devstack/exercises/volumes.sh
Dean Troyer 27e326995a Simplify die_if_error
* Replace die_if_error() with the simpler die()
* Attempt to clean up unnecessary trace output
* Formatting cleanups on all exercise scripts

Change-Id: I72a542b3a59ee9bf12bee6bcc605edd7579205e0
2012-03-19 10:06:15 -05:00

177 lines
5.2 KiB
Bash
Executable File

#!/usr/bin/env bash
# **volumes.sh**
# Test nova volumes with the nova command from python-novaclient
echo "*********************************************************************"
echo "Begin DevStack Exercise: $0"
echo "*********************************************************************"
# This script exits on an error so that errors don't compound and you see
# only the first error that occured.
set -o errexit
# Print the commands being run so that we can see the command that triggers
# an error. It is also useful for following allowing as the install occurs.
set -o xtrace
# Settings
# ========
# Keep track of the current directory
EXERCISE_DIR=$(cd $(dirname "$0") && pwd)
TOP_DIR=$(cd $EXERCISE_DIR/..; pwd)
# Import common functions
source $TOP_DIR/functions
# Import configuration
source $TOP_DIR/openrc
# Import exercise configuration
source $TOP_DIR/exerciserc
# Instance type to create
DEFAULT_INSTANCE_TYPE=${DEFAULT_INSTANCE_TYPE:-m1.tiny}
# Boot this image, use first AMi image if unset
DEFAULT_IMAGE_NAME=${DEFAULT_IMAGE_NAME:-ami}
# Launching a server
# ==================
# List servers for tenant:
nova list
# Images
# ------
# Nova has a **deprecated** way of listing images.
nova image-list
# But we recommend using glance directly
glance -f index
# Grab the id of the image to launch
IMAGE=`glance -f index | egrep $DEFAULT_IMAGE_NAME | head -1 | cut -d" " -f1`
# determinine instance type
# -------------------------
# List of instance types:
nova flavor-list
INSTANCE_TYPE=`nova flavor-list | grep $DEFAULT_INSTANCE_TYPE | get_field 1`
if [[ -z "$INSTANCE_TYPE" ]]; then
# grab the first flavor in the list to launch if default doesn't exist
INSTANCE_TYPE=`nova flavor-list | head -n 4 | tail -n 1 | get_field 1`
fi
NAME="ex-vol"
VM_UUID=`nova boot --flavor $INSTANCE_TYPE --image $IMAGE $NAME --security_groups=$SECGROUP | grep ' id ' | get_field 2`
die_if_not_set VM_UUID "Failure launching $NAME"
# Testing
# =======
# First check if it spins up (becomes active and responds to ping on
# internal ip). If you run this script from a nova node, you should
# bypass security groups and have direct access to the server.
# Waiting for boot
# ----------------
# check that the status is active within ACTIVE_TIMEOUT seconds
if ! timeout $ACTIVE_TIMEOUT sh -c "while ! nova show $VM_UUID | grep status | grep -q ACTIVE; do sleep 1; done"; then
echo "server didn't become active!"
exit 1
fi
# get the IP of the server
IP=`nova show $VM_UUID | grep "private network" | get_field 2`
die_if_not_set IP "Failure retrieving IP address"
# for single node deployments, we can ping private ips
MULTI_HOST=${MULTI_HOST:-0}
if [ "$MULTI_HOST" = "0" ]; then
# sometimes the first ping fails (10 seconds isn't enough time for the VM's
# network to respond?), so let's ping for a default of 15 seconds with a
# timeout of a second for each ping.
if ! timeout $BOOT_TIMEOUT sh -c "while ! ping -c1 -w1 $IP; do sleep 1; done"; then
echo "Couldn't ping server"
exit 1
fi
else
# On a multi-host system, without vm net access, do a sleep to wait for the boot
sleep $BOOT_TIMEOUT
fi
# Volumes
# -------
VOL_NAME="myvol-$(openssl rand -hex 4)"
# Verify it doesn't exist
if [[ -n "`nova volume-list | grep $VOL_NAME | head -1 | get_field 2`" ]]; then
echo "Volume $VOL_NAME already exists"
exit 1
fi
# Create a new volume
nova volume-create --display_name $VOL_NAME --display_description "test volume: $VOL_NAME" 1
if [[ $? != 0 ]]; then
echo "Failure creating volume $VOL_NAME"
exit 1
fi
if ! timeout $ACTIVE_TIMEOUT sh -c "while ! nova volume-list | grep $VOL_NAME | grep available; do sleep 1; done"; then
echo "Volume $VOL_NAME not created"
exit 1
fi
# Get volume ID
VOL_ID=`nova volume-list | grep $VOL_NAME | head -1 | get_field 1`
die_if_not_set VOL_ID "Failure retrieving volume ID for $VOL_NAME"
# Attach to server
DEVICE=/dev/vdb
nova volume-attach $VM_UUID $VOL_ID $DEVICE || \
die "Failure attaching volume $VOL_NAME to $NAME"
if ! timeout $ACTIVE_TIMEOUT sh -c "while ! nova volume-list | grep $VOL_NAME | grep in-use; do sleep 1; done"; then
echo "Volume $VOL_NAME not attached to $NAME"
exit 1
fi
VOL_ATTACH=`nova volume-list | grep $VOL_NAME | head -1 | get_field -1`
die_if_not_set VOL_ATTACH "Failure retrieving $VOL_NAME status"
if [[ "$VOL_ATTACH" != $VM_UUID ]]; then
echo "Volume not attached to correct instance"
exit 1
fi
# Detach volume
nova volume-detach $VM_UUID $VOL_ID || die "Failure detaching volume $VOL_NAME from $NAME"
if ! timeout $ACTIVE_TIMEOUT sh -c "while ! nova volume-list | grep $VOL_NAME | grep available; do sleep 1; done"; then
echo "Volume $VOL_NAME not detached from $NAME"
exit 1
fi
# Delete volume
nova volume-delete $VOL_ID || die "Failure deleting volume $VOL_NAME"
if ! timeout $ACTIVE_TIMEOUT sh -c "while ! nova volume-list | grep $VOL_NAME; do sleep 1; done"; then
echo "Volume $VOL_NAME not deleted"
exit 1
fi
# shutdown the server
nova delete $NAME || die "Failure deleting instance $NAME"
set +o xtrace
echo "*********************************************************************"
echo "SUCCESS: End DevStack Exercise: $0"
echo "*********************************************************************"