Add Collectd as a Monitor Type

A plugin for collectd is added to use collectd on compute as a
monitor type. Monitor files are updated accordingly.
The inspector now listens on all interfaces instead of only
localhost to enable it to communicate with compute node.

JIRA: DOCTOR-86
JIRA: DOCTOR-101

Change-Id: Idc834d428152e4687020eff7d8db36a652b1bf86
Signed-off-by: Umar Farooq <umar.farooq@neclab.eu>

4 files changed, 410 insertions, 0 deletions

diff --git a/tests/lib/monitors/collectd/collectd b/tests/lib/monitors/collectd/collectd
new file mode 100644
index 00000000..f5096658
--- /dev/null
+++ b/tests/lib/monitors/collectd/collectd
@@ -0,0 +1,101 @@
+#!/bin/bash
+
+function start_monitor_collectd {
+    ## CONTROL_IP is the IP of primary interface of control node i.e.
+    ## eth0, eno1. It is used by collectd monitor to communicate with
+    ## sample inspector.
+    ## @TODO (umar) see if mgmt IP of control is a better option. Also
+    ## primary interface may not be the right option
+    CONTROL_IP="$(ip a | sed -En 's/127.0.0.1//;s/.*inet (addr:)?(([0-9]*\.){3}[0-9]*).*/\2/p' | sed -n 1p)"
+    #CONTROL_IP=192.168.98.6
+
+    echo "
+Hostname \"$COMPUTE_HOST\"
+FQDNLookup false
+Interval 1
+MaxReadInterval 2
+
+<LoadPlugin python>
+    Globals true
+</LoadPlugin>
+LoadPlugin ovs_events
+LoadPlugin logfile
+
+<Plugin logfile>
+  File \"/var/log/collectd.log\"
+  Timestamp true
+  LogLevel \"info\"
+</Plugin>
+
+<Plugin python>
+    ModulePath \"/home/$COMPUTE_USER\"
+    LogTraces true
+    Interactive false
+    Import \"collectd_plugin\"
+    <Module \"collectd_plugin\">
+        control_ip \"$CONTROL_IP\"
+        compute_ip \"$COMPUTE_IP\"
+        compute_host \"$COMPUTE_HOST\"
+        compute_user \"$COMPUTE_USER\"
+        inspector_type \"$INSPECTOR_TYPE\"
+        os_auth_url \"$OS_AUTH_URL\"
+        os_username \"$OS_USERNAME\"
+        os_password \"$OS_PASSWORD\"
+        os_project_name \"$OS_PROJECT_NAME\"
+        os_user_domain_name \"$OS_USER_DOMAIN_NAME\"
+        os_user_domain_id \"$OS_USER_DOMAIN_ID\"
+        os_project_domain_name \"$OS_PROJECT_DOMAIN_NAME\"
+        os_project_domain_id \"$OS_PROJECT_DOMAIN_ID\"
+    </Module>
+</Plugin>
+
+<Plugin ovs_events>
+    Port 6640
+    Socket \"/var/run/openvswitch/db.sock\"
+    Interfaces \"@INTERFACE_NAME@\"
+    SendNotification true
+    DispatchValues false
+</Plugin>
+
+" > $TOP_DIR/lib/monitors/collectd.conf
+
+    scp $ssh_opts_cpu $TOP_DIR/lib/monitors/collectd.conf $COMPUTE_USER@$COMPUTE_IP:
+    ## @TODO (umar) Always assuming that the interface is assigned an IP if
+    ## interface name is not provided. See if there is a better approach
+    ssh $ssh_opts_cpu "$COMPUTE_USER@$COMPUTE_IP" "
+        if [ -n \"$INTERFACE_NAME\" ]; then
+            dev=$INTERFACE_NAME
+        else
+            dev=\$(sudo ip a | awk '/ $COMPUTE_IP\//{print \$NF}')
+        fi
+        sed -i -e \"s/@INTERFACE_NAME@/\$dev/\" collectd.conf
+        collectd_conf=/opt/collectd/etc/collectd.conf
+        if [ -e \$collectd_conf ]; then
+            sudo cp \$collectd_conf \${collectd_conf}-doctor-saved
+        else
+            sudo touch \${collectd_conf}-doctor-created
+        fi
+        sudo mv collectd.conf /opt/collectd/etc/collectd.conf"
+
+    scp $ssh_opts_cpu $TOP_DIR/lib/monitors/collectd/collectd_plugin.py $COMPUTE_USER@$COMPUTE_IP:collectd_plugin.py
+    ssh $ssh_opts_cpu "$COMPUTE_USER@$COMPUTE_IP" "sudo pkill collectd
+                                                   sudo /opt/collectd/sbin/collectd"
+}
+
+function stop_monitor_collectd {
+    ssh $ssh_opts_cpu "$COMPUTE_USER@$COMPUTE_IP" 'sudo pkill collectd'
+}
+
+function cleanup_monitor_collectd {
+    ssh $ssh_opts_cpu "$COMPUTE_USER@$COMPUTE_IP" "
+        collectd_conf=/opt/collectd/etc/collectd.conf
+        if [ -e \"\${collectd_conf}-doctor-created\" ]; then
+            sudo rm \"\${collectd_conf}-doctor-created\"
+            sudo rm \$collectd_conf
+        elif [ -e \"\${collectd_conf}-doctor-saved\" ]; then
+            sudo cp -f \"\${collectd_conf}-doctor-saved\" \$collectd_conf
+            sudo rm \"\${collectd_conf}-doctor-saved\"
+        fi"
+
+    rm $TOP_DIR/lib/monitors/collectd.conf
+}
diff --git a/tests/lib/monitors/collectd/collectd_plugin.py b/tests/lib/monitors/collectd/collectd_plugin.py
new file mode 100644
index 00000000..70fcf26e
--- /dev/null
+++ b/tests/lib/monitors/collectd/collectd_plugin.py
@@ -0,0 +1,167 @@
+##############################################################################
+# Copyright (c) 2017 NEC Corporation and others.
+#
+# All rights reserved. This program and the accompanying materials
+# are made available under the terms of the Apache License, Version 2.0
+# which accompanies this distribution, and is available at
+# http://www.apache.org/licenses/LICENSE-2.0
+##############################################################################
+
+import collectd
+import sys
+from netifaces import interfaces, ifaddresses, AF_INET
+from datetime import datetime
+import json
+import requests
+import time
+from requests.exceptions import ConnectionError
+
+from keystoneauth1 import loading
+from keystoneauth1 import session
+from congressclient.v1 import client
+
+
+def write_debug(str_write, write_type, compute_user):
+    file_name = ('/home/%s/monitor.log' % compute_user)
+    file_tmp = open(file_name, write_type)
+    file_tmp.write( "%s" % str_write)
+    file_tmp.close()
+
+
+class DoctorMonitorCollectd(object):
+    def __init__(self):
+        self.control_ip = ''
+        self.compute_user = ''
+        self.compute_ip = ''
+        self.host_name = ''
+        self.inspector_type = ''
+        self.inspector_url = ''
+        self.os_auth_url = ''
+        self.os_username = ''
+        self.os_password = ''
+        self.os_project_name = ''
+        self.os_user_domain_name = ''
+        self.os_user_domain_id = ''
+        self.os_project_domain_name = ''
+        self.os_project_domain_id = ''
+        self.sess = ''
+        self.auth = ''
+        self.inspector_notified = 0
+        self.start_notifications = 0
+        self.monitor_type = 'sample'
+
+    def config_func(self, config):
+        for node in config.children:
+            key = node.key.lower()
+            val = node.values[0]
+
+            if key == 'compute_host':
+                self.host_name = val
+            elif key == 'control_ip':
+                self.control_ip = val
+            elif key == 'compute_ip':
+                self.compute_ip = val
+            elif key == 'compute_user':
+                self.compute_user = val
+            elif key == 'inspector_type':
+                self.inspector_type = val
+            elif key == 'os_auth_url':
+                self.os_auth_url = val
+            elif key == 'os_username':
+                self.os_username = val
+            elif key == 'os_password':
+                self.os_password = val
+            elif key == 'os_project_name':
+                self.os_project_name = val
+            elif key == 'os_user_domain_name':
+                self.os_user_domain_name = val
+            elif key == 'os_user_domain_id':
+                self.os_user_domain_id = val
+            elif key == 'os_project_domain_name':
+                self.os_project_domain_name = val
+            elif key == 'os_project_domain_id':
+                self.os_project_domain_id = val
+            else:
+                collectd.info('Unknown config key "%s"' % key)
+
+    def init_collectd(self):
+        write_debug("Compute node collectd monitor start at %s\n\n" % datetime.now().isoformat(), "w", self.compute_user)
+
+        if self.inspector_type == 'sample':
+            self.inspector_url = ('http://%s:12345/events' % self.control_ip)
+        elif self.inspector_type == 'congress':
+            loader = loading.get_plugin_loader('password')
+            self.auth = loader.load_from_options(auth_url=self.os_auth_url,
+                        username=self.os_username,
+                        password=self.os_password,
+                        project_name=self.os_project_name,
+                        user_domain_name=self.os_user_domain_name,
+                        user_domain_id=self.os_user_domain_id,
+                        project_domain_name=self.os_project_domain_name,
+                        project_domain_id=self.os_project_domain_id)
+            self.sess=session.Session(auth=self.auth)
+            congress = client.Client(session=self.sess, service_type='policy')
+            ds = congress.list_datasources()['results']
+            doctor_ds = next((item for item in ds if item['driver'] == 'doctor'),
+                         None)
+
+            congress_endpoint = congress.httpclient.get_endpoint(auth=self.auth)
+            self.inspector_url = ('%s/v1/data-sources/%s/tables/events/rows' %
+                              (congress_endpoint, doctor_ds['id']))
+        else:
+            sys.exit()
+        self.start_notifications = 1
+
+
+    def notify_inspector(self):
+        event_type = "compute.host.down"
+        payload = [
+            {
+                 'id': ("monitor_%s_id1" % self.monitor_type),
+                 'time': datetime.now().isoformat(),
+                 'type': event_type,
+                 'details': {
+                     'hostname': self.host_name,
+                     'status': 'down',
+                     'monitor': ("monitor_%s" % self.monitor_type),
+                     'monitor_event_id': ("monitor_%s_event1" % self.monitor_type)
+                 },
+             },
+        ]
+        data = json.dumps(payload)
+        self.inspector_notified = 1
+
+        if self.inspector_type == 'sample':
+            headers = {'content-type': 'application/json'}
+            try:
+                requests.post(self.inspector_url, data=data, headers=headers)
+            except ConnectionError as err:
+                print err
+        elif self.inspector_type == 'congress':
+            # TODO(umar) enhance for token expiry case
+            headers = {
+                'Content-Type': 'application/json',
+                'Accept': 'application/json',
+                'X-Auth-Token': self.sess.get_token()
+            }
+            requests.put(self.inspector_url, data=data, headers=headers)
+
+
+    def handle_notif(self, notification, data=None):
+        if (notification.severity == collectd.NOTIF_FAILURE or
+            notification.severity == collectd.NOTIF_WARNING):
+            if (self.start_notifications == 1 and self.inspector_notified == 0):
+                write_debug("Received down notification: doctor monitor detected at %s\n" % time.time(), "a", self.compute_user)
+                self.notify_inspector()
+
+        elif notification.severity == collectd.NOTIF_OKAY:
+            collectd.info("Interface status: UP again %s\n" % time.time())
+        else:
+            collectd.info("Unknown notification severity %s\n" % notification.severity)
+
+
+monitor = DoctorMonitorCollectd()
+
+collectd.register_config(monitor.config_func)
+collectd.register_init(monitor.init_collectd)
+collectd.register_notification(monitor.handle_notif)
diff --git a/tests/lib/monitors/sample/monitor.py b/tests/lib/monitors/sample/monitor.py
new file mode 100644
index 00000000..7450c534
--- /dev/null
+++ b/tests/lib/monitors/sample/monitor.py
@@ -0,0 +1,124 @@
+##############################################################################
+# Copyright (c) 2016 NEC Corporation and others.
+#
+# All rights reserved. This program and the accompanying materials
+# are made available under the terms of the Apache License, Version 2.0
+# which accompanies this distribution, and is available at
+# http://www.apache.org/licenses/LICENSE-2.0
+##############################################################################
+
+import argparse
+from datetime import datetime
+import json
+import logger as doctor_log
+import requests
+import socket
+import time
+
+from keystoneauth1 import session
+from congressclient.v1 import client
+
+import identity_auth
+
+# NOTE: icmp message with all zero data (checksum = 0xf7ff)
+#       see https://tools.ietf.org/html/rfc792
+ICMP_ECHO_MESSAGE = '\x08\x00\xf7\xff\x00\x00\x00\x00'
+
+SUPPORTED_INSPECTOR_TYPES = ['sample', 'congress']
+
+LOG = doctor_log.Logger('doctor_monitor').getLogger()
+
+
+class DoctorMonitorSample(object):
+
+    interval = 0.1  # second
+    timeout = 0.1  # second
+    event_type = "compute.host.down"
+
+    def __init__(self, args):
+        if args.inspector_type not in SUPPORTED_INSPECTOR_TYPES:
+            raise Exception("Inspector type '%s' not supported", args.inspector_type)
+
+        self.hostname = args.hostname
+        self.inspector_type = args.inspector_type
+        self.ip_addr = args.ip or socket.gethostbyname(self.hostname)
+
+        if self.inspector_type == 'sample':
+            self.inspector_url = 'http://127.0.0.1:12345/events'
+        elif self.inspector_type == 'congress':
+            auth=identity_auth.get_identity_auth()
+            self.session=session.Session(auth=auth)
+            congress = client.Client(session=self.session, service_type='policy')
+            ds = congress.list_datasources()['results']
+            doctor_ds = next((item for item in ds if item['driver'] == 'doctor'),
+                             None)
+
+            congress_endpoint = congress.httpclient.get_endpoint(auth=auth)
+            self.inspector_url = ('%s/v1/data-sources/%s/tables/events/rows' %
+                                  (congress_endpoint, doctor_ds['id']))
+
+    def start_loop(self):
+        LOG.debug("start ping to host %(h)s (ip=%(i)s)" % {'h': self.hostname,
+                                                       'i': self.ip_addr})
+        sock = socket.socket(socket.AF_INET, socket.SOCK_RAW,
+                             socket.IPPROTO_ICMP)
+        sock.settimeout(self.timeout)
+        while True:
+            try:
+                sock.sendto(ICMP_ECHO_MESSAGE, (self.ip_addr, 0))
+                data = sock.recv(4096)
+            except socket.timeout:
+                LOG.info("doctor monitor detected at %s" % time.time())
+                self.report_error()
+                LOG.info("ping timeout, quit monitoring...")
+                return
+            time.sleep(self.interval)
+
+    def report_error(self):
+        payload = [
+            {
+                'id': 'monitor_sample_id1',
+                'time': datetime.now().isoformat(),
+                'type': self.event_type,
+                'details': {
+                    'hostname': self.hostname,
+                    'status': 'down',
+                    'monitor': 'monitor_sample',
+                    'monitor_event_id': 'monitor_sample_event1'
+                },
+            },
+        ]
+        data = json.dumps(payload)
+
+        if self.inspector_type == 'sample':
+            headers = {'content-type': 'application/json'}
+            requests.post(self.inspector_url, data=data, headers=headers)
+        elif self.inspector_type == 'congress':
+            headers = {
+                'Content-Type': 'application/json',
+                'Accept': 'application/json',
+                'X-Auth-Token':self.session.get_token(),
+            }
+            requests.put(self.inspector_url, data=data, headers=headers)
+
+
+def get_args():
+    parser = argparse.ArgumentParser(description='Doctor Sample Monitor')
+    parser.add_argument('hostname', metavar='HOSTNAME', type=str, nargs='?',
+                        help='a hostname to monitor connectivity')
+    parser.add_argument('ip', metavar='IP', type=str, nargs='?',
+                        help='an IP address to monitor connectivity')
+    parser.add_argument('inspector_type', metavar='INSPECTOR_TYPE', type=str, nargs='?',
+                        help='inspector to report',
+                        default='sample')
+    return parser.parse_args()
+
+
+def main():
+    args = get_args()
+    monitor = DoctorMonitorSample(args)
+    monitor.start_loop()
+
+
+if __name__ == '__main__':
+    main()
diff --git a/tests/lib/monitors/sample/sample b/tests/lib/monitors/sample/sample
new file mode 100644
index 00000000..1d310333
--- /dev/null
+++ b/tests/lib/monitors/sample/sample
@@ -0,0 +1,18 @@
+#!/bin/bash
+
+function start_monitor_sample {
+    cp $TOP_DIR/lib/monitors/sample/monitor.py $TOP_DIR/monitor.py
+    pgrep -f "python monitor.py" && return 0
+    sudo -E python monitor.py "$COMPUTE_HOST" "$COMPUTE_IP" "$INSPECTOR_TYPE" \
+        > monitor.log 2>&1 &
+}
+
+function stop_monitor_sample {
+    pgrep -f "python monitor.py" || return 0
+    sudo kill $(pgrep -f "python monitor.py")
+}
+
+function cleanup_monitor_sample {
+    rm monitor.py
+    return
+}