NETWORKING" == "True" ]; then
return 0
fi
return 1
}
# Return just the <major>.<minor> for the given python interpreter
function _get_python_version {
local interp=$1
local version
# disable erroring out here, otherwise if python 3 doesn't exist we fail hard.
if [[ -x $(which $interp 2> /dev/null) ]]; then
version=$($interp -c 'import sys; print("%s.%s" % sys.version_info[0:2])')
fi
echo ${version}
}
# Return the current python as "python<major>.<minor>"
function python_version {
local python_version
python_version=$(_get_python_version python2)
echo "python${python_version}"
}
function python3_version {
local python3_version
python3_version=$(_get_python_version python3)
echo "python${python3_version}"
}
# Service wrapper to restart services
# restart_service service-name
function restart_service {
if [ -x /bin/systemctl ]; then
sudo /bin/systemctl restart $1
else
sudo service $1 restart
fi
}
# Only change permissions of a file or directory if it is not on an
# NFS filesystem.
function safe_chmod {
_safe_permission_operation chmod $@
}
# Only change ownership of a file or directory if it is not on an NFS
# filesystem.
function safe_chown {
_safe_permission_operation chown $@
}
# Service wrapper to start services
# start_service service-name
function start_service {
if [ -x /bin/systemctl ]; then
sudo /bin/systemctl start $1
else
sudo service $1 start
fi
}
# Service wrapper to stop services
# stop_service service-name
function stop_service {
if [ -x /bin/systemctl ]; then
sudo /bin/systemctl stop $1
else
sudo service $1 stop
fi
}
# Service wrapper to reload services
# If the service was not in running state it will start it
# reload_service service-name
function reload_service {
if [ -x /bin/systemctl ]; then
sudo /bin/systemctl reload-or-restart $1
else
sudo service $1 reload
fi
}
# Test with a finite retry loop.
#
function test_with_retry {
local testcmd=$1
local failmsg=$2
local until=${3:-10}
local sleep=${4:-0.5}
time_start "test_with_retry"
if ! timeout $until sh -c "while ! $testcmd; do sleep $sleep; done"; then
die $LINENO "$failmsg"
fi
time_stop "test_with_retry"
}
# Like sudo but forwarding http_proxy https_proxy no_proxy environment vars.
# If it is run as superuser then sudo is replaced by env.
#
function sudo_with_proxies {
local sudo
[[ "$(id -u)" = "0" ]] && sudo="env" || sudo="sudo"
$sudo http_proxy="${http_proxy:-}" https_proxy="${https_proxy:-}"\
no_proxy="${no_proxy:-}" "$@"
}
# Timing infrastructure - figure out where large blocks of time are
# used in DevStack
#
# The timing infrastructure for DevStack is about collecting buckets
# of time that are spend in some subtask. For instance, that might be
# 'apt', 'pip', 'osc', even database migrations. We do this by a pair
# of functions: time_start / time_stop.
#
# These take a single parameter: $name - which specifies the name of
# the bucket to be accounted against. time_totals function spits out
# the results.
#
# Resolution is only in whole seconds, so should be used for long
# running activities.
declare -A -g _TIME_TOTAL
declare -A -g _TIME_START
declare -r -g _TIME_BEGIN=$(date +%s)
# time_start $name
#
# starts the clock for a timer by name. Errors if that clock is
# already started.
function time_start {
local name=$1
local start_time=${_TIME_START[$name]}
if [[ -n "$start_time" ]]; then
die $LINENO "Trying to start the clock on $name, but it's already been started"
fi
_TIME_START[$name]=$(date +%s%3N)
}
# time_stop $name
#
# stops the clock for a timer by name, and accumulate that time in the
# global counter for that name. Errors if that clock had not
# previously been started.
function time_stop {
local name
local end_time
local elapsed_time
local total
local start_time
name=$1
start_time=${_TIME_START[$name]}
if [[ -z "$start_time" ]]; then
die $LINENO "Trying to stop the clock on $name, but it was never started"
fi
end_time=$(date +%s%3N)
elapsed_time=$(($end_time - $start_time))
total=${_TIME_TOTAL[$name]:-0}
# reset the clock so we can start it in the future
_TIME_START[$name]=""
_TIME_TOTAL[$name]=$(($total + $elapsed_time))
}
function oscwrap {
local xtrace
xtrace=$(set +o | grep xtrace)
set +o xtrace
local out
local rc
local start
local end
# Cannot use timer_start and timer_stop as we run in subshells
# and those rely on modifying vars in the same process (which cannot
# happen from a subshell.
start=$(date +%s%3N)
out=$(command openstack "$@")
rc=$?
end=$(date +%s%3N)
echo $((end - start)) >> $OSCWRAP_TIMER_FILE
echo "$out"
$xtrace
return $rc
}
function install_oscwrap {
# File to accumulate our timing data
OSCWRAP_TIMER_FILE=$(mktemp)
# Bash by default doesn't expand aliases, allow it for the aliases
# we want to whitelist.
shopt -s expand_aliases
# Remove all aliases that might be expanded to preserve old unexpanded
# behavior
unalias -a
# Add only the alias we want for openstack
alias openstack=oscwrap
}
function cleanup_oscwrap {
local total=0
total=$(cat $OSCWRAP_TIMER_FILE | $PYTHON -c "import sys; print(sum(int(l) for l in sys.stdin))")
_TIME_TOTAL["osc"]=$total
rm $OSCWRAP_TIMER_FILE
}
# time_totals
# Print out total time summary
function time_totals {
local elapsed_time
local end_time
local len=20
local xtrace
local unaccounted_time
end_time=$(date +%s)
elapsed_time=$(($end_time - $_TIME_BEGIN))
unaccounted_time=$elapsed_time
# pad 1st column this far
for t in ${!_TIME_TOTAL[*]}; do
if [[ ${#t} -gt $len ]]; then
len=${#t}
fi
done
cleanup_oscwrap
xtrace=$(set +o | grep xtrace)
set +o xtrace
echo
echo "========================="
echo "DevStack Component Timing"
echo " (times are in seconds) "
echo "========================="
for t in ${!_TIME_TOTAL[*]}; do
local v=${_TIME_TOTAL[$t]}
# because we're recording in milliseconds
v=$(($v / 1000))
printf "%-${len}s %3d\n" "$t" "$v"
unaccounted_time=$(($unaccounted_time - $v))
done
echo "-------------------------"
printf "%-${len}s %3d\n" "Unaccounted time" "$unaccounted_time"
echo "========================="
printf "%-${len}s %3d\n" "Total runtime" "$elapsed_time"
$xtrace
}
# Restore xtrace
$_XTRACE_FUNCTIONS_COMMON
# Local variables:
# mode: shell-script
# End: