Add multiple VM support

Add support to run Doctor test with more than one VM by setting
VM_COUNT to wanted amount of VMs. Normally half of those VMs
gets scheduled to host that is used for Doctor testing.

JIRA: DOCTOR-77

Change-Id: I9fac96c6f53c290fccc35a2fd933b9fb106b96f2
Signed-off-by: Tomi Juvonen <tomi.juvonen@nokia.com>

2 files changed, 180 insertions, 70 deletions

diff --git a/tests/inspector.py b/tests/inspector.py
index c1f95697..7195969a 100644
--- a/tests/inspector.py
+++ b/tests/inspector.py
@@ -14,6 +14,7 @@ from flask import request
 import json
 import logger as doctor_log
 import os
+import threading
 import time
 
 import novaclient.client as novaclient
@@ -23,19 +24,41 @@ import nova_force_down
 LOG = doctor_log.Logger('doctor_inspector').getLogger()
 
 
+class ThreadedResetState(threading.Thread):
+
+    def __init__(self, nova, state, server):
+        threading.Thread.__init__(self)
+        self.nova = nova
+        self.state = state
+        self.server = server
+
+    def run(self):
+        self.nova.servers.reset_state(self.server, self.state)
+        LOG.info('doctor mark vm(%s) error at %s' % (self.server, time.time()))
+
+
 class DoctorInspectorSample(object):
 
-    nova_api_version = '2.11'
+    NOVA_API_VERSION = '2.11'
+    NUMBER_OF_CLIENTS = 50
+    # TODO(tojuvone): This could be enhanced in future with dynamic
+    # reuse of self.novaclients when all threads in use and
+    # self.NUMBER_OF_CLIENTS based on amount of cores or overriden by input
+    # argument
 
     def __init__(self):
         self.servers = collections.defaultdict(list)
-        self.nova = novaclient.Client(self.nova_api_version,
-                                      os.environ['OS_USERNAME'],
-                                      os.environ['OS_PASSWORD'],
-                                      os.environ['OS_TENANT_NAME'],
-                                      os.environ['OS_AUTH_URL'],
-                                      connection_pool=True)
-        # check nova is available
+        self.novaclients = list()
+        # Pool of novaclients for redundant usage
+        for i in range(self.NUMBER_OF_CLIENTS):
+            self.novaclients.append(novaclient.Client(self.NOVA_API_VERSION,
+                                    os.environ['OS_USERNAME'],
+                                    os.environ['OS_PASSWORD'],
+                                    os.environ['OS_TENANT_NAME'],
+                                    os.environ['OS_AUTH_URL'],
+                                    connection_pool=True))
+        # Normally we use this client for non redundant API calls
+        self.nova=self.novaclients[0]
         self.nova.servers.list(detailed=False)
         self.init_servers_list()
 
@@ -52,10 +75,18 @@ class DoctorInspectorSample(object):
                 LOG.error('can not get hostname from server=%s' % server)
 
     def disable_compute_host(self, hostname):
-        for server in self.servers[hostname]:
-            self.nova.servers.reset_state(server, 'error')
-            LOG.info('doctor mark vm(%s) error at %s' % (server, time.time()))
-
+        threads = []
+        if len(self.servers[hostname]) > self.NUMBER_OF_CLIENTS:
+            # TODO(tojuvone): This could be enhanced in future with dynamic
+            # reuse of self.novaclients when all threads in use
+            LOG.error('%d servers in %s. Can handle only %d'%(
+                      self.servers[hostname], hostname, self.NUMBER_OF_CLIENTS))
+        for nova, server in zip(self.novaclients, self.servers[hostname]):
+            t = ThreadedResetState(nova, "error", server)
+            t.start()
+            threads.append(t)
+        for t in threads:
+            t.join()
         # NOTE: We use our own client here instead of this novaclient for a
         #       workaround.  Once keystone provides v2.1 nova api endpoint
         #       in the service catalog which is configured by OpenStack
diff --git a/tests/run.sh b/tests/run.sh
index 32562221..855ecfc6 100755
--- a/tests/run.sh
+++ b/tests/run.sh
@@ -17,9 +17,11 @@ IMAGE_URL=https://launchpad.net/cirros/trunk/0.3.0/+download/cirros-0.3.0-x86_64
 IMAGE_NAME=${IMAGE_NAME:-cirros}
 IMAGE_FILE="${IMAGE_NAME}.img"
 IMAGE_FORMAT=qcow2
-VM_NAME=doctor_vm1
+VM_BASENAME=doctor_vm
 VM_FLAVOR=m1.tiny
-ALARM_NAME=doctor_alarm1
+#if VM_COUNT set, use that instead
+VM_COUNT=${VM_COUNT:-1}
+ALARM_BASENAME=doctor_alarm
 INSPECTOR_PORT=12345
 CONSUMER_PORT=12346
 DOCTOR_USER=doctor
@@ -38,8 +40,8 @@ as_doctor_user="--os-username $DOCTOR_USER --os-password $DOCTOR_PW
 # Functions
 
 get_compute_host_info() {
-    # get computer host info which VM boot in
-    COMPUTE_HOST=$(openstack $as_doctor_user server show $VM_NAME |
+    # get computer host info which first VM boot in
+    COMPUTE_HOST=$(openstack $as_doctor_user server show ${VM_BASENAME}1 |
                    grep "OS-EXT-SRV-ATTR:host" | awk '{ print $4 }')
     compute_host_in_undercloud=${COMPUTE_HOST%%.*}
     die_if_not_set $LINENO COMPUTE_HOST "Failed to get compute hostname"
@@ -113,31 +115,60 @@ create_test_user() {
         openstack role add "$DOCTOR_ROLE" --user "$DOCTOR_USER" \
                            --project "$DOCTOR_PROJECT"
     }
+    # tojuvone: openstack quota show is broken and have to use nova
+    # https://bugs.launchpad.net/manila/+bug/1652118
+    # Note! while it is encouraged to use openstack client it has proven
+    # quite buggy.
+    # QUOTA=$(openstack quota show $DOCTOR_PROJECT)
+    DOCTOR_QUOTA=$(nova quota-show --tenant DOCTOR_PROJECT)
+    # We make sure that quota allows number of instances and cores
+    OLD_INSTANCE_QUOTA=$(echo "${DOCTOR_QUOTA}" | grep " instances " | \
+                         awk '{print $4}')
+    if [ $OLD_INSTANCE_QUOTA -lt $VM_COUNT ]; then
+        openstack quota set --instances $VM_COUNT \
+                  $DOCTOR_USER
+    fi
+    OLD_CORES_QUOTA=$(echo "${DOCTOR_QUOTA}" | grep " cores " | \
+                      awk '{print $4}')
+    if [ $OLD_CORES_QUOTA -lt $VM_COUNT ]; then
+        openstack quota set --cores $VM_COUNT \
+                  $DOCTOR_USER
+    fi
 }
 
 boot_vm() {
     # test VM done with test user, so can test non-admin
-    openstack $as_doctor_user server list | grep -q " $VM_NAME " && return 0
-    openstack $as_doctor_user server create --flavor "$VM_FLAVOR" \
+    servers=$(openstack $as_doctor_user server list)
+    for i in `seq $VM_COUNT`; do
+        echo "${servers}" | grep -q " $VM_BASENAME$i " && continue
+        openstack $as_doctor_user server create --flavor "$VM_FLAVOR" \
                             --image "$IMAGE_NAME" \
-                            "$VM_NAME"
+                            "$VM_BASENAME$i"
+    done
     sleep 1
 }
 
 create_alarm() {
     # get vm_id as test user
-    ceilometer $as_doctor_user alarm-list | grep -q " $ALARM_NAME " && return 0
-    vm_id=$(openstack $as_doctor_user server list | grep " $VM_NAME " | awk '{print $2}')
-    # TODO(r-mibu): change notification endpoint from localhost to the consumer
-    # IP address (functest container).
-    ceilometer $as_doctor_user alarm-event-create --name "$ALARM_NAME" \
-        --alarm-action "http://localhost:$CONSUMER_PORT/failure" \
-        --description "VM failure" \
-        --enabled True \
-        --repeat-actions False \
-        --severity "moderate" \
-        --event-type compute.instance.update \
-        -q "traits.state=string::error; traits.instance_id=string::$vm_id"
+    alarm_list=$(ceilometer $as_doctor_user alarm-list)
+    vms=$(openstack $as_doctor_user server list)
+    for i in `seq $VM_COUNT`; do
+        echo "${alarm_list}" | grep -q " $ALARM_BASENAME$i " || {
+            vm_id=$(echo "${vms}" | grep " $VM_BASENAME$i " | awk '{print $2}')
+            # TODO(r-mibu): change notification endpoint from localhost to the
+            # consumer. IP address (functest container).
+            ceilometer $as_doctor_user alarm-event-create \
+                       --name "$ALARM_BASENAME$i" \
+                       --alarm-action "http://localhost:$CONSUMER_PORT/failure" \
+                       --description "VM failure" \
+                       --enabled True \
+                       --repeat-actions False \
+                       --severity "moderate" \
+                       --event-type compute.instance.update \
+                       -q "traits.state=string::error; \
+                       traits.instance_id=string::$vm_id"
+            }
+     done
 }
 
 start_monitor() {
@@ -189,19 +220,31 @@ wait_for_vm_launch() {
     count=0
     while [[ ${count} -lt 60 ]]
     do
-        state=$(openstack $as_doctor_user server list | grep " $VM_NAME " | awk '{print $6}')
-        if [[ "$state" == "ACTIVE" ]]; then
-            # NOTE(cgoncalves): sleeping for a bit to stabilize
-            # See python-openstackclient/functional/tests/compute/v2/test_server.py:wait_for_status
-            sleep 5
+        active_count=0
+        vms=$(openstack $as_doctor_user server list)
+        for i in `seq $VM_COUNT`; do
+            state=$(echo "${vms}" | grep " $VM_BASENAME$i " | awk '{print $6}')
+            if [[ "$state" == "ACTIVE" ]]; then
+                active_count=$(($active_count+1))
+            elif [[ "$state" == "ERROR" ]]; then
+                die $LINENO "vm state $VM_BASENAME$i is ERROR"
+            else
+                #This VM not yet active
+                count=$(($count+1))
+                sleep 5
+                continue
+            fi
+        done
+        [[ $active_count -eq $VM_COUNT ]] && {
+            echo "get computer host info..."
+            get_compute_host_info
+            VMS_ON_FAILED_HOST=$(openstack $as_doctor_user server list --host \
+                         $COMPUTE_HOST | grep " ${VM_BASENAME}" |  wc -l)
             return 0
-        fi
-        if [[ "$state" == "ERROR" ]]; then
-            openstack $as_doctor_user server show $VM_NAME
-            die $LINENO "vm state is ERROR"
-        fi
+        }
+        #Not all VMs active
         count=$(($count+1))
-        sleep 1
+        sleep 5
     done
     die $LINENO "Time out while waiting for VM launch"
 }
@@ -227,17 +270,27 @@ END_TXT
     triggered=$(date "+%s.%N")
 }
 
-calculate_notification_time() {
-    if ! grep -q "doctor consumer notified at" consumer.log ; then
-        die $LINENO "Consumer hasn't received fault notification."
-    fi
+wait_consumer() {
+    local interval=1
+    local rounds=$(($1 / $interval))
+    for i in `seq $rounds`; do
+        notified_count=$(grep "doctor consumer notified at" consumer.log | wc -l)
+        if [[ $notified_count -eq  $VMS_ON_FAILED_HOST ]]; then
+            return 0
+        fi
+        sleep $interval
+    done
+    die $LINENO "Consumer hasn't received fault notification."
+}
 
+calculate_notification_time() {
+    wait_consumer 60
     #keep 'at' as the last keyword just before the value, and
     #use regex to get value instead of the fixed column
     detected=$(grep "doctor monitor detected at" monitor.log |\
                sed -e "s/^.* at //")
     notified=$(grep "doctor consumer notified at" consumer.log |\
-               sed -e "s/^.* at //")
+               sed -e "s/^.* at //" | tail -1)
 
     echo "$notified $detected" | \
         awk '{
@@ -247,17 +300,42 @@ calculate_notification_time() {
         }'
 }
 
+wait_ping() {
+    local interval=5
+    local rounds=$(($1 / $interval))
+    for i in `seq $rounds`; do
+        ping -c 1 "$COMPUTE_IP"
+        if [[ $? -ne 0 ]] ; then
+            sleep $interval
+            continue
+        fi
+        return 0
+    done
+}
+
 check_host_status() {
+    # Check host related to first Doctor VM is in wanted state
+    # $1    Expected state
+    # $2    Seconds to wait to have wanted state
     expected_state=$1
-
-    host_status_line=$(openstack $as_doctor_user --os-compute-api-version 2.16 \
-                       server show $VM_NAME | grep "host_status")
-    host_status=$(echo $host_status_line | awk '{print $4}')
-    die_if_not_set $LINENO host_status "host_status not reported by: nova show $VM_NAME"
+    local interval=5
+    local rounds=$(($2 / $interval))
+    for i in `seq $rounds`; do
+        host_status_line=$(openstack $as_doctor_user --os-compute-api-version \
+                           2.16 server show ${VM_BASENAME}1 | grep "host_status")
+        host_status=$(echo $host_status_line | awk '{print $4}')
+        die_if_not_set $LINENO host_status "host_status not reported by: nova show ${VM_BASENAME}1"
+        if [[ "$expected_state" =~ "$host_status" ]] ; then
+            echo "${VM_BASENAME}1 showing host_status: $host_status"
+            return 0
+        else
+            sleep $interval
+        fi
+    done
     if [[ "$expected_state" =~ "$host_status" ]] ; then
-        echo "$VM_NAME showing host_status: $host_status"
+        echo "${VM_BASENAME}1 showing host_status: $host_status"
     else
-        die $LINENO "host_status:$host_status not equal to expected_state: $expected_state"
+        die $LINENO  "host_status:$host_status not equal to expected_state: $expected_state"
     fi
 }
 
@@ -320,18 +398,23 @@ cleanup() {
     stop_consumer
 
     unset_forced_down_hosts
-    # TODO: We need to make sure the target compute host is back to IP
-    #       reachable. wait_ping() will be added by tojuvone .
-    sleep 110
-    if is_set COMPUTE_IP; then
-        scp $ssh_opts_cpu "$COMPUTE_USER@$COMPUTE_IP:disable_network.log" .
-    fi
 
-    openstack $as_doctor_user server list | grep -q " $VM_NAME " && openstack $as_doctor_user server delete "$VM_NAME"
-    sleep 1
-    alarm_id=$(ceilometer $as_doctor_user alarm-list | grep " $ALARM_NAME " | awk '{print $2}')
-    sleep 1
-    [ -n "$alarm_id" ] && ceilometer $as_doctor_user alarm-delete "$alarm_id"
+    wait_ping 120
+
+    scp $ssh_opts_cpu "$COMPUTE_USER@$COMPUTE_IP:disable_network.log" .
+    vms=$(openstack $as_doctor_user server list)
+    vmstodel=""
+    for i in `seq $VM_COUNT`; do
+        $(echo "${vms}" | grep -q " $VM_BASENAME$i ") &&
+        vmstodel+=" $VM_BASENAME$i"
+    done
+    [[ $vmstodel ]] && openstack $as_doctor_user server delete $vmstodel
+    alarm_list=$(ceilometer $as_doctor_user alarm-list)
+    for i in `seq $VM_COUNT`; do
+        alarm_id=$(echo "${alarm_list}" | grep " $ALARM_BASENAME$i " |
+                   awk '{print $2}')
+        [ -n "$alarm_id" ] && ceilometer $as_doctor_user alarm-delete "$alarm_id"
+    done
     sleep 1
 
     image_id=$(openstack image list | grep " $IMAGE_NAME " | awk '{print $2}')
@@ -373,9 +456,6 @@ echo "creating VM..."
 boot_vm
 wait_for_vm_launch
 
-echo "get computer host info..."
-get_compute_host_info
-
 echo "creating alarm..."
 #TODO: change back to use, network problems depends on infra and installers
 #get_consumer_ip
@@ -389,9 +469,8 @@ start_consumer
 sleep 60
 echo "injecting host failure..."
 inject_failure
-sleep 60
 
-check_host_status "(DOWN|UNKNOWN)"
+check_host_status "(DOWN|UNKNOWN)" 60
 calculate_notification_time
 collect_logs
 run_profiler