Calipso initial release for OPNFV

Change-Id: I7210c244b0c10fa80bfa8c77cb86c9d6ddf8bc88
Signed-off-by: Yaron Yogev <yaronyogev@gmail.com>

9 files changed, 536 insertions, 0 deletions

diff --git a/app/monitoring/checks/binary_converter.py b/app/monitoring/checks/binary_converter.py
new file mode 100644
index 0000000..4da1107
--- /dev/null
+++ b/app/monitoring/checks/binary_converter.py
@@ -0,0 +1,17 @@
+###############################################################################
+# Copyright (c) 2017 Koren Lev (Cisco Systems), Yaron Yogev (Cisco Systems)   #
+# 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                                  #
+###############################################################################
+def binary2str(txt):
+    if not isinstance(txt, bytes):
+        return str(txt)
+    try:
+        s = txt.decode("utf-8")
+    except TypeError:
+        s = str(txt)
+    return s
diff --git a/app/monitoring/checks/check_interface.py b/app/monitoring/checks/check_interface.py
new file mode 100755
index 0000000..4140dfe
--- /dev/null
+++ b/app/monitoring/checks/check_interface.py
@@ -0,0 +1,50 @@
+#!/usr/bin/env python3
+###############################################################################
+# Copyright (c) 2017 Koren Lev (Cisco Systems), Yaron Yogev (Cisco Systems)   #
+# 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 re
+import sys
+import subprocess
+
+from binary_converter import binary2str
+
+
+if len(sys.argv) < 2:
+    print('name of interface must be specified')
+    exit(2)
+nic_name = str(sys.argv[1])
+
+rc = 0
+
+try:
+    out = subprocess.check_output(["ifconfig " + nic_name],
+                                  stderr=subprocess.STDOUT,
+                                  shell=True)
+    out = binary2str(out)
+    lines = out.splitlines()
+    line_number = 1
+    line = -1
+    while line_number < len(lines):
+        line = lines[line_number]
+        if ' BROADCAST ' in line:
+            break
+        line_number += 1
+    state_match = re.match('^\W+([A-Z]+)', line)
+    if not state_match:
+        rc = 2
+        print('Error: failed to find status in ifconfig output: ' + out)
+    else:
+        rc = 0 if state_match.group(1) == 'UP' else 2
+        print(out)
+except subprocess.CalledProcessError as e:
+    print("Error finding NIC {}: {}\n".format(nic_name, binary2str(e.output)))
+    rc = 2
+
+exit(rc)
diff --git a/app/monitoring/checks/check_ping.py b/app/monitoring/checks/check_ping.py
new file mode 100755
index 0000000..35e7234
--- /dev/null
+++ b/app/monitoring/checks/check_ping.py
@@ -0,0 +1,121 @@
+#!/usr/bin/env python3
+###############################################################################
+# Copyright (c) 2017 Koren Lev (Cisco Systems), Yaron Yogev (Cisco Systems)   #
+# 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
+import re
+import sys
+import subprocess
+
+from binary_converter import binary2str
+
+
+if len(sys.argv) < 2:
+    raise ValueError('destination address must be specified')
+
+
+def thresholds_string(string):
+    matches = re.match('\d+%/\d+([.]\d+)?/\d+([.]\d+)?', string)
+    if not matches:
+        msg = "%r is not a valid thresholds string" % string
+        raise argparse.ArgumentTypeError(msg)
+    return string
+
+
+def get_args():
+    # try to read scan plan from command line parameters
+    parser = argparse.ArgumentParser()
+
+    parser.add_argument("-W", "--warning", nargs="?",
+                        type=thresholds_string,
+                        default='1%/300/600',
+                        help="warning thresholds: packet-loss "
+                             "(%)/avg-rtt (ms)/max-rtt (ms)"
+                             "(example: 1%/300ms/600ms)")
+    parser.add_argument("-C", "--critical", nargs="?",
+                        type=thresholds_string,
+                        default='10%/1000/2000',
+                        help="critical thresholds: packet-loss "
+                             "(%)/avg-rtt (ms)/max-rtt (ms)"
+                             "(example: 1%/300ms/600ms)")
+    parser.add_argument("-f", "--source", nargs="?", type=str, default='',
+                        help="source address")
+    parser.add_argument("-t", "--target", nargs="?", type=str, default='',
+                        help="target address")
+    parser.add_argument("-c", "--count", nargs="?", type=int, default=5,
+                        help="how many packets will be sent")
+    parser.add_argument("-i", "--interval", nargs="?", type=float, default=0.5,
+                        help="seconds between sending each packet")
+    parser.add_argument("-p", "--pattern", nargs="?", type=str,
+                        default='OS-DNA', help="pattern to pad packet with")
+    parser.add_argument("-w", "--wait", nargs="?", type=int, default=5,
+                        help="seconds to wait for completion of all responses")
+    parser.add_argument("-s", "--packetsize", nargs="?", type=int, default=256,
+                        help="size of packet vseconds to wait for completion "
+                             "of all responses")
+    return parser.parse_args()
+
+args = get_args()
+
+if not args.target:
+    raise ValueError('target address must be specified')
+
+rc = 0
+
+try:
+    cmd = "ping -c {} -i {} -p {} -w {} -s {} {}{} {}".format(
+        args.count, args.interval,
+        args.pattern, args.wait,
+        args.packetsize,
+        '-I ' if args.source else '',
+        args.source, args.target)
+    out = subprocess.check_output([cmd],
+                                  stderr=subprocess.STDOUT,
+                                  shell=True)
+    out = binary2str(out)
+except subprocess.CalledProcessError as e:
+    print("Error doing ping: {}\n".format(binary2str(e.output)))
+
+# find packet loss data
+packet_loss_match = re.search('(\d+)[%] packet loss', out, re.M)
+if not packet_loss_match:
+    out += '\npacket loss data not found'
+    rc = 2
+
+# find rtt avg/max data
+rtt_results = None
+if rc < 2:
+    regexp = 'rtt min/avg/max/mdev = [0-9.]+/([0-9.]+)/([0-9.]+)/[0-9.]+ ms'
+    rtt_results = re.search(regexp, out, re.M)
+    if not rtt_results:
+        out += '\nrtt results not found'
+        rc = 2
+if rc < 2:
+    packet_loss = int(packet_loss_match.group(1))
+    avg_rtt = float(rtt_results.group(1))
+    max_rtt = float(rtt_results.group(2))
+    thresholds_regexp = r'(\d+)%/(\d+[.0-9]*)/(\d+[.0-9]*)'
+    warn_threshold_match = re.match(thresholds_regexp, args.warning)
+    critical_threshold_match = re.match(thresholds_regexp, args.critical)
+    packet_loss_warn = int(warn_threshold_match.group(1))
+    packet_loss_critical = int(critical_threshold_match.group(1))
+    avg_rtt_warn = float(warn_threshold_match.group(2))
+    avg_rtt_critical = float(critical_threshold_match.group(2))
+    max_rtt_warn = float(warn_threshold_match.group(3))
+    max_rtt_critical = float(critical_threshold_match.group(3))
+    if packet_loss > packet_loss_critical or avg_rtt >= avg_rtt_critical or \
+            max_rtt >= max_rtt_critical:
+        rc = 2
+    elif packet_loss > packet_loss_warn or avg_rtt >= avg_rtt_warn or \
+            max_rtt >= max_rtt_warn:
+        rc = 1
+
+print(out)
+exit(rc)
diff --git a/app/monitoring/checks/check_pnic_vpp.py b/app/monitoring/checks/check_pnic_vpp.py
new file mode 100755
index 0000000..942fdc2
--- /dev/null
+++ b/app/monitoring/checks/check_pnic_vpp.py
@@ -0,0 +1,53 @@
+#!/usr/bin/env python3
+###############################################################################
+# Copyright (c) 2017 Koren Lev (Cisco Systems), Yaron Yogev (Cisco Systems)   #
+# 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                                  #
+###############################################################################
+"""
+sudo vppctl show hardware-interfaces:
+
+take only the virtual interfaces, e.g. "VirtualEthernet0/0/0"
+Status: "OK" if "up" is detected in the interface line, CRITICAL otherwise
+
+return full text of "vppctl show hardware-interfaces"
+"""
+
+import re
+import subprocess
+
+from binary_converter import binary2str
+
+
+NAME_RE = '^[a-zA-Z]*GigabitEthernet'
+
+rc = 0
+
+try:
+    out = subprocess.check_output(["sudo vppctl show hardware-interfaces"],
+                                  stderr=subprocess.STDOUT,
+                                  shell=True)
+    out = binary2str(out)
+    lines = out.splitlines()
+    name_re = re.compile(NAME_RE)
+    matching_lines = [l for l in lines if name_re.search(l)]
+    matching_line = matching_lines[0] if matching_lines else None
+    if matching_line:
+        rc = 0 if "up" in matching_line.split() else 2
+        print('output from "vppctl show hardware-interfaces":\n{}'
+              .format(out))
+    else:
+        rc = 2
+        print('Error: failed to find pNic in output of '
+              '"vppctl show hardware-interfaces": {}'
+              .format(out))
+except subprocess.CalledProcessError as e:
+    print("Error running 'vppctl show hardware-interfaces': {}"
+          .format(binary2str(e.output)))
+    rc = 2
+
+exit(rc)
diff --git a/app/monitoring/checks/check_vedge_ovs.py b/app/monitoring/checks/check_vedge_ovs.py
new file mode 100755
index 0000000..849af66
--- /dev/null
+++ b/app/monitoring/checks/check_vedge_ovs.py
@@ -0,0 +1,43 @@
+#!/usr/bin/env python3
+###############################################################################
+# Copyright (c) 2017 Koren Lev (Cisco Systems), Yaron Yogev (Cisco Systems)   #
+# 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                                  #
+###############################################################################
+"""
+Check OVS vEdge health
+
+Run command: 
+ps -aux | grep "\(ovs-vswitchd\|ovsdb-server\)"
+
+OK if for both ovs-vswitchd AND ovsdb-server processes we see '(healthy)'
+otherwise CRITICAL
+
+return full text output of the command
+"""
+
+import subprocess
+
+from binary_converter import binary2str
+
+
+rc = 0
+cmd = 'ps aux | grep "\(ovs-vswitchd\|ovsdb-server\): monitoring" | ' + \
+      'grep -v grep'
+
+try:
+    out = subprocess.check_output([cmd], stderr=subprocess.STDOUT, shell=True)
+    out = binary2str(out)
+    lines = out.splitlines()
+    matching_lines = [l for l in lines if '(healthy)']
+    rc = 0 if len(matching_lines) == 2 else 2
+    print(out)
+except subprocess.CalledProcessError as e:
+    print("Error finding expected output: {}".format(binary2str(e.output)))
+    rc = 2
+
+exit(rc)
diff --git a/app/monitoring/checks/check_vedge_vpp.py b/app/monitoring/checks/check_vedge_vpp.py
new file mode 100755
index 0000000..346feae
--- /dev/null
+++ b/app/monitoring/checks/check_vedge_vpp.py
@@ -0,0 +1,50 @@
+#!/usr/bin/env python3
+###############################################################################
+# Copyright (c) 2017 Koren Lev (Cisco Systems), Yaron Yogev (Cisco Systems)   #
+# 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                                  #
+###############################################################################
+"""
+sudo vppctl show runtime:
+
+test 1: was the return value not null?
+test 2: is startup-config-process = done?
+1 and 2 = vedge status ok
+1 and not 2 = vedge status warning
+not 1 = vedge status critical
+
+return full text of "vppctl show runtime"
+"""
+
+import re
+import subprocess
+
+from binary_converter import binary2str
+
+
+rc = 0
+search_pattern = re.compile("^startup-config-process ")
+
+try:
+    out = subprocess.check_output(["sudo vppctl show runtime"],
+                                  stderr=subprocess.STDOUT,
+                                  shell=True)
+    out = binary2str(out)
+    lines = out.splitlines()
+    matching_lines = [l for l in lines if search_pattern.match(l)]
+    matching_line = matching_lines[0] if matching_lines else None
+    if matching_line and "done" in matching_line.split():
+        print(out)
+    else:
+        rc = 1
+        print('Error: failed to find status in ifconfig output: ' + out)
+except subprocess.CalledProcessError as e:
+    print("Error finding 'vppctl show runtime': {}"
+          .format(binary2str(e.output)))
+    rc = 2
+
+exit(rc)
diff --git a/app/monitoring/checks/check_vnic_vconnector.py b/app/monitoring/checks/check_vnic_vconnector.py
new file mode 100755
index 0000000..b0f96cd
--- /dev/null
+++ b/app/monitoring/checks/check_vnic_vconnector.py
@@ -0,0 +1,72 @@
+#!/usr/bin/env python3
+###############################################################################
+# Copyright (c) 2017 Koren Lev (Cisco Systems), Yaron Yogev (Cisco Systems)   #
+# 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                                  #
+###############################################################################
+
+# find status of vnic-vconnector link
+# vconnector object name defines name of bridge
+# use "brctl showmacs <bridge>", then look for the MAC address
+
+import re
+import sys
+import subprocess
+
+from binary_converter import binary2str
+
+
+if len(sys.argv) < 3:
+    print('usage: ' + sys.argv[0] + ' <bridge> <mac_address>')
+    exit(2)
+bridge_name = str(sys.argv[1])
+mac_address = str(sys.argv[2])
+
+rc = 0
+
+try:
+    out = subprocess.check_output(["brctl showmacs " + bridge_name],
+                                  stderr=subprocess.STDOUT,
+                                  shell=True)
+    out = binary2str(out)
+    lines = out.splitlines()
+    line_number = 1
+    line = ''
+    found = False
+    while line_number < len(lines):
+        line = lines[line_number]
+        if mac_address in line:
+            found = True
+            break
+        line_number += 1
+    state_match = re.match('^\W+([A-Z]+)', line)
+    if not found:
+        rc = 2
+        print('Error: failed to find MAC {}:\n{}\n'
+              .format(mac_address, out))
+    else:
+        # grab "is local?" and "ageing timer" values
+        line_parts = line.split()  # port, mac address, is local?, ageing timer
+        is_local = line_parts[2]
+        ageing_timer = line_parts[3]
+        msg_format =\
+            'vConnector bridge name: {}\n'\
+            'vNIC MAC address: {}\n'\
+            'is local: {}\n'\
+            'ageing timer: {}\n'\
+            'vNIC MAC address: {}\n'\
+            'command: brctl showmacs {}\n'\
+            'output:\n{}'
+        msg = msg_format.format(bridge_name, mac_address, is_local,
+                                ageing_timer, mac_address, bridge_name, out)
+        print(msg)
+except subprocess.CalledProcessError as e:
+    print("Error finding MAC {}: {}\n"
+          .format(mac_address, binary2str(e.output)))
+    rc = 2
+
+exit(rc)
diff --git a/app/monitoring/checks/check_vnic_vpp.py b/app/monitoring/checks/check_vnic_vpp.py
new file mode 100755
index 0000000..0f77ddd
--- /dev/null
+++ b/app/monitoring/checks/check_vnic_vpp.py
@@ -0,0 +1,48 @@
+#!/usr/bin/env python3
+###############################################################################
+# Copyright (c) 2017 Koren Lev (Cisco Systems), Yaron Yogev (Cisco Systems)   #
+# 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                                  #
+###############################################################################
+"""
+sudo vppctl show hardware-interfaces:
+
+take only the virtual interfaces, e.g. "VirtualEthernet0/0/0"
+Status: "OK" if "up" is detected in the interface line, CRITICAL otherwise
+
+return full text of "vppctl show hardware-interfaces"
+"""
+
+import re
+import subprocess
+
+from binary_converter import binary2str
+
+rc = 0
+search_pattern = re.compile("^Virtual")
+
+try:
+    out = subprocess.check_output(["sudo vppctl show hardware-interfaces"],
+                                  stderr=subprocess.STDOUT,
+                                  shell=True)
+    out = binary2str(out)
+    lines = out.splitlines()
+    matching_lines = [l for l in lines if search_pattern.match(l)]
+    matching_line = matching_lines[0] if matching_lines else None
+    if matching_line and "up" in matching_line.split():
+        print('output of "vppctl show hardware-interfaces":\n{}'
+              .format(out))
+    else:
+        rc = 2
+        print('Error: failed to find status in output of '
+              '"vppctl show hardware-interfaces": {}'.format(out))
+except subprocess.CalledProcessError as e:
+    print("Error finding 'vppctl show hardware-interfaces': {}"
+          .format(binary2str(e.output)))
+    rc = 2
+
+exit(rc)
diff --git a/app/monitoring/checks/check_vservice.py b/app/monitoring/checks/check_vservice.py
new file mode 100644
index 0000000..a95a46a
--- /dev/null
+++ b/app/monitoring/checks/check_vservice.py
@@ -0,0 +1,82 @@
+#!/usr/bin/env python3
+###############################################################################
+# Copyright (c) 2017 Koren Lev (Cisco Systems), Yaron Yogev (Cisco Systems)   #
+# 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                                  #
+###############################################################################
+
+"""
+for vservice with type T and id X
+run on the corresponding host:
+ip netns pid X
+response is pid(s), for example:
+32075
+
+For DHCP there are multiple pid, we will take the dnsmasq process
+
+then run :
+ps -uf -p 32075
+
+get STAT - "S" and "R" = OK
+"""
+
+import subprocess
+import sys
+
+from binary_converter import binary2str
+
+
+rc = 0
+
+args = sys.argv
+if len(args) < 3:
+    print('usage: check_vservice.py <vService type> <vService ID>')
+    exit(2)
+
+vservice_type = args[1]
+vservice_id = args[2]
+netns_cmd = 'sudo ip netns pid {}'.format(vservice_id)
+pid = ''
+ps_cmd = ''
+try:
+    out = subprocess.check_output([netns_cmd], stderr=subprocess.STDOUT,
+                                  shell=True)
+    out = binary2str(out)
+    lines = out.splitlines()
+    if not lines:
+        print('no matching vservice: {}\ncommand: {}\noutput: {}'
+              .format(vservice_id, netns_cmd, out))
+        exit(2)
+    pid = lines[0]
+except subprocess.CalledProcessError as e:
+    print("Error running '{}': {}"
+          .format(netns_cmd, binary2str(e.output)))
+    exit(2)
+try:
+    ps_cmd = 'ps -uf -p {}'.format(pid)
+    out = subprocess.check_output([ps_cmd], stderr=subprocess.STDOUT,
+                                  shell=True)
+    ps_out = binary2str(out)
+    lines = ps_out.splitlines()
+    if not lines:
+        print('no matching vservice: {}\noutput of {}:\n{}'
+              .format(vservice_id, netns_cmd, out))
+        exit(2)
+    headers = lines[0].split()
+    lines = lines[1:]
+    if vservice_type == 'dhcp' and len(lines) > 1:
+        lines = [line for line in lines if 'dnsmasq' in line]
+    values = lines[0].split()
+    stat_index = headers.index('STAT')
+    status = values[stat_index]
+    rc = 0 if status in ['S', 'R'] else 2
+    print('{}\n{}\n{}'.format(netns_cmd, ps_cmd, ps_out))
+except subprocess.CalledProcessError as e:
+    print("Error running '{}': {}".format(ps_cmd, binary2str(e.output)))
+    rc = 2
+
+exit(rc)