Change the deployment mode from 'nonha' to 'noha'
[joid.git] / ci / deploy.sh
index c455727..8491ad4 100755 (executable)
@@ -6,13 +6,16 @@ set -ex
 # Below parameters are the default and we can according the release
 
 opnfvsdn=nosdn
-opnfvtype=nonha
-openstack=mitaka
+opnfvtype=noha
+openstack=ocata
 opnfvlab=default
-opnfvrel=c
+opnfvrel=e
 opnfvfeature=none
 opnfvdistro=xenial
 opnfvarch=amd64
+opnfvmodel=openstack
+
+jujuver=`juju --version`
 
 read_config() {
     opnfvrel=`grep release: deploy.yaml | cut -d ":" -f2`
@@ -23,15 +26,16 @@ read_config() {
 }
 
 usage() { echo "Usage: $0 [-s <nosdn|odl|opencontrail>]
-                         [-t <nonha|ha|tip>]
+                         [-t <noha|ha|tip>]
                          [-o <juno|liberty>]
                          [-l <default|intelpod5>]
                          [-f <ipv6,dpdk,lxd,dvr>]
                          [-d <trusty|xenial>]
                          [-a <amd64>]
+                         [-m <openstack|kubernetes>]
                          [-r <a|b>]" 1>&2 exit 1; }
 
-while getopts ":s:t:o:l:h:r:f:d:a:" opt; do
+while getopts ":s:t:o:l:h:r:f:d:a:m:" opt; do
     case "${opt}" in
         s)
             opnfvsdn=${OPTARG}
@@ -57,6 +61,9 @@ while getopts ":s:t:o:l:h:r:f:d:a:" opt; do
         a)
             opnfvarch=${OPTARG}
             ;;
+        m)
+            opnfvmodel=${OPTARG}
+            ;;
         h)
             usage
             ;;
@@ -65,70 +72,91 @@ while getopts ":s:t:o:l:h:r:f:d:a:" opt; do
     esac
 done
 
-deploy_dep() {
-    sudo apt-add-repository ppa:juju/stable -y
-    sudo apt-get update
-    sudo apt-get install juju git juju-deployer -y
-    juju init -f
-    cp environments.yaml ~/.juju/
-}
-
 #by default maas creates two VMs in case of three more VM needed.
 createresource() {
-    maas_ip=`grep " ip_address" deployment.yaml | cut -d " "  -f 10`
-    apikey=`grep maas-oauth: environments.yaml | cut -d "'" -f 2`
-    maas login maas http://${maas_ip}/MAAS/api/1.0 ${apikey}
-
-    nodeexist=`maas maas nodes list hostname=node3-control`
-
-    if [ $nodeexist != *node3* ]; then
-        sudo virt-install --connect qemu:///system --name node3-control --ram 8192 --vcpus 4 --disk size=120,format=qcow2,bus=virtio,io=native,pool=default --network bridge=virbr0,model=virtio --network bridge=virbr0,model=virtio --boot network,hd,menu=off --noautoconsole --vnc --print-xml | tee node3-control
-
-        sudo virt-install --connect qemu:///system --name node4-control --ram 8192 --vcpus 4 --disk size=120,format=qcow2,bus=virtio,io=native,pool=default --network bridge=virbr0,model=virtio --network bridge=virbr0,model=virtio --boot network,hd,menu=off --noautoconsole --vnc --print-xml | tee node4-control
-
-        node3controlmac=`grep  "mac address" node3-control | head -1 | cut -d "'" -f 2`
-        node4controlmac=`grep  "mac address" node4-control | head -1 | cut -d "'" -f 2`
-
-        sudo virsh -c qemu:///system define --file node3-control
-        sudo virsh -c qemu:///system define --file node4-control
-
-        controlnodeid=`maas maas nodes new autodetect_nodegroup='yes' name='node3-control' tags='control' hostname='node3-control' power_type='virsh' mac_addresses=$node3controlmac power_parameters_power_address='qemu+ssh://'$USER'@192.168.122.1/system' architecture='amd64/generic' power_parameters_power_id='node3-control' | grep system_id | cut -d '"' -f 4 `
-
-        maas maas tag update-nodes control add=$controlnodeid
-
-        controlnodeid=`maas maas nodes new autodetect_nodegroup='yes' name='node4-control' tags='control' hostname='node4-control' power_type='virsh' mac_addresses=$node4controlmac power_parameters_power_address='qemu+ssh://'$USER'@192.168.122.1/system' architecture='amd64/generic' power_parameters_power_id='node4-control' | grep system_id | cut -d '"' -f 4 `
-
-        maas maas tag update-nodes control add=$controlnodeid
-
-    fi
+    # TODO: make sure this function run with the same parameters used in 03-maasdeploy.sh
+    PROFILE=${PROFILE:-ubuntu}
+    MAAS_IP=$(grep " ip_address" deployconfig.yaml | cut -d ':' -f 2 | sed -e 's/ //')
+    API_SERVER="http://$MAAS_IP/MAAS/api/2.0"
+    API_KEY=`sudo maas-region apikey --username=ubuntu`
+    maas login $PROFILE $API_SERVER $API_KEY
+
+    for node in node3-control node4-control
+    do
+        node_id=$(maas $PROFILE machines read | \
+                  jq -r ".[] | select(.hostname == \"$node\").system_id")
+        if [[ -z "$node_id" ]]; then
+            sudo virt-install --connect qemu:///system --name $node \
+                --ram 8192 --cpu host --vcpus 4 \
+                --disk size=120,format=qcow2,bus=virtio,cache=directsync,io=native,pool=default \
+                --network bridge=virbr0,model=virtio \
+                --network bridge=virbr0,model=virtio \
+                --boot network,hd,menu=off \
+                --noautoconsole --vnc --print-xml | tee _node.xml
+            node_mac=$(grep "mac address" _node.xml | head -1 | cut -d "'" -f 2)
+            sudo virsh -c qemu:///system define --file _node.xml
+            rm -f _node.xml
+
+            maas $PROFILE nodes new autodetect_nodegroup='yes' name=$node \
+                tags='control' hostname=$name power_type='virsh' \
+                mac_addresses=$node3controlmac \
+                power_parameters_power_address="qemu+ssh://$USER@192.168.122.1/system" \
+                architecture='amd64/generic' power_parameters_power_id='node3-control'
+            node_id=$(maas $PROFILE machines read | \
+                  jq -r ".[] | select(.hostname == \"$node\").system_id")
+        fi
+        if [[ -z "$node_id" ]]; then
+            echo "Error: failed to create node $node ."
+            exit 1
+        fi
+        maas $PROFILE tag update-nodes control add=$node_id || true
+    done
 }
 
 #copy the files and create extra resources needed for HA deployment
 # in case of default VM labs.
 deploy() {
-
-    if [ ! -f ./environments.yaml ] && [ -e ~/.juju/environments.yaml ]; then
-        cp ~/.juju/environments.yaml ./environments.yaml
-    fi
-    if [ ! -f ./deployment.yaml ] && [ -e ~/.juju/deployment.yaml ]; then
-        cp ~/.juju/deployment.yaml ./deployment.yaml
-    fi
-    if [ ! -f ./labconfig.yaml ] && [ -e ~/.juju/labconfig.yaml ]; then
-        cp ~/.juju/labconfig.yaml ./labconfig.yaml
+    if [[ "$jujuver" > "2" ]]; then
+        if [ ! -f ./labconfig.yaml ] && [ -e ~/joid_config/labconfig.yaml ]; then
+            cp ~/joid_config/labconfig.yaml ./labconfig.yaml
+
+            if [ ! -f ./deployconfig.yaml ] && [ -e ~/joid_config/deployconfig.yaml ]; then
+                cp ~/joid_config/deployconfig.yaml ./deployconfig.yaml
+            else
+                python genDeploymentConfig.py -l labconfig.yaml > deployconfig.yaml
+            fi
+        else
+            if [ -e ./labconfig.yaml ]; then
+                if [ ! -f ./deployconfig.yaml ] && [ -e ~/joid_config/deployconfig.yaml ]; then
+                    cp ~/joid_config/deployconfig.yaml ./deployconfig.yaml
+                else
+                    python genDeploymentConfig.py -l labconfig.yaml > deployconfig.yaml
+                fi
+            else
+                echo " MAAS not deployed please deploy MAAS first."
+            fi
+        fi
+#create json file which is missing in case of new deployment after maas and git tree cloned freshly.
+
+        python -c 'import sys, yaml, json; json.dump(yaml.load(sys.stdin), sys.stdout, indent=4)' < labconfig.yaml > labconfig.json
+        python -c 'import sys, yaml, json; json.dump(yaml.load(sys.stdin), sys.stdout, indent=4)' < deployconfig.yaml > deployconfig.json
+
+    else
+        if [ ! -f ./environments.yaml ] && [ -e ~/.juju/environments.yaml ]; then
+            cp ~/.juju/environments.yaml ./environments.yaml
+        elif [ ! -f ./environments.yaml ] && [ -e ~/joid_config/environments.yaml ]; then
+            cp ~/joid_config/environments.yaml ./environments.yaml
+        fi
+        #copy the script which needs to get deployed as part of ofnfv release
+        echo "...... deploying now ......"
+        echo "   " >> environments.yaml
+        echo "        enable-os-refresh-update: false" >> environments.yaml
+        echo "        enable-os-upgrade: false" >> environments.yaml
+        echo "        admin-secret: admin" >> environments.yaml
+        echo "        default-series: $opnfvdistro" >> environments.yaml
+        cp environments.yaml ~/.juju/
+        cp environments.yaml ~/joid_config/
     fi
-    if [ ! -f ./deployconfig.yaml ] && [ -e ~/.juju/deployconfig.yaml ]; then
-        cp ~/.juju/deployconfig.yaml ./deployconfig.yaml
-    fi
-
-    #copy the script which needs to get deployed as part of ofnfv release
-    echo "...... deploying now ......"
-    echo "   " >> environments.yaml
-    echo "        enable-os-refresh-update: false" >> environments.yaml
-    echo "        enable-os-upgrade: false" >> environments.yaml
-    echo "        admin-secret: admin" >> environments.yaml
-    echo "        default-series: $opnfvdistro" >> environments.yaml
-
-    cp environments.yaml ~/.juju/
 
     if [[ "$opnfvtype" = "ha" && "$opnfvlab" = "default" ]]; then
         createresource
@@ -137,48 +165,109 @@ deploy() {
     #bootstrap the node
     ./01-bootstrap.sh
 
+    if [[ "$jujuver" > "2" ]]; then
+        juju model-config default-series=$opnfvdistro enable-os-refresh-update=false enable-os-upgrade=false
+    fi
+
     #case default deploy the opnfv platform:
-    ./02-deploybundle.sh $opnfvtype $openstack $opnfvlab $opnfvsdn $opnfvfeature $opnfvdistro
+    ./02-deploybundle.sh $opnfvtype $openstack $opnfvlab $opnfvsdn $opnfvfeature $opnfvdistro $opnfvmodel
 }
 
 #check whether charms are still executing the code even juju-deployer says installed.
 check_status() {
+    waitstatus=$1
     retval=0
     timeoutiter=0
+
+    echo -n "executing the reltionship within charms ."
     while [ $retval -eq 0 ]; do
-       sleep 30
-       juju status > status.txt
-       if [ "$(grep -c "executing" status.txt )" -ge 1 ]; then
-           echo " still executing the reltionship within charms ..."
-           if [ $timeoutiter -ge 60 ]; then
+        if juju status | grep -q $waitstatus; then
+           echo -n '.'
+           if [ $timeoutiter -ge 120 ]; then
+               echo 'timed out'
                retval=1
+           else
+               sleep 30
            fi
            timeoutiter=$((timeoutiter+1))
        else
+           echo 'done'
            retval=1
        fi
     done
 
-    juju expose ceph-radosgw
-    #juju ssh ceph/0 \ 'sudo radosgw-admin user create --uid="ubuntu" --display-name="Ubuntu Ceph"'
+    if [[ "$opnfvmodel" = "openstack" ]]; then
+        juju expose ceph-radosgw || true
+        #juju ssh ceph/0 \ 'sudo radosgw-admin user create --uid="ubuntu" --display-name="Ubuntu Ceph"'
+    fi
 
     echo "...... deployment finishing ......."
-}
+ }
 
 echo "...... deployment started ......"
 deploy
 
-check_status
+check_status executing
 
 echo "...... deployment finished  ......."
 
-./openstack.sh "$opnfvsdn" "$opnfvlab" "$opnfvdistro" "$openstack" || true
 
-# creating heat domain after puching the public API into /etc/hosts
-status=`juju action do heat/0 domain-setup`
-echo $status
+echo "...... configuring public access  ......."
+
+# translate bundle.yaml to json
+python -c 'import sys, yaml, json; json.dump(yaml.load(sys.stdin), sys.stdout, indent=4)' < bundles.yaml > bundles.json
+# get services list having a public interface
+srv_list=$(cat bundles.json | jq -r ".services | to_entries[] | {\"key\": .key, \"value\": .value[\"bindings\"]} | select (.value!=null) | select(.value[] | contains(\"public-api\"))".key)
+# get cnt list from service list
+cnt_list=$(for cnt in $srv_list; do juju status $cnt --format=json | jq -r ".machines[].containers | to_entries[]".key; done)
+# get public network gateway (supposing it is the first ip of the network)
+public_api_gw=$(cat labconfig.json | jq --raw-output ".opnfv.spaces[] | select(.type==\"public\")".gateway)
+admin_gw=$(cat labconfig.json | jq --raw-output ".opnfv.spaces[] | select(.type==\"admin\")".gateway)
+
+if ([ $admin_gw ] && [ $admin_gw != "null" ]); then
+    # set default gateway to public api gateway
+    for cnt in $cnt_list; do
+        echo "changing default gw on $cnt"
+        if ([ $public_api_gw ] && [ $public_api_gw != "null" ]); then
+            juju ssh $cnt "sudo ip r d default && sudo ip r a default via $public_api_gw";
+            juju ssh $cnt "gw_dev=\$(ip  r l | grep 'via $public_api_gw' | cut -d \  -f5) &&\
+                   sudo cp /etc/network/interfaces /etc/network/interfaces.bak &&\
+                   echo 'removing old default gateway' &&\
+                   sudo perl -i -pe 's/^\ *gateway $admin_gw\n$//' /etc/network/interfaces &&\
+                   sudo perl -i -pe \"s/iface \$gw_dev inet static/iface \$gw_dev inet static\\n  gateway $public_api_gw/\" /etc/network/interfaces \
+                   ";
+        fi
+    done
+fi
+
+echo "...... configure  ......."
+
+if ([ $opnfvmodel == "openstack" ]); then
+    ./openstack.sh "$opnfvsdn" "$opnfvlab" "$opnfvdistro" "$openstack" || true
+
+    # creating heat domain after puching the public API into /etc/hosts
+    if [[ "$jujuver" > "2" ]]; then
+        status=`juju run-action heat/0 domain-setup`
+        echo $status
+    else
+        status=`juju action do heat/0 domain-setup`
+        echo $status
+    fi
+
+    sudo ../juju/get-cloud-images || true
+    ../juju/joid-configure-openstack || true
+
+    if grep -q 'openbaton' bundles.yaml; then
+        juju add-relation openbaton keystone
+    fi
+
+elif ([ $opnfvmodel == "kubernetes" ]); then
+    ./k8.sh
+fi
+
+# expose the juju gui-url to login into juju gui
 
-sudo ../juju/get-cloud-images || true
-../juju/joid-configure-openstack || true
+echo " ...... JUJU GUI can be access using the below URL ...... "
+juju gui --show-credentials --no-browser
 
 echo "...... finished  ......."