devstack/stack.sh
Johannes Erdfelt 965716e168 Explicitly set connection_type to libvirt
A future merge to nova (http://review.openstack.org/2918) will require this to be set

Change-Id: I92c28336a6022c325a84d6594764779d8fe5e655
2012-01-23 15:03:43 -08:00

1557 lines
61 KiB
Bash
Executable File
Raw Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

#!/usr/bin/env bash
# **stack.sh** is an opinionated openstack developer installation.
# This script installs and configures *nova*, *glance*, *horizon* and *keystone*
# This script allows you to specify configuration options of what git
# repositories to use, enabled services, network configuration and various
# passwords. If you are crafty you can run the script on multiple nodes using
# shared settings for common resources (mysql, rabbitmq) and build a multi-node
# developer install.
# To keep this script simple we assume you are running on an **Ubuntu 11.10
# Oneiric** machine. It should work in a VM or physical server. Additionally
# we put the list of *apt* and *pip* dependencies and other configuration files
# in this repo. So start by grabbing this script and the dependencies.
# Learn more and get the most recent version at http://devstack.org
# Sanity Check
# ============
# Warn users who aren't on oneiric, but allow them to override check and attempt
# installation with ``FORCE=yes ./stack``
DISTRO=$(lsb_release -c -s)
if [[ ! ${DISTRO} =~ (oneiric) ]]; then
echo "WARNING: this script has only been tested on oneiric"
if [[ "$FORCE" != "yes" ]]; then
echo "If you wish to run this script anyway run with FORCE=yes"
exit 1
fi
fi
# Keep track of the current devstack directory.
TOP_DIR=$(cd $(dirname "$0") && pwd)
# stack.sh keeps the list of **apt** and **pip** dependencies in external
# files, along with config templates and other useful files. You can find these
# in the ``files`` directory (next to this script). We will reference this
# directory using the ``FILES`` variable in this script.
FILES=$TOP_DIR/files
if [ ! -d $FILES ]; then
echo "ERROR: missing devstack/files - did you grab more than just stack.sh?"
exit 1
fi
# Settings
# ========
# This script is customizable through setting environment variables. If you
# want to override a setting you can either::
#
# export MYSQL_PASSWORD=anothersecret
# ./stack.sh
#
# You can also pass options on a single line ``MYSQL_PASSWORD=simple ./stack.sh``
#
# Additionally, you can put any local variables into a ``localrc`` file, like::
#
# MYSQL_PASSWORD=anothersecret
# MYSQL_USER=hellaroot
#
# We try to have sensible defaults, so you should be able to run ``./stack.sh``
# in most cases.
#
# We support HTTP and HTTPS proxy servers via the usual environment variables
# http_proxy and https_proxy. They can be set in localrc if necessary or
# on the command line::
#
# http_proxy=http://proxy.example.com:3128/ ./stack.sh
#
# We source our settings from ``stackrc``. This file is distributed with devstack
# and contains locations for what repositories to use. If you want to use other
# repositories and branches, you can add your own settings with another file called
# ``localrc``
#
# If ``localrc`` exists, then ``stackrc`` will load those settings. This is
# useful for changing a branch or repository to test other versions. Also you
# can store your other settings like **MYSQL_PASSWORD** or **ADMIN_PASSWORD** instead
# of letting devstack generate random ones for you.
source ./stackrc
# Destination path for installation ``DEST``
DEST=${DEST:-/opt/stack}
# apt-get wrapper to just get arguments set correctly
function apt_get() {
[[ "$OFFLINE" = "True" ]] && return
local sudo="sudo"
[ "$(id -u)" = "0" ] && sudo="env"
$sudo DEBIAN_FRONTEND=noninteractive \
http_proxy=$http_proxy https_proxy=$https_proxy \
apt-get --option "Dpkg::Options::=--force-confold" --assume-yes "$@"
}
# Check to see if we are already running a stack.sh
if screen -ls | egrep -q "[0-9].stack"; then
echo "You are already running a stack.sh session."
echo "To rejoin this session type 'screen -x stack'."
echo "To destroy this session, kill the running screen."
exit 1
fi
# OpenStack is designed to be run as a regular user (Horizon will fail to run
# as root, since apache refused to startup serve content from root user). If
# stack.sh is run as root, it automatically creates a stack user with
# sudo privileges and runs as that user.
if [[ $EUID -eq 0 ]]; then
ROOTSLEEP=${ROOTSLEEP:-10}
echo "You are running this script as root."
echo "In $ROOTSLEEP seconds, we will create a user 'stack' and run as that user"
sleep $ROOTSLEEP
# since this script runs as a normal user, we need to give that user
# ability to run sudo
dpkg -l sudo || apt_get update && apt_get install sudo
if ! getent passwd stack >/dev/null; then
echo "Creating a user called stack"
useradd -U -G sudo -s /bin/bash -d $DEST -m stack
fi
echo "Giving stack user passwordless sudo priviledges"
# some uec images sudoers does not have a '#includedir'. add one.
grep -q "^#includedir.*/etc/sudoers.d" /etc/sudoers ||
echo "#includedir /etc/sudoers.d" >> /etc/sudoers
( umask 226 && echo "stack ALL=(ALL) NOPASSWD:ALL" \
> /etc/sudoers.d/50_stack_sh )
echo "Copying files to stack user"
STACK_DIR="$DEST/${PWD##*/}"
cp -r -f -T "$PWD" "$STACK_DIR"
chown -R stack "$STACK_DIR"
if [[ "$SHELL_AFTER_RUN" != "no" ]]; then
exec su -c "set -e; cd $STACK_DIR; bash stack.sh; bash" stack
else
exec su -c "set -e; cd $STACK_DIR; bash stack.sh" stack
fi
exit 1
else
# Our user needs passwordless priviledges for certain commands which nova
# uses internally.
# Natty uec images sudoers does not have a '#includedir'. add one.
sudo grep -q "^#includedir.*/etc/sudoers.d" /etc/sudoers ||
echo "#includedir /etc/sudoers.d" | sudo tee -a /etc/sudoers
TEMPFILE=`mktemp`
cat $FILES/sudo/nova > $TEMPFILE
sed -e "s,%USER%,$USER,g" -i $TEMPFILE
chmod 0440 $TEMPFILE
sudo chown root:root $TEMPFILE
sudo mv $TEMPFILE /etc/sudoers.d/stack_sh_nova
fi
# Normalize config values to True or False
# VAR=`trueorfalse default-value test-value`
function trueorfalse() {
local default=$1
local testval=$2
[[ -z "$testval" ]] && { echo "$default"; return; }
[[ "0 no false False FALSE" =~ "$testval" ]] && { echo "False"; return; }
[[ "1 yes true True TRUE" =~ "$testval" ]] && { echo "True"; return; }
echo "$default"
}
# Set True to configure stack.sh to run cleanly without Internet access.
# stack.sh must have been previously run with Internet access to install
# prerequisites and initialize $DEST.
OFFLINE=`trueorfalse False $OFFLINE`
# Set the destination directories for openstack projects
NOVA_DIR=$DEST/nova
HORIZON_DIR=$DEST/horizon
GLANCE_DIR=$DEST/glance
KEYSTONE_DIR=$DEST/keystone
NOVACLIENT_DIR=$DEST/python-novaclient
KEYSTONECLIENT_DIR=$DEST/python-keystoneclient
OPENSTACKX_DIR=$DEST/openstackx
NOVNC_DIR=$DEST/noVNC
SWIFT_DIR=$DEST/swift
SWIFT_KEYSTONE_DIR=$DEST/swift-keystone2
QUANTUM_DIR=$DEST/quantum
# Default Quantum Plugin
Q_PLUGIN=${Q_PLUGIN:-openvswitch}
# Default Quantum Port
Q_PORT=${Q_PORT:-9696}
# Default Quantum Host
Q_HOST=${Q_HOST:-localhost}
# Specify which services to launch. These generally correspond to screen tabs
ENABLED_SERVICES=${ENABLED_SERVICES:-g-api,g-reg,key,n-api,n-cpu,n-net,n-sch,n-novnc,n-xvnc,n-cauth,horizon,mysql,rabbit,openstackx}
# Name of the lvm volume group to use/create for iscsi volumes
VOLUME_GROUP=${VOLUME_GROUP:-nova-volumes}
VOLUME_NAME_PREFIX=${VOLUME_NAME_PREFIX:-volume-}
INSTANCE_NAME_PREFIX=${INSTANCE_NAME_PREFIX:-instance-}
# Nova hypervisor configuration. We default to libvirt whth **kvm** but will
# drop back to **qemu** if we are unable to load the kvm module. Stack.sh can
# also install an **LXC** based system.
VIRT_DRIVER=${VIRT_DRIVER:-libvirt}
LIBVIRT_TYPE=${LIBVIRT_TYPE:-kvm}
# nova supports pluggable schedulers. ``SimpleScheduler`` should work in most
# cases unless you are working on multi-zone mode.
SCHEDULER=${SCHEDULER:-nova.scheduler.simple.SimpleScheduler}
# Use the eth0 IP unless an explicit is set by ``HOST_IP`` environment variable
if [ ! -n "$HOST_IP" ]; then
HOST_IP=`LC_ALL=C /sbin/ifconfig eth0 | grep -m 1 'inet addr:'| cut -d: -f2 | awk '{print $1}'`
if [ "$HOST_IP" = "" ]; then
echo "Could not determine host ip address."
echo "If this is not your first run of stack.sh, it is "
echo "possible that nova moved your eth0 ip address to the FLAT_NETWORK_BRIDGE."
echo "Please specify your HOST_IP in your localrc."
exit 1
fi
fi
# Allow the use of an alternate hostname (such as localhost/127.0.0.1) for service endpoints.
SERVICE_HOST=${SERVICE_HOST:-$HOST_IP}
# Configure services to syslog instead of writing to individual log files
SYSLOG=`trueorfalse False $SYSLOG`
SYSLOG_HOST=${SYSLOG_HOST:-$HOST_IP}
SYSLOG_PORT=${SYSLOG_PORT:-516}
# Service startup timeout
SERVICE_TIMEOUT=${SERVICE_TIMEOUT:-60}
# Generic helper to configure passwords
function read_password {
set +o xtrace
var=$1; msg=$2
pw=${!var}
localrc=$TOP_DIR/localrc
# If the password is not defined yet, proceed to prompt user for a password.
if [ ! $pw ]; then
# If there is no localrc file, create one
if [ ! -e $localrc ]; then
touch $localrc
fi
# Presumably if we got this far it can only be that our localrc is missing
# the required password. Prompt user for a password and write to localrc.
echo ''
echo '################################################################################'
echo $msg
echo '################################################################################'
echo "This value will be written to your localrc file so you don't have to enter it "
echo "again. Use only alphanumeric characters."
echo "If you leave this blank, a random default value will be used."
pw=" "
while true; do
echo "Enter a password now:"
read -e $var
pw=${!var}
[[ "$pw" = "`echo $pw | tr -cd [:alnum:]`" ]] && break
echo "Invalid chars in password. Try again:"
done
if [ ! $pw ]; then
pw=`openssl rand -hex 10`
fi
eval "$var=$pw"
echo "$var=$pw" >> $localrc
fi
set -o xtrace
}
# Nova Network Configuration
# --------------------------
# FIXME: more documentation about why these are important flags. Also
# we should make sure we use the same variable names as the flag names.
PUBLIC_INTERFACE=${PUBLIC_INTERFACE:-eth0}
FIXED_RANGE=${FIXED_RANGE:-10.0.0.0/24}
FIXED_NETWORK_SIZE=${FIXED_NETWORK_SIZE:-256}
FLOATING_RANGE=${FLOATING_RANGE:-172.24.4.224/28}
NET_MAN=${NET_MAN:-FlatDHCPManager}
EC2_DMZ_HOST=${EC2_DMZ_HOST:-$SERVICE_HOST}
FLAT_NETWORK_BRIDGE=${FLAT_NETWORK_BRIDGE:-br100}
VLAN_INTERFACE=${VLAN_INTERFACE:-$PUBLIC_INTERFACE}
# Test floating pool and range are used for testing. They are defined
# here until the admin APIs can replace nova-manage
TEST_FLOATING_POOL=${TEST_FLOATING_POOL:-test}
TEST_FLOATING_RANGE=${TEST_FLOATING_RANGE:-192.168.253.0/29}
# Multi-host is a mode where each compute node runs its own network node. This
# allows network operations and routing for a VM to occur on the server that is
# running the VM - removing a SPOF and bandwidth bottleneck.
MULTI_HOST=${MULTI_HOST:-False}
# If you are using FlatDHCP on multiple hosts, set the ``FLAT_INTERFACE``
# variable but make sure that the interface doesn't already have an
# ip or you risk breaking things.
#
# **DHCP Warning**: If your flat interface device uses DHCP, there will be a
# hiccup while the network is moved from the flat interface to the flat network
# bridge. This will happen when you launch your first instance. Upon launch
# you will lose all connectivity to the node, and the vm launch will probably
# fail.
#
# If you are running on a single node and don't need to access the VMs from
# devices other than that node, you can set the flat interface to the same
# value as ``FLAT_NETWORK_BRIDGE``. This will stop the network hiccup from
# occurring.
FLAT_INTERFACE=${FLAT_INTERFACE:-eth0}
## FIXME(ja): should/can we check that FLAT_INTERFACE is sane?
# Using Quantum networking:
#
# Make sure that quantum is enabled in ENABLED_SERVICES. If it is the network
# manager will be set to the QuantumManager. If you want to run Quantum on
# this host, make sure that q-svc is also in ENABLED_SERVICES.
#
# If you're planning to use the Quantum openvswitch plugin, set Q_PLUGIN to
# "openvswitch" and make sure the q-agt service is enabled in
# ENABLED_SERVICES.
#
# With Quantum networking the NET_MAN variable is ignored.
# MySQL & RabbitMQ
# ----------------
# We configure Nova, Horizon, Glance and Keystone to use MySQL as their
# database server. While they share a single server, each has their own
# database and tables.
# By default this script will install and configure MySQL. If you want to
# use an existing server, you can pass in the user/password/host parameters.
# You will need to send the same ``MYSQL_PASSWORD`` to every host if you are doing
# a multi-node devstack installation.
MYSQL_HOST=${MYSQL_HOST:-localhost}
MYSQL_USER=${MYSQL_USER:-root}
read_password MYSQL_PASSWORD "ENTER A PASSWORD TO USE FOR MYSQL."
# don't specify /db in this string, so we can use it for multiple services
BASE_SQL_CONN=${BASE_SQL_CONN:-mysql://$MYSQL_USER:$MYSQL_PASSWORD@$MYSQL_HOST}
# Rabbit connection info
RABBIT_HOST=${RABBIT_HOST:-localhost}
read_password RABBIT_PASSWORD "ENTER A PASSWORD TO USE FOR RABBIT."
# Glance connection info. Note the port must be specified.
GLANCE_HOSTPORT=${GLANCE_HOSTPORT:-$SERVICE_HOST:9292}
# SWIFT
# -----
# TODO: implement glance support
# TODO: add logging to different location.
# By default the location of swift drives and objects is located inside
# the swift source directory. SWIFT_DATA_LOCATION variable allow you to redefine
# this.
SWIFT_DATA_LOCATION=${SWIFT_DATA_LOCATION:-${SWIFT_DIR}/data}
# We are going to have the configuration files inside the source
# directory, change SWIFT_CONFIG_LOCATION if you want to adjust that.
SWIFT_CONFIG_LOCATION=${SWIFT_CONFIG_LOCATION:-${SWIFT_DIR}/config}
# devstack will create a loop-back disk formatted as XFS to store the
# swift data. By default the disk size is 1 gigabyte. The variable
# SWIFT_LOOPBACK_DISK_SIZE specified in bytes allow you to change
# that.
SWIFT_LOOPBACK_DISK_SIZE=${SWIFT_LOOPBACK_DISK_SIZE:-1000000}
# The ring uses a configurable number of bits from a paths MD5 hash as
# a partition index that designates a device. The number of bits kept
# from the hash is known as the partition power, and 2 to the partition
# power indicates the partition count. Partitioning the full MD5 hash
# ring allows other parts of the cluster to work in batches of items at
# once which ends up either more efficient or at least less complex than
# working with each item separately or the entire cluster all at once.
# By default we define 9 for the partition count (which mean 512).
SWIFT_PARTITION_POWER_SIZE=${SWIFT_PARTITION_POWER_SIZE:-9}
# We only ask for Swift Hash if we have enabled swift service.
if [[ "$ENABLED_SERVICES" =~ "swift" ]]; then
# SWIFT_HASH is a random unique string for a swift cluster that
# can never change.
read_password SWIFT_HASH "ENTER A RANDOM SWIFT HASH."
fi
# Keystone
# --------
# Service Token - Openstack components need to have an admin token
# to validate user tokens.
read_password SERVICE_TOKEN "ENTER A SERVICE_TOKEN TO USE FOR THE SERVICE ADMIN TOKEN."
# Horizon currently truncates usernames and passwords at 20 characters
read_password ADMIN_PASSWORD "ENTER A PASSWORD TO USE FOR HORIZON AND KEYSTONE (20 CHARS OR LESS)."
# Set Keystone interface configuration
KEYSTONE_AUTH_HOST=${KEYSTONE_AUTH_HOST:-$SERVICE_HOST}
KEYSTONE_AUTH_PORT=${KEYSTONE_AUTH_PORT:-35357}
KEYSTONE_AUTH_PROTOCOL=${KEYSTONE_AUTH_PROTOCOL:-http}
KEYSTONE_SERVICE_HOST=${KEYSTONE_SERVICE_HOST:-$SERVICE_HOST}
KEYSTONE_SERVICE_PORT=${KEYSTONE_SERVICE_PORT:-5000}
KEYSTONE_SERVICE_PROTOCOL=${KEYSTONE_SERVICE_PROTOCOL:-http}
# Log files
# ---------
# Set up logging for stack.sh
# Set LOGFILE to turn on logging
# We append '.xxxxxxxx' to the given name to maintain history
# where xxxxxxxx is a representation of the date the file was created
if [[ -n "$LOGFILE" ]]; then
# First clean up old log files. Use the user-specified LOGFILE
# as the template to search for, appending '.*' to match the date
# we added on earlier runs.
LOGDAYS=${LOGDAYS:-7}
LOGDIR=$(dirname "$LOGFILE")
LOGNAME=$(basename "$LOGFILE")
find $LOGDIR -maxdepth 1 -name $LOGNAME.\* -mtime +$LOGDAYS -exec rm {} \;
TIMESTAMP_FORMAT=${TIMESTAMP_FORMAT:-"%F-%H%M%S"}
LOGFILE=$LOGFILE.$(date "+$TIMESTAMP_FORMAT")
# Redirect stdout/stderr to tee to write the log file
exec 1> >( tee "${LOGFILE}" ) 2>&1
echo "stack.sh log $LOGFILE"
# Specified logfile name always links to the most recent log
ln -sf $LOGFILE $LOGDIR/$LOGNAME
fi
# So that errors don't compound we exit on any errors so you see only the
# first error that occurred.
trap failed ERR
failed() {
local r=$?
set +o xtrace
[ -n "$LOGFILE" ] && echo "${0##*/} failed: full log in $LOGFILE"
exit $r
}
# Print the commands being run so that we can see the command that triggers
# an error. It is also useful for following along as the install occurs.
set -o xtrace
# create the destination directory and ensure it is writable by the user
sudo mkdir -p $DEST
if [ ! -w $DEST ]; then
sudo chown `whoami` $DEST
fi
# Install Packages
# ================
#
# Openstack uses a fair number of other projects.
# - We are going to install packages only for the services needed.
# - We are parsing the packages files and detecting metadatas.
# - If there is a NOPRIME as comment mean we are not doing the install
# just yet.
# - If we have the meta-keyword dist:DISTRO or
# dist:DISTRO1,DISTRO2 it will be installed only for those
# distros (case insensitive).
function get_packages() {
local file_to_parse="general"
local service
for service in ${ENABLED_SERVICES//,/ }; do
# Allow individual services to specify dependencies
if [[ -e $FILES/apts/${service} ]]; then
file_to_parse="${file_to_parse} $service"
fi
if [[ $service == n-* ]]; then
if [[ ! $file_to_parse =~ nova ]]; then
file_to_parse="${file_to_parse} nova"
fi
elif [[ $service == g-* ]]; then
if [[ ! $file_to_parse =~ glance ]]; then
file_to_parse="${file_to_parse} glance"
fi
elif [[ $service == key* ]]; then
if [[ ! $file_to_parse =~ keystone ]]; then
file_to_parse="${file_to_parse} keystone"
fi
fi
done
for file in ${file_to_parse}; do
local fname=${FILES}/apts/${file}
local OIFS line package distros distro
[[ -e $fname ]] || { echo "missing: $fname"; exit 1 ;}
OIFS=$IFS
IFS=$'\n'
for line in $(<${fname}); do
if [[ $line =~ "NOPRIME" ]]; then
continue
fi
if [[ $line =~ (.*)#.*dist:([^ ]*) ]]; then # We are using BASH regexp matching feature.
package=${BASH_REMATCH[1]}
distros=${BASH_REMATCH[2]}
for distro in ${distros//,/ }; do #In bash ${VAR,,} will lowecase VAR
[[ ${distro,,} == ${DISTRO,,} ]] && echo $package
done
continue
fi
echo ${line%#*}
done
IFS=$OIFS
done
}
function pip_install {
[[ "$OFFLINE" = "True" ]] && return
sudo PIP_DOWNLOAD_CACHE=/var/cache/pip \
HTTP_PROXY=$http_proxy \
HTTPS_PROXY=$https_proxy \
pip install --use-mirrors $@
}
# install apt requirements
apt_get update
apt_get install $(get_packages)
# install python requirements
pip_install `cat $FILES/pips/* | uniq`
# git clone only if directory doesn't exist already. Since ``DEST`` might not
# be owned by the installation user, we create the directory and change the
# ownership to the proper user.
function git_clone {
[[ "$OFFLINE" = "True" ]] && return
GIT_REMOTE=$1
GIT_DEST=$2
GIT_BRANCH=$3
if echo $GIT_BRANCH | egrep -q "^refs"; then
# If our branch name is a gerrit style refs/changes/...
if [ ! -d $GIT_DEST ]; then
git clone $GIT_REMOTE $GIT_DEST
fi
cd $GIT_DEST
git fetch $GIT_REMOTE $GIT_BRANCH && git checkout FETCH_HEAD
else
# do a full clone only if the directory doesn't exist
if [ ! -d $GIT_DEST ]; then
git clone $GIT_REMOTE $GIT_DEST
cd $GIT_DEST
# This checkout syntax works for both branches and tags
git checkout $GIT_BRANCH
elif [[ "$RECLONE" == "yes" ]]; 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 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 $GIT_DEST -name '*.pyc' -delete
git checkout -f origin/$GIT_BRANCH
# a local branch might not exist
git branch -D $GIT_BRANCH || true
git checkout -b $GIT_BRANCH
fi
fi
}
# compute service
git_clone $NOVA_REPO $NOVA_DIR $NOVA_BRANCH
# python client library to nova that horizon (and others) use
git_clone $NOVACLIENT_REPO $NOVACLIENT_DIR $NOVACLIENT_BRANCH
# glance, swift middleware and nova api needs keystone middleware
if [[ "$ENABLED_SERVICES" =~ "key" ||
"$ENABLED_SERVICES" =~ "g-api" ||
"$ENABLED_SERVICES" =~ "n-api" ||
"$ENABLED_SERVICES" =~ "swift" ]]; then
# unified auth system (manages accounts/tokens)
git_clone $KEYSTONE_REPO $KEYSTONE_DIR $KEYSTONE_BRANCH
fi
if [[ "$ENABLED_SERVICES" =~ "swift" ]]; then
# storage service
git_clone $SWIFT_REPO $SWIFT_DIR $SWIFT_BRANCH
# swift + keystone middleware
git_clone $SWIFT_KEYSTONE_REPO $SWIFT_KEYSTONE_DIR $SWIFT_KEYSTONE_BRANCH
fi
if [[ "$ENABLED_SERVICES" =~ "g-api" ||
"$ENABLED_SERVICES" =~ "n-api" ]]; then
# image catalog service
git_clone $GLANCE_REPO $GLANCE_DIR $GLANCE_BRANCH
fi
if [[ "$ENABLED_SERVICES" =~ "n-novnc" ]]; then
# a websockets/html5 or flash powered VNC console for vm instances
git_clone $NOVNC_REPO $NOVNC_DIR $NOVNC_BRANCH
fi
if [[ "$ENABLED_SERVICES" =~ "horizon" ]]; then
# django powered web control panel for openstack
git_clone $HORIZON_REPO $HORIZON_DIR $HORIZON_BRANCH $HORIZON_TAG
git_clone $KEYSTONECLIENT_REPO $KEYSTONECLIENT_DIR $KEYSTONECLIENT_BRANCH
fi
if [[ "$ENABLED_SERVICES" =~ "openstackx" ]]; then
# openstackx is a collection of extensions to openstack.compute & nova
# that is *deprecated*. The code is being moved into python-novaclient & nova.
git_clone $OPENSTACKX_REPO $OPENSTACKX_DIR $OPENSTACKX_BRANCH
fi
if [[ "$ENABLED_SERVICES" =~ "q-svc" ]]; then
# quantum
git_clone $QUANTUM_REPO $QUANTUM_DIR $QUANTUM_BRANCH
fi
# Initialization
# ==============
# setup our checkouts so they are installed into python path
# allowing ``import nova`` or ``import glance.client``
if [[ "$ENABLED_SERVICES" =~ "key" ||
"$ENABLED_SERVICES" =~ "g-api" ||
"$ENABLED_SERVICES" =~ "n-api" ||
"$ENABLED_SERVICES" =~ "swift" ]]; then
cd $KEYSTONE_DIR; sudo python setup.py develop
fi
if [[ "$ENABLED_SERVICES" =~ "swift" ]]; then
cd $SWIFT_DIR; sudo python setup.py develop
cd $SWIFT_KEYSTONE_DIR; sudo python setup.py develop
fi
if [[ "$ENABLED_SERVICES" =~ "g-api" ||
"$ENABLED_SERVICES" =~ "n-api" ]]; then
cd $GLANCE_DIR; sudo python setup.py develop
fi
cd $NOVACLIENT_DIR; sudo python setup.py develop
cd $NOVA_DIR; sudo python setup.py develop
if [[ "$ENABLED_SERVICES" =~ "openstackx" ]]; then
cd $OPENSTACKX_DIR; sudo python setup.py develop
fi
if [[ "$ENABLED_SERVICES" =~ "horizon" ]]; then
cd $KEYSTONECLIENT_DIR; sudo python setup.py develop
cd $HORIZON_DIR/horizon; sudo python setup.py develop
cd $HORIZON_DIR/openstack-dashboard; sudo python setup.py develop
fi
if [[ "$ENABLED_SERVICES" =~ "q-svc" ]]; then
cd $QUANTUM_DIR; sudo python setup.py develop
fi
# Syslog
# ---------
if [[ $SYSLOG != "False" ]]; then
apt_get install -y rsyslog-relp
if [[ "$SYSLOG_HOST" = "$HOST_IP" ]]; then
# Configure the master host to receive
cat <<EOF >/tmp/90-stack-m.conf
\$ModLoad imrelp
\$InputRELPServerRun $SYSLOG_PORT
EOF
sudo mv /tmp/90-stack-m.conf /etc/rsyslog.d
else
# Set rsyslog to send to remote host
cat <<EOF >/tmp/90-stack-s.conf
*.* :omrelp:$SYSLOG_HOST:$SYSLOG_PORT
EOF
sudo mv /tmp/90-stack-s.conf /etc/rsyslog.d
fi
sudo /usr/sbin/service rsyslog restart
fi
# Rabbit
# ---------
if [[ "$ENABLED_SERVICES" =~ "rabbit" ]]; then
# Install and start rabbitmq-server
# the temp file is necessary due to LP: #878600
tfile=$(mktemp)
apt_get install rabbitmq-server > "$tfile" 2>&1
cat "$tfile"
rm -f "$tfile"
# change the rabbit password since the default is "guest"
sudo rabbitmqctl change_password guest $RABBIT_PASSWORD
fi
# Mysql
# ---------
if [[ "$ENABLED_SERVICES" =~ "mysql" ]]; then
# Seed configuration with mysql password so that apt-get install doesn't
# prompt us for a password upon install.
cat <<MYSQL_PRESEED | sudo debconf-set-selections
mysql-server-5.1 mysql-server/root_password password $MYSQL_PASSWORD
mysql-server-5.1 mysql-server/root_password_again password $MYSQL_PASSWORD
mysql-server-5.1 mysql-server/start_on_boot boolean true
MYSQL_PRESEED
# while ``.my.cnf`` is not needed for openstack to function, it is useful
# as it allows you to access the mysql databases via ``mysql nova`` instead
# of having to specify the username/password each time.
if [[ ! -e $HOME/.my.cnf ]]; then
cat <<EOF >$HOME/.my.cnf
[client]
user=$MYSQL_USER
password=$MYSQL_PASSWORD
host=$MYSQL_HOST
EOF
chmod 0600 $HOME/.my.cnf
fi
# Install and start mysql-server
apt_get install mysql-server
# Update the DB to give user $MYSQL_USER@% full control of the all databases:
sudo mysql -uroot -p$MYSQL_PASSWORD -h127.0.0.1 -e "GRANT ALL PRIVILEGES ON *.* TO '$MYSQL_USER'@'%' identified by '$MYSQL_PASSWORD';"
# Edit /etc/mysql/my.cnf to change bind-address from localhost (127.0.0.1) to any (0.0.0.0) and restart the mysql service:
sudo sed -i 's/127.0.0.1/0.0.0.0/g' /etc/mysql/my.cnf
sudo service mysql restart
fi
# Horizon
# ---------
# Setup the django horizon application to serve via apache/wsgi
if [[ "$ENABLED_SERVICES" =~ "horizon" ]]; then
# Install apache2, which is NOPRIME'd
apt_get install apache2 libapache2-mod-wsgi
# Horizon currently imports quantum even if you aren't using it. Instead
# of installing quantum we can create a simple module that will pass the
# initial imports
mkdir -p $HORIZON_DIR/openstack-dashboard/quantum || true
touch $HORIZON_DIR/openstack-dashboard/quantum/__init__.py
touch $HORIZON_DIR/openstack-dashboard/quantum/client.py
# ``local_settings.py`` is used to override horizon default settings.
local_settings=$HORIZON_DIR/openstack-dashboard/local/local_settings.py
cp $FILES/horizon_settings.py $local_settings
# Enable quantum in dashboard, if requested
if [[ "$ENABLED_SERVICES" =~ "quantum" ]]; then
sudo sed -e "s,QUANTUM_ENABLED = False,QUANTUM_ENABLED = True,g" -i $local_settings
fi
# Initialize the horizon database (it stores sessions and notices shown to
# users). The user system is external (keystone).
cd $HORIZON_DIR/openstack-dashboard
python manage.py syncdb
# create an empty directory that apache uses as docroot
sudo mkdir -p $HORIZON_DIR/.blackhole
## Configure apache's 000-default to run horizon
sudo cp $FILES/000-default.template /etc/apache2/sites-enabled/000-default
sudo sed -e "s,%USER%,$USER,g" -i /etc/apache2/sites-enabled/000-default
sudo sed -e "s,%HORIZON_DIR%,$HORIZON_DIR,g" -i /etc/apache2/sites-enabled/000-default
sudo service apache2 restart
fi
# Glance
# ------
if [[ "$ENABLED_SERVICES" =~ "g-reg" ]]; then
GLANCE_IMAGE_DIR=$DEST/glance/images
# Delete existing images
rm -rf $GLANCE_IMAGE_DIR
# Use local glance directories
mkdir -p $GLANCE_IMAGE_DIR
# (re)create glance database
mysql -u$MYSQL_USER -p$MYSQL_PASSWORD -e 'DROP DATABASE IF EXISTS glance;'
mysql -u$MYSQL_USER -p$MYSQL_PASSWORD -e 'CREATE DATABASE glance;'
function glance_config {
sudo sed -e "
s,%KEYSTONE_AUTH_HOST%,$KEYSTONE_AUTH_HOST,g;
s,%KEYSTONE_AUTH_PORT%,$KEYSTONE_AUTH_PORT,g;
s,%KEYSTONE_AUTH_PROTOCOL%,$KEYSTONE_AUTH_PROTOCOL,g;
s,%KEYSTONE_SERVICE_HOST%,$KEYSTONE_SERVICE_HOST,g;
s,%KEYSTONE_SERVICE_PORT%,$KEYSTONE_SERVICE_PORT,g;
s,%KEYSTONE_SERVICE_PROTOCOL%,$KEYSTONE_SERVICE_PROTOCOL,g;
s,%SQL_CONN%,$BASE_SQL_CONN/glance,g;
s,%SERVICE_TOKEN%,$SERVICE_TOKEN,g;
s,%DEST%,$DEST,g;
s,%SYSLOG%,$SYSLOG,g;
" -i $1
}
# Copy over our glance configurations and update them
GLANCE_REGISTRY_CONF=$GLANCE_DIR/etc/glance-registry.conf
cp $FILES/glance-registry.conf $GLANCE_REGISTRY_CONF
glance_config $GLANCE_REGISTRY_CONF
if [[ -e $FILES/glance-registry-paste.ini ]]; then
GLANCE_REGISTRY_PASTE_INI=$GLANCE_DIR/etc/glance-registry-paste.ini
cp $FILES/glance-registry-paste.ini $GLANCE_REGISTRY_PASTE_INI
glance_config $GLANCE_REGISTRY_PASTE_INI
# During the transition for Glance to the split config files
# we cat them together to handle both pre- and post-merge
cat $GLANCE_REGISTRY_PASTE_INI >>$GLANCE_REGISTRY_CONF
fi
GLANCE_API_CONF=$GLANCE_DIR/etc/glance-api.conf
cp $FILES/glance-api.conf $GLANCE_API_CONF
glance_config $GLANCE_API_CONF
if [[ -e $FILES/glance-api-paste.ini ]]; then
GLANCE_API_PASTE_INI=$GLANCE_DIR/etc/glance-api-paste.ini
cp $FILES/glance-api-paste.ini $GLANCE_API_PASTE_INI
glance_config $GLANCE_API_PASTE_INI
# During the transition for Glance to the split config files
# we cat them together to handle both pre- and post-merge
cat $GLANCE_API_PASTE_INI >>$GLANCE_API_CONF
fi
fi
# Nova
# ----
if [[ "$ENABLED_SERVICES" =~ "n-api" ]]; then
# We are going to use a sample http middleware configuration based on the
# one from the keystone project to launch nova. This paste config adds
# the configuration required for nova to validate keystone tokens.
# First we add a some extra data to the default paste config from nova
cat $NOVA_DIR/etc/nova/api-paste.ini $FILES/nova-api-paste.ini > $NOVA_DIR/bin/nova-api-paste.ini
# Then we add our own service token to the configuration
sed -e "s,%SERVICE_TOKEN%,$SERVICE_TOKEN,g" -i $NOVA_DIR/bin/nova-api-paste.ini
# Finally, we change the pipelines in nova to use keystone
function replace_pipeline() {
sed "/\[pipeline:$1\]/,/\[/s/^pipeline = .*/pipeline = $2/" -i $NOVA_DIR/bin/nova-api-paste.ini
}
replace_pipeline "ec2cloud" "ec2faultwrap logrequest totoken authtoken keystonecontext cloudrequest authorizer ec2executor"
replace_pipeline "ec2admin" "ec2faultwrap logrequest totoken authtoken keystonecontext adminrequest authorizer ec2executor"
replace_pipeline "openstack_api_v2" "faultwrap authtoken keystonecontext ratelimit osapi_app_v2"
replace_pipeline "openstack_compute_api_v2" "faultwrap authtoken keystonecontext ratelimit osapi_compute_app_v2"
replace_pipeline "openstack_volume_api_v1" "faultwrap authtoken keystonecontext ratelimit osapi_volume_app_v1"
fi
# Helper to clean iptables rules
function clean_iptables() {
# Delete rules
sudo iptables -S -v | sed "s/-c [0-9]* [0-9]* //g" | grep "nova" | grep "\-A" | sed "s/-A/-D/g" | awk '{print "sudo iptables",$0}' | bash
# Delete nat rules
sudo iptables -S -v -t nat | sed "s/-c [0-9]* [0-9]* //g" | grep "nova" | grep "\-A" | sed "s/-A/-D/g" | awk '{print "sudo iptables -t nat",$0}' | bash
# Delete chains
sudo iptables -S -v | sed "s/-c [0-9]* [0-9]* //g" | grep "nova" | grep "\-N" | sed "s/-N/-X/g" | awk '{print "sudo iptables",$0}' | bash
# Delete nat chains
sudo iptables -S -v -t nat | sed "s/-c [0-9]* [0-9]* //g" | grep "nova" | grep "\-N" | sed "s/-N/-X/g" | awk '{print "sudo iptables -t nat",$0}' | bash
}
if [[ "$ENABLED_SERVICES" =~ "n-cpu" ]]; then
# Virtualization Configuration
# ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
apt_get install libvirt-bin
# attempt to load modules: network block device - used to manage qcow images
sudo modprobe nbd || true
# Check for kvm (hardware based virtualization). If unable to initialize
# kvm, we drop back to the slower emulation mode (qemu). Note: many systems
# come with hardware virtualization disabled in BIOS.
if [[ "$LIBVIRT_TYPE" == "kvm" ]]; then
sudo modprobe kvm || true
if [ ! -e /dev/kvm ]; then
echo "WARNING: Switching to QEMU"
LIBVIRT_TYPE=qemu
fi
fi
# Install and configure **LXC** if specified. LXC is another approach to
# splitting a system into many smaller parts. LXC uses cgroups and chroot
# to simulate multiple systems.
if [[ "$LIBVIRT_TYPE" == "lxc" ]]; then
if [[ "$DISTRO" > natty ]]; then
apt_get install cgroup-lite
else
cgline="none /cgroup cgroup cpuacct,memory,devices,cpu,freezer,blkio 0 0"
sudo mkdir -p /cgroup
if ! grep -q cgroup /etc/fstab; then
echo "$cgline" | sudo tee -a /etc/fstab
fi
if ! mount -n | grep -q cgroup; then
sudo mount /cgroup
fi
fi
fi
# The user that nova runs as needs to be member of libvirtd group otherwise
# nova-compute will be unable to use libvirt.
sudo usermod -a -G libvirtd `whoami`
# libvirt detects various settings on startup, as we potentially changed
# the system configuration (modules, filesystems), we need to restart
# libvirt to detect those changes.
sudo /etc/init.d/libvirt-bin restart
# Instance Storage
# ~~~~~~~~~~~~~~~~
# Nova stores each instance in its own directory.
mkdir -p $NOVA_DIR/instances
# You can specify a different disk to be mounted and used for backing the
# virtual machines. If there is a partition labeled nova-instances we
# mount it (ext filesystems can be labeled via e2label).
if [ -L /dev/disk/by-label/nova-instances ]; then
if ! mount -n | grep -q $NOVA_DIR/instances; then
sudo mount -L nova-instances $NOVA_DIR/instances
sudo chown -R `whoami` $NOVA_DIR/instances
fi
fi
# Clean iptables from previous runs
clean_iptables
# Destroy old instances
instances=`virsh list --all | grep $INSTANCE_NAME_PREFIX | sed "s/.*\($INSTANCE_NAME_PREFIX[0-9a-fA-F]*\).*/\1/g"`
if [ ! $instances = "" ]; then
echo $instances | xargs -n1 virsh destroy || true
echo $instances | xargs -n1 virsh undefine || true
fi
# Clean out the instances directory.
sudo rm -rf $NOVA_DIR/instances/*
fi
if [[ "$ENABLED_SERVICES" =~ "n-net" ]]; then
# Delete traces of nova networks from prior runs
sudo killall dnsmasq || true
clean_iptables
rm -rf $NOVA_DIR/networks
mkdir -p $NOVA_DIR/networks
fi
# Storage Service
if [[ "$ENABLED_SERVICES" =~ "swift" ]]; then
# We first do a bit of setup by creating the directories and
# changing the permissions so we can run it as our user.
USER_GROUP=$(id -g)
sudo mkdir -p ${SWIFT_DATA_LOCATION}/drives
sudo chown -R $USER:${USER_GROUP} ${SWIFT_DATA_LOCATION}
# We then create a loopback disk and format it to XFS.
# TODO: Reset disks on new pass.
if [[ ! -e ${SWIFT_DATA_LOCATION}/drives/images/swift.img ]]; then
mkdir -p ${SWIFT_DATA_LOCATION}/drives/images
sudo touch ${SWIFT_DATA_LOCATION}/drives/images/swift.img
sudo chown $USER: ${SWIFT_DATA_LOCATION}/drives/images/swift.img
dd if=/dev/zero of=${SWIFT_DATA_LOCATION}/drives/images/swift.img \
bs=1024 count=0 seek=${SWIFT_LOOPBACK_DISK_SIZE}
mkfs.xfs -f -i size=1024 ${SWIFT_DATA_LOCATION}/drives/images/swift.img
fi
# After the drive being created we mount the disk with a few mount
# options to make it most efficient as possible for swift.
mkdir -p ${SWIFT_DATA_LOCATION}/drives/sdb1
if ! egrep -q ${SWIFT_DATA_LOCATION}/drives/sdb1 /proc/mounts; then
sudo mount -t xfs -o loop,noatime,nodiratime,nobarrier,logbufs=8 \
${SWIFT_DATA_LOCATION}/drives/images/swift.img ${SWIFT_DATA_LOCATION}/drives/sdb1
fi
# We then create link to that mounted location so swift would know
# where to go.
for x in {1..4}; do sudo ln -sf ${SWIFT_DATA_LOCATION}/drives/sdb1/$x ${SWIFT_DATA_LOCATION}/$x; done
# We now have to emulate a few different servers into one we
# create all the directories needed for swift
tmpd=""
for d in ${SWIFT_DATA_LOCATION}/drives/sdb1/{1..4} \
${SWIFT_CONFIG_LOCATION}/{object,container,account}-server \
${SWIFT_DATA_LOCATION}/{1..4}/node/sdb1 /var/run/swift; do
[[ -d $d ]] && continue
sudo install -o ${USER} -g $USER_GROUP -d $d
done
# We do want to make sure this is all owned by our user.
sudo chown -R $USER: ${SWIFT_DATA_LOCATION}/{1..4}/node
sudo chown -R $USER: ${SWIFT_CONFIG_LOCATION}
# swift-init has a bug using /etc/swift until bug #885595 is fixed
# we have to create a link
sudo ln -sf ${SWIFT_CONFIG_LOCATION} /etc/swift
# Swift use rsync to syncronize between all the different
# partitions (which make more sense when you have a multi-node
# setup) we configure it with our version of rsync.
sed -e "s/%GROUP%/${USER_GROUP}/;s/%USER%/$USER/;s,%SWIFT_DATA_LOCATION%,$SWIFT_DATA_LOCATION," $FILES/swift/rsyncd.conf | sudo tee /etc/rsyncd.conf
sudo sed -i '/^RSYNC_ENABLE=false/ { s/false/true/ }' /etc/default/rsync
# By default Swift will be installed with the tempauth middleware
# which has some default username and password if you have
# configured keystone it will checkout the directory.
if [[ "$ENABLED_SERVICES" =~ "key" ]]; then
swift_auth_server=keystone
# We install the memcache server as this is will be used by the
# middleware to cache the tokens auths for a long this is needed.
apt_get install memcached
# We need a special version of bin/swift which understand the
# OpenStack api 2.0, we download it until this is getting
# integrated in swift.
sudo https_proxy=$https_proxy curl -s -o/usr/local/bin/swift \
'https://review.openstack.org/gitweb?p=openstack/swift.git;a=blob_plain;f=bin/swift;hb=48bfda6e2fdf3886c98bd15649887d54b9a2574e'
else
swift_auth_server=tempauth
fi
# We do the install of the proxy-server and swift configuration
# replacing a few directives to match our configuration.
sed "s,%SWIFT_CONFIG_LOCATION%,${SWIFT_CONFIG_LOCATION},;s/%USER%/$USER/;s/%SERVICE_TOKEN%/${SERVICE_TOKEN}/;s/%AUTH_SERVER%/${swift_auth_server}/" \
$FILES/swift/proxy-server.conf|sudo tee ${SWIFT_CONFIG_LOCATION}/proxy-server.conf
sed -e "s/%SWIFT_HASH%/$SWIFT_HASH/" $FILES/swift/swift.conf > ${SWIFT_CONFIG_LOCATION}/swift.conf
# We need to generate a object/account/proxy configuration
# emulating 4 nodes on different ports we have a little function
# that help us doing that.
function generate_swift_configuration() {
local server_type=$1
local bind_port=$2
local log_facility=$3
local node_number
for node_number in {1..4}; do
node_path=${SWIFT_DATA_LOCATION}/${node_number}
sed -e "s,%SWIFT_CONFIG_LOCATION%,${SWIFT_CONFIG_LOCATION},;s,%USER%,$USER,;s,%NODE_PATH%,${node_path},;s,%BIND_PORT%,${bind_port},;s,%LOG_FACILITY%,${log_facility}," \
$FILES/swift/${server_type}-server.conf > ${SWIFT_CONFIG_LOCATION}/${server_type}-server/${node_number}.conf
bind_port=$(( ${bind_port} + 10 ))
log_facility=$(( ${log_facility} + 1 ))
done
}
generate_swift_configuration object 6010 2
generate_swift_configuration container 6011 2
generate_swift_configuration account 6012 2
# We have some specific configuration for swift for rsyslog. See
# the file /etc/rsyslog.d/10-swift.conf for more info.
swift_log_dir=${SWIFT_DATA_LOCATION}/logs
rm -rf ${swift_log_dir}
mkdir -p ${swift_log_dir}/hourly
sudo chown -R syslog:adm ${swift_log_dir}
sed "s,%SWIFT_LOGDIR%,${swift_log_dir}," $FILES/swift/rsyslog.conf | sudo \
tee /etc/rsyslog.d/10-swift.conf
sudo restart rsyslog
# We create two helper scripts :
#
# - swift-remakerings
# Allow to recreate rings from scratch.
# - swift-startmain
# Restart your full cluster.
#
sed -e "s,%SWIFT_CONFIG_LOCATION%,${SWIFT_CONFIG_LOCATION},;s/%SWIFT_PARTITION_POWER_SIZE%/$SWIFT_PARTITION_POWER_SIZE/" $FILES/swift/swift-remakerings | \
sudo tee /usr/local/bin/swift-remakerings
sudo install -m755 $FILES/swift/swift-startmain /usr/local/bin/
sudo chmod +x /usr/local/bin/swift-*
# We then can start rsync.
sudo /etc/init.d/rsync restart || :
# Create our ring for the object/container/account.
/usr/local/bin/swift-remakerings
# And now we launch swift-startmain to get our cluster running
# ready to be tested.
/usr/local/bin/swift-startmain || :
unset s swift_hash swift_auth_server tmpd
fi
# Volume Service
# --------------
if [[ "$ENABLED_SERVICES" =~ "n-vol" ]]; then
#
# Configure a default volume group called 'nova-volumes' for the nova-volume
# service if it does not yet exist. If you don't wish to use a file backed
# volume group, create your own volume group called 'nova-volumes' before
# invoking stack.sh.
#
# By default, the backing file is 2G in size, and is stored in /opt/stack.
# install the package
apt_get install tgt
if ! sudo vgs $VOLUME_GROUP; then
VOLUME_BACKING_FILE=${VOLUME_BACKING_FILE:-$DEST/nova-volumes-backing-file}
VOLUME_BACKING_FILE_SIZE=${VOLUME_BACKING_FILE_SIZE:-2052M}
# Only create if the file doesn't already exists
[[ -f $VOLUME_BACKING_FILE ]] || truncate -s $VOLUME_BACKING_FILE_SIZE $VOLUME_BACKING_FILE
DEV=`sudo losetup -f --show $VOLUME_BACKING_FILE`
# Only create if the loopback device doesn't contain $VOLUME_GROUP
if ! sudo vgs $VOLUME_GROUP; then sudo vgcreate $VOLUME_GROUP $DEV; fi
fi
if sudo vgs $VOLUME_GROUP; then
# Clean out existing volumes
for lv in `sudo lvs --noheadings -o lv_name $VOLUME_GROUP`; do
# VOLUME_NAME_PREFIX prefixes the LVs we want
if [[ "${lv#$VOLUME_NAME_PREFIX}" != "$lv" ]]; then
tid=`egrep "^tid.+$lv" /proc/net/iet/volume | cut -f1 -d' ' | tr ':' '='`
if [[ -n "$tid" ]]; then
lun=`egrep "lun.+$lv" /proc/net/iet/volume | cut -f1 -d' ' | tr ':' '=' | tr -d '\t'`
sudo ietadm --op delete --$tid --$lun
fi
sudo lvremove -f $VOLUME_GROUP/$lv
fi
done
fi
# tgt in oneiric doesn't restart properly if tgtd isn't running
# do it in two steps
sudo stop tgt || true
sudo start tgt
fi
function add_nova_flag {
echo "$1" >> $NOVA_DIR/bin/nova.conf
}
# (re)create nova.conf
rm -f $NOVA_DIR/bin/nova.conf
add_nova_flag "--verbose"
add_nova_flag "--allow_admin_api"
add_nova_flag "--scheduler_driver=$SCHEDULER"
add_nova_flag "--dhcpbridge_flagfile=$NOVA_DIR/bin/nova.conf"
add_nova_flag "--fixed_range=$FIXED_RANGE"
if [[ "$ENABLED_SERVICES" =~ "quantum" ]]; then
add_nova_flag "--network_manager=nova.network.quantum.manager.QuantumManager"
add_nova_flag "--quantum_connection_host=$Q_HOST"
add_nova_flag "--quantum_connection_port=$Q_PORT"
if [[ "$ENABLED_SERVICES" =~ "q-svc" && "$Q_PLUGIN" = "openvswitch" ]]; then
add_nova_flag "--libvirt_vif_type=ethernet"
add_nova_flag "--libvirt_vif_driver=nova.virt.libvirt.vif.LibvirtOpenVswitchDriver"
add_nova_flag "--linuxnet_interface_driver=nova.network.linux_net.LinuxOVSInterfaceDriver"
add_nova_flag "--quantum_use_dhcp"
fi
else
add_nova_flag "--network_manager=nova.network.manager.$NET_MAN"
fi
if [[ "$ENABLED_SERVICES" =~ "n-vol" ]]; then
add_nova_flag "--volume_group=$VOLUME_GROUP"
add_nova_flag "--volume_name_template=${VOLUME_NAME_PREFIX}%08x"
# oneiric no longer supports ietadm
add_nova_flag "--iscsi_helper=tgtadm"
fi
add_nova_flag "--my_ip=$HOST_IP"
add_nova_flag "--public_interface=$PUBLIC_INTERFACE"
add_nova_flag "--vlan_interface=$VLAN_INTERFACE"
add_nova_flag "--sql_connection=$BASE_SQL_CONN/nova"
add_nova_flag "--libvirt_type=$LIBVIRT_TYPE"
add_nova_flag "--instance_name_template=${INSTANCE_NAME_PREFIX}%08x"
if [[ "$ENABLED_SERVICES" =~ "openstackx" ]]; then
add_nova_flag "--osapi_compute_extension=nova.api.openstack.compute.contrib.standard_extensions"
add_nova_flag "--osapi_compute_extension=extensions.admin.Admin"
fi
if [[ "$ENABLED_SERVICES" =~ "n-novnc" ]]; then
NOVNCPROXY_URL=${NOVNCPROXY_URL:-"http://$SERVICE_HOST:6080/vnc_auto.html"}
add_nova_flag "--novncproxy_base_url=$NOVNCPROXY_URL"
fi
if [[ "$ENABLED_SERVICES" =~ "n-xvnc" ]]; then
XVPVNCPROXY_URL=${XVPVNCPROXY_URL:-"http://$SERVICE_HOST:6081/console"}
add_nova_flag "--xvpvncproxy_base_url=$XVPVNCPROXY_URL"
fi
if [ "$VIRT_DRIVER" = 'xenserver' ]; then
VNCSERVER_PROXYCLIENT_ADDRESS=${VNCSERVER_PROXYCLIENT_ADDRESS=169.254.0.1}
else
VNCSERVER_PROXYCLIENT_ADDRESS=${VNCSERVER_PROXYCLIENT_ADDRESS=127.0.0.1}
fi
add_nova_flag "--vncserver_proxyclient_address=$VNCSERVER_PROXYCLIENT_ADDRESS"
add_nova_flag "--api_paste_config=$NOVA_DIR/bin/nova-api-paste.ini"
add_nova_flag "--image_service=nova.image.glance.GlanceImageService"
add_nova_flag "--ec2_dmz_host=$EC2_DMZ_HOST"
add_nova_flag "--rabbit_host=$RABBIT_HOST"
add_nova_flag "--rabbit_password=$RABBIT_PASSWORD"
add_nova_flag "--glance_api_servers=$GLANCE_HOSTPORT"
add_nova_flag "--force_dhcp_release"
if [ -n "$INSTANCES_PATH" ]; then
add_nova_flag "--instances_path=$INSTANCES_PATH"
fi
if [ "$MULTI_HOST" != "False" ]; then
add_nova_flag "--multi_host"
add_nova_flag "--send_arp_for_ha"
fi
if [ "$SYSLOG" != "False" ]; then
add_nova_flag "--use_syslog"
fi
# You can define extra nova conf flags by defining the array EXTRA_FLAGS,
# For Example: EXTRA_FLAGS=(--foo --bar=2)
for I in "${EXTRA_FLAGS[@]}"; do
add_nova_flag $I
done
# XenServer
# ---------
if [ "$VIRT_DRIVER" = 'xenserver' ]; then
read_password XENAPI_PASSWORD "ENTER A PASSWORD TO USE FOR XEN."
add_nova_flag "--connection_type=xenapi"
add_nova_flag "--xenapi_connection_url=http://169.254.0.1"
add_nova_flag "--xenapi_connection_username=root"
add_nova_flag "--xenapi_connection_password=$XENAPI_PASSWORD"
add_nova_flag "--noflat_injected"
add_nova_flag "--flat_interface=eth1"
add_nova_flag "--flat_network_bridge=xapi1"
add_nova_flag "--public_interface=eth3"
# Need to avoid crash due to new firewall support
XEN_FIREWALL_DRIVER=${XEN_FIREWALL_DRIVER:-"nova.virt.firewall.IptablesFirewallDriver"}
add_nova_flag "--firewall_driver=$XEN_FIREWALL_DRIVER"
else
add_nova_flag "--connection_type=libvirt"
add_nova_flag "--flat_network_bridge=$FLAT_NETWORK_BRIDGE"
if [ -n "$FLAT_INTERFACE" ]; then
add_nova_flag "--flat_interface=$FLAT_INTERFACE"
fi
fi
# Nova Database
# ~~~~~~~~~~~~~
# All nova components talk to a central database. We will need to do this step
# only once for an entire cluster.
if [[ "$ENABLED_SERVICES" =~ "mysql" ]]; then
# (re)create nova database
mysql -u$MYSQL_USER -p$MYSQL_PASSWORD -e 'DROP DATABASE IF EXISTS nova;'
mysql -u$MYSQL_USER -p$MYSQL_PASSWORD -e 'CREATE DATABASE nova;'
# (re)create nova database
$NOVA_DIR/bin/nova-manage db sync
fi
# Keystone
# --------
if [[ "$ENABLED_SERVICES" =~ "key" ]]; then
# (re)create keystone database
mysql -u$MYSQL_USER -p$MYSQL_PASSWORD -e 'DROP DATABASE IF EXISTS keystone;'
mysql -u$MYSQL_USER -p$MYSQL_PASSWORD -e 'CREATE DATABASE keystone;'
# Configure keystone.conf
KEYSTONE_CONF=$KEYSTONE_DIR/etc/keystone.conf
cp $FILES/keystone.conf $KEYSTONE_CONF
sudo sed -e "s,%SQL_CONN%,$BASE_SQL_CONN/keystone,g" -i $KEYSTONE_CONF
sudo sed -e "s,%DEST%,$DEST,g" -i $KEYSTONE_CONF
# keystone_data.sh creates our admin user and our ``SERVICE_TOKEN``.
KEYSTONE_DATA=$KEYSTONE_DIR/bin/keystone_data.sh
cp $FILES/keystone_data.sh $KEYSTONE_DATA
sudo sed -e "
s,%KEYSTONE_AUTH_HOST%,$KEYSTONE_AUTH_HOST,g;
s,%KEYSTONE_AUTH_PORT%,$KEYSTONE_AUTH_PORT,g;
s,%KEYSTONE_AUTH_PROTOCOL%,$KEYSTONE_AUTH_PROTOCOL,g;
s,%KEYSTONE_SERVICE_HOST%,$KEYSTONE_SERVICE_HOST,g;
s,%KEYSTONE_SERVICE_PORT%,$KEYSTONE_SERVICE_PORT,g;
s,%KEYSTONE_SERVICE_PROTOCOL%,$KEYSTONE_SERVICE_PROTOCOL,g;
s,%SERVICE_HOST%,$SERVICE_HOST,g;
s,%SERVICE_TOKEN%,$SERVICE_TOKEN,g;
s,%ADMIN_PASSWORD%,$ADMIN_PASSWORD,g;
" -i $KEYSTONE_DATA
# initialize keystone with default users/endpoints
ENABLED_SERVICES=$ENABLED_SERVICES BIN_DIR=$KEYSTONE_DIR/bin bash $KEYSTONE_DATA
if [ "$SYSLOG" != "False" ]; then
sed -i -e '/^handlers=devel$/s/=devel/=production/' \
$KEYSTONE_DIR/etc/logging.cnf
sed -i -e "/^log_file/s/log_file/\#log_file/" \
$KEYSTONE_DIR/etc/keystone.conf
KEYSTONE_LOG_CONFIG="--log-config $KEYSTONE_DIR/etc/logging.cnf"
fi
fi
# Launch Services
# ===============
# nova api crashes if we start it with a regular screen command,
# so send the start command by forcing text into the window.
# Only run the services specified in ``ENABLED_SERVICES``
# our screen helper to launch a service in a hidden named screen
function screen_it {
NL=`echo -ne '\015'`
if [[ "$ENABLED_SERVICES" =~ "$1" ]]; then
if [[ "$USE_TMUX" =~ "yes" ]]; then
tmux new-window -t stack -a -n "$1" "bash"
tmux send-keys "$2" C-M
else
screen -S stack -X screen -t $1
# sleep to allow bash to be ready to be send the command - we are
# creating a new window in screen and then sends characters, so if
# bash isn't running by the time we send the command, nothing happens
sleep 1.5
screen -S stack -p $1 -X stuff "$2$NL"
fi
fi
}
# create a new named screen to run processes in
screen -d -m -S stack -t stack
sleep 1
# set a reasonable statusbar
screen -r stack -X hardstatus alwayslastline "%-Lw%{= BW}%50>%n%f* %t%{-}%+Lw%< %= %H"
# launch the glance registry service
if [[ "$ENABLED_SERVICES" =~ "g-reg" ]]; then
screen_it g-reg "cd $GLANCE_DIR; bin/glance-registry --config-file=etc/glance-registry.conf"
fi
# launch the glance api and wait for it to answer before continuing
if [[ "$ENABLED_SERVICES" =~ "g-api" ]]; then
screen_it g-api "cd $GLANCE_DIR; bin/glance-api --config-file=etc/glance-api.conf"
echo "Waiting for g-api ($GLANCE_HOSTPORT) to start..."
if ! timeout $SERVICE_TIMEOUT sh -c "while ! http_proxy= wget -q -O- http://$GLANCE_HOSTPORT; do sleep 1; done"; then
echo "g-api did not start"
exit 1
fi
fi
# launch the keystone and wait for it to answer before continuing
if [[ "$ENABLED_SERVICES" =~ "key" ]]; then
screen_it key "cd $KEYSTONE_DIR && $KEYSTONE_DIR/bin/keystone --config-file $KEYSTONE_CONF $KEYSTONE_LOG_CONFIG -d"
echo "Waiting for keystone to start..."
if ! timeout $SERVICE_TIMEOUT sh -c "while ! http_proxy= wget -q -O- $KEYSTONE_SERVICE_PROTOCOL://$KEYSTONE_SERVICE_HOST:$KEYSTONE_SERVICE_PORT; do sleep 1; done"; then
echo "keystone did not start"
exit 1
fi
fi
# launch the nova-api and wait for it to answer before continuing
if [[ "$ENABLED_SERVICES" =~ "n-api" ]]; then
screen_it n-api "cd $NOVA_DIR && $NOVA_DIR/bin/nova-api"
echo "Waiting for nova-api to start..."
if ! timeout $SERVICE_TIMEOUT sh -c "while ! http_proxy= wget -q -O- http://127.0.0.1:8774; do sleep 1; done"; then
echo "nova-api did not start"
exit 1
fi
fi
# Quantum service
if [[ "$ENABLED_SERVICES" =~ "q-svc" ]]; then
if [[ "$Q_PLUGIN" = "openvswitch" ]]; then
# Install deps
# FIXME add to files/apts/quantum, but don't install if not needed!
apt_get install openvswitch-switch openvswitch-datapath-dkms
# Create database for the plugin/agent
if [[ "$ENABLED_SERVICES" =~ "mysql" ]]; then
mysql -u$MYSQL_USER -p$MYSQL_PASSWORD -e 'CREATE DATABASE IF NOT EXISTS ovs_quantum;'
else
echo "mysql must be enabled in order to use the $Q_PLUGIN Quantum plugin."
exit 1
fi
QUANTUM_PLUGIN_INI_FILE=$QUANTUM_DIR/etc/plugins.ini
# Make sure we're using the openvswitch plugin
sed -i -e "s/^provider =.*$/provider = quantum.plugins.openvswitch.ovs_quantum_plugin.OVSQuantumPlugin/g" $QUANTUM_PLUGIN_INI_FILE
fi
screen_it q-svc "cd $QUANTUM_DIR && PYTHONPATH=.:$PYTHONPATH python $QUANTUM_DIR/bin/quantum-server $QUANTUM_DIR/etc/quantum.conf"
fi
# Quantum agent (for compute nodes)
if [[ "$ENABLED_SERVICES" =~ "q-agt" ]]; then
if [[ "$Q_PLUGIN" = "openvswitch" ]]; then
# Set up integration bridge
OVS_BRIDGE=${OVS_BRIDGE:-br-int}
sudo ovs-vsctl --no-wait -- --if-exists del-br $OVS_BRIDGE
sudo ovs-vsctl --no-wait add-br $OVS_BRIDGE
sudo ovs-vsctl --no-wait br-set-external-id $OVS_BRIDGE bridge-id br-int
# Start up the quantum <-> openvswitch agent
QUANTUM_OVS_CONFIG_FILE=$QUANTUM_DIR/etc/quantum/plugins/openvswitch/ovs_quantum_plugin.ini
sed -i -e "s/^sql_connection =.*$/sql_connection = mysql:\/\/$MYSQL_USER:$MYSQL_PASSWORD@$MYSQL_HOST\/ovs_quantum/g" $QUANTUM_OVS_CONFIG_FILE
screen_it q-agt "sleep 4; sudo python $QUANTUM_DIR/quantum/plugins/openvswitch/agent/ovs_quantum_agent.py $QUANTUM_OVS_CONFIG_FILE -v"
fi
fi
# If we're using Quantum (i.e. q-svc is enabled), network creation has to
# happen after we've started the Quantum service.
if [[ "$ENABLED_SERVICES" =~ "mysql" ]]; then
# create a small network
$NOVA_DIR/bin/nova-manage network create private $FIXED_RANGE 1 $FIXED_NETWORK_SIZE
if [[ "$ENABLED_SERVICES" =~ "q-svc" ]]; then
echo "Not creating floating IPs (not supported by QuantumManager)"
else
# create some floating ips
$NOVA_DIR/bin/nova-manage floating create $FLOATING_RANGE
# create a second pool
$NOVA_DIR/bin/nova-manage floating create --ip_range=$TEST_FLOATING_RANGE --pool=$TEST_FLOATING_POOL
fi
fi
# Launching nova-compute should be as simple as running ``nova-compute`` but
# have to do a little more than that in our script. Since we add the group
# ``libvirtd`` to our user in this script, when nova-compute is run it is
# within the context of our original shell (so our groups won't be updated).
# Use 'sg' to execute nova-compute as a member of the libvirtd group.
screen_it n-cpu "cd $NOVA_DIR && sg libvirtd $NOVA_DIR/bin/nova-compute"
screen_it n-vol "cd $NOVA_DIR && $NOVA_DIR/bin/nova-volume"
screen_it n-net "cd $NOVA_DIR && $NOVA_DIR/bin/nova-network"
screen_it n-sch "cd $NOVA_DIR && $NOVA_DIR/bin/nova-scheduler"
if [[ "$ENABLED_SERVICES" =~ "n-novnc" ]]; then
screen_it n-novnc "cd $NOVNC_DIR && ./utils/nova-novncproxy --flagfile $NOVA_DIR/bin/nova.conf --web ."
fi
if [[ "$ENABLED_SERVICES" =~ "n-xvnc" ]]; then
screen_it n-xvnc "cd $NOVA_DIR && ./bin/nova-xvpvncproxy --flagfile $NOVA_DIR/bin/nova.conf"
fi
if [[ "$ENABLED_SERVICES" =~ "n-cauth" ]]; then
screen_it n-cauth "cd $NOVA_DIR && ./bin/nova-consoleauth"
fi
if [[ "$ENABLED_SERVICES" =~ "horizon" ]]; then
screen_it horizon "cd $HORIZON_DIR && sudo tail -f /var/log/apache2/error.log"
fi
# Install Images
# ==============
# Upload an image to glance.
#
# The default image is a small ***TTY*** testing image, which lets you login
# the username/password of root/password.
#
# TTY also uses cloud-init, supporting login via keypair and sending scripts as
# userdata. See https://help.ubuntu.com/community/CloudInit for more on cloud-init
#
# Override ``IMAGE_URLS`` with a comma-separated list of uec images.
#
# * **natty**: http://uec-images.ubuntu.com/natty/current/natty-server-cloudimg-amd64.tar.gz
# * **oneiric**: http://uec-images.ubuntu.com/oneiric/current/oneiric-server-cloudimg-amd64.tar.gz
if [[ "$ENABLED_SERVICES" =~ "g-reg" ]]; then
# Create a directory for the downloaded image tarballs.
mkdir -p $FILES/images
# Option to upload legacy ami-tty, which works with xenserver
if [ $UPLOAD_LEGACY_TTY ]; then
if [ ! -f $FILES/tty.tgz ]; then
wget -c http://images.ansolabs.com/tty.tgz -O $FILES/tty.tgz
fi
tar -zxf $FILES/tty.tgz -C $FILES/images
RVAL=`glance add -A $SERVICE_TOKEN name="tty-kernel" is_public=true container_format=aki disk_format=aki < $FILES/images/aki-tty/image`
KERNEL_ID=`echo $RVAL | cut -d":" -f2 | tr -d " "`
RVAL=`glance add -A $SERVICE_TOKEN name="tty-ramdisk" is_public=true container_format=ari disk_format=ari < $FILES/images/ari-tty/image`
RAMDISK_ID=`echo $RVAL | cut -d":" -f2 | tr -d " "`
glance add -A $SERVICE_TOKEN name="tty" is_public=true container_format=ami disk_format=ami kernel_id=$KERNEL_ID ramdisk_id=$RAMDISK_ID < $FILES/images/ami-tty/image
fi
for image_url in ${IMAGE_URLS//,/ }; do
# Downloads the image (uec ami+aki style), then extracts it.
IMAGE_FNAME=`basename "$image_url"`
if [ ! -f $FILES/$IMAGE_FNAME ]; then
wget -c $image_url -O $FILES/$IMAGE_FNAME
fi
KERNEL=""
RAMDISK=""
case "$IMAGE_FNAME" in
*.tar.gz|*.tgz)
# Extract ami and aki files
[ "${IMAGE_FNAME%.tar.gz}" != "$IMAGE_FNAME" ] &&
IMAGE_NAME="${IMAGE_FNAME%.tar.gz}" ||
IMAGE_NAME="${IMAGE_FNAME%.tgz}"
xdir="$FILES/images/$IMAGE_NAME"
rm -Rf "$xdir";
mkdir "$xdir"
tar -zxf $FILES/$IMAGE_FNAME -C "$xdir"
KERNEL=$(for f in "$xdir/"*-vmlinuz*; do
[ -f "$f" ] && echo "$f" && break; done; true)
RAMDISK=$(for f in "$xdir/"*-initrd*; do
[ -f "$f" ] && echo "$f" && break; done; true)
IMAGE=$(for f in "$xdir/"*.img; do
[ -f "$f" ] && echo "$f" && break; done; true)
[ -n "$IMAGE_NAME" ]
IMAGE_NAME=$(basename "$IMAGE" ".img")
;;
*.img)
IMAGE="$FILES/$IMAGE_FNAME";
IMAGE_NAME=$(basename "$IMAGE" ".img")
;;
*.img.gz)
IMAGE="$FILES/${IMAGE_FNAME}"
IMAGE_NAME=$(basename "$IMAGE" ".img.gz")
;;
*) echo "Do not know what to do with $IMAGE_FNAME"; false;;
esac
# Use glance client to add the kernel the root filesystem.
# We parse the results of the first upload to get the glance ID of the
# kernel for use when uploading the root filesystem.
KERNEL_ID=""; RAMDISK_ID="";
if [ -n "$KERNEL" ]; then
RVAL=`glance add -A $SERVICE_TOKEN name="$IMAGE_NAME-kernel" is_public=true container_format=aki disk_format=aki < "$KERNEL"`
KERNEL_ID=`echo $RVAL | cut -d":" -f2 | tr -d " "`
fi
if [ -n "$RAMDISK" ]; then
RVAL=`glance add -A $SERVICE_TOKEN name="$IMAGE_NAME-ramdisk" is_public=true container_format=ari disk_format=ari < "$RAMDISK"`
RAMDISK_ID=`echo $RVAL | cut -d":" -f2 | tr -d " "`
fi
glance add -A $SERVICE_TOKEN name="${IMAGE_NAME%.img}" is_public=true container_format=ami disk_format=ami ${KERNEL_ID:+kernel_id=$KERNEL_ID} ${RAMDISK_ID:+ramdisk_id=$RAMDISK_ID} < <(zcat --force "${IMAGE}")
done
fi
# Fin
# ===
set +o xtrace
# Using the cloud
# ===============
echo ""
echo ""
echo ""
# If you installed the horizon on this server, then you should be able
# to access the site using your browser.
if [[ "$ENABLED_SERVICES" =~ "horizon" ]]; then
echo "horizon is now available at http://$SERVICE_HOST/"
fi
# If keystone is present, you can point nova cli to this server
if [[ "$ENABLED_SERVICES" =~ "key" ]]; then
echo "keystone is serving at $KEYSTONE_SERVICE_PROTOCOL://$KEYSTONE_SERVICE_HOST:$KEYSTONE_SERVICE_PORT/v2.0/"
echo "examples on using novaclient command line is in exercise.sh"
echo "the default users are: admin and demo"
echo "the password: $ADMIN_PASSWORD"
fi
# Echo HOST_IP - useful for build_uec.sh, which uses dhcp to give the instance an address
echo "This is your host ip: $HOST_IP"
# Indicate how long this took to run (bash maintained variable 'SECONDS')
echo "stack.sh completed in $SECONDS seconds."