Add option to exclude alarms and usage elements

NETWAYS · Aug 21, 2023 · fae7be6 · fae7be6
1 parent ad21861
commit fae7be6
Show file tree

Hide file tree

Showing 3 changed files with 123 additions and 14 deletions.
diff --git a/README.md b/README.md
@@ -16,7 +16,7 @@ Please prefer installation via system packages like `python3-requests`.
 
 Alternatively you can install with pip:
 
-    pip3 install requests
+    pip3 install -r requirements.txt
 
 Make sure to modify the shebang to your environment, one of the following should be fine.
 
@@ -43,12 +43,23 @@ optional arguments:
   --password PASSWORD, -p PASSWORD
                         Password for Basic Auth
   --mode MODE, -m MODE  Check mode
+  --exclude [EXCLUDE ...]
+                        Exclude alarms or usage from the check results. Can be used multiple times and supports regular expressions.
   --max-age MAX_AGE, -M MAX_AGE
                         Max age in minutes for capacity usage updates. Defaults to 5
   --version, -V         Print version
   --insecure            Do not verify TLS certificate. Be careful with this option, please
 ```
 
+The `--exclude` parameter will match against alarms and capacity-usage. It uses the following string representation (whitespaces included) to match against:
+
+* alarms: `severity` `node_display_name` `feature_display_name` `event_type_display_name`
+* capacity-usage: `severity` `display_name`
+
+## Examples
+
+Mode: cluster-status
+
 ```
 $ ./check_vmware_nsxt.py --api 'https://vmware-nsx.local' -u icinga -p password --mode cluster-status
 [OK] control_cluster_status=STABLE - mgmt_cluster_status=STABLE - control_cluster_status=STABLE - nodes_online=3
@@ -66,14 +77,25 @@ $ ./check_vmware_nsxt.py --api 'https://vmware-nsx.local' -u icinga -p password
 | nodes_online=3;;;0
 ```
 
+Mode: alarms
+
 ```
 $ ./check_vmware_nsxt.py --api 'https://vmware-nsx.local' -u icinga -p password --mode alarms
 [WARNING] 1 alarms - 1 medium
 
 [MEDIUM] (2021-04-26 17:25:18) (node1) Intelligence Health/Storage Latency High - Intelligence node storage latency is high.
+| alarms=1;;;0 alarms.medium=1;;;0
+```
+
+```
+$ ./check_vmware_nsxt.py --api 'https://vmware-nsx.local' -u icinga -p password --mode alarms --exclude "LOW"
+# Excluded alerts will still be counted, but are not factored into the exit code
+[OK] 1 alarms
 | alarms=1;;;0
 ```
 
+Mode: capacity-usage
+
 ```
 $ ./check_vmware_nsxt.py --api 'https://vmware-nsx.local' -u icinga -p password --mode capacity-usage
 [OK] 28 info - no usages - last update: 2021-04-29 19:06:12

diff --git a/check_vmware_nsxt.py b/check_vmware_nsxt.py
@@ -37,6 +37,7 @@
 import logging
 import datetime
 import ssl
+import re
 from urllib.parse import urljoin
 import urllib3
 import requests
@@ -127,26 +128,26 @@ def request(self, url, method='GET'):
         except Exception as json_exc:
             raise CriticalException('Could not decode API JSON: ' + str(json_exc)) # pylint: disable=raise-missing-from
 
-    def get_cluster_status(self):
+    def get_cluster_status(self, excludes=None):
         """
         GET and build ClusterStatus
         """
-        return ClusterStatus(self.request('cluster/status'))
+        return ClusterStatus(self.request('cluster/status'), excludes)
 
-    def get_alarms(self):
+    def get_alarms(self, excludes=None):
         """
         GET and build Alarms
         """
         status = "OPEN"
         # status = "RESOLVED" # for testing
         result = self.request('alarms?page_size=100&status=%s&sort_ascending=false' % status)
-        return Alarms(result['results'])
+        return Alarms(data=result['results'], excludes=excludes)
 
-    def get_capacity_usage(self):
+    def get_capacity_usage(self, excludes=None):
         """
         GET and build CapacityUsage
         """
-        return CapacityUsage(self.request('capacity/usage'), self.max_age)
+        return CapacityUsage(self.request('capacity/usage'), self.max_age, excludes)
 
 
 class CheckResult:
@@ -203,9 +204,12 @@ class ClusterStatus(CheckResult):
     https://vdc-download.vmware.com/vmwb-repository/dcr-public/787988e9-6348-4b2a-8617-e6d672c690ee/a187360c-77d5-4c0c-92a8-8e07aa161a27/api_includes/method_ReadClusterStatus.html
     """
 
-    def __init__(self, data):
+    def __init__(self, data, excludes):
         super().__init__()
         self.data = data
+        self.excludes = excludes
+        if excludes is None:
+            self.excludes = []
 
     def build_output(self):
         for area in ['control_cluster_status', 'mgmt_cluster_status', 'control_cluster_status']:
@@ -234,14 +238,33 @@ class Alarms(CheckResult):
     https://vdc-download.vmware.com/vmwb-repository/dcr-public/787988e9-6348-4b2a-8617-e6d672c690ee/a187360c-77d5-4c0c-92a8-8e07aa161a27/api_includes/method_GetAlarms.html
     """
 
-    def __init__(self, data):
+    def __init__(self, data, excludes):
         super().__init__()
         self.data = data
+        self.excludes = excludes
+        if excludes is None:
+            self.excludes = []
+
+    def _is_excluded(self, alarm):
+        # to exclude via --exclude
+        identifier = "%s %s %s %s" % (
+            alarm['severity'],
+            alarm['node_display_name'],
+            alarm['feature_display_name'],
+            alarm['event_type_display_name'])
+        for exclude in self.excludes:
+            regexp = re.compile(exclude)
+            if bool(regexp.search(identifier)):
+                return True
+        return False
 
     def build_output(self):
         states = {}
 
         for alarm in self.data:
+            if self._is_excluded(alarm):
+                continue
+
             severity = alarm['severity']
             if severity in states:
                 states[severity] += 1
@@ -270,7 +293,11 @@ def build_status(self):
         states = []
 
         for alarm in self.data:
-            state = WARNING if alarm['severity'] in ['MEDIUM', 'LOW'] else CRITICAL  # CRITICAL, HIGH
+            if self._is_excluded(alarm):
+                continue
+
+            # HIGH == CRITICAL
+            state = WARNING if alarm['severity'] in ['MEDIUM', 'LOW'] else CRITICAL
             states.append(state)
 
         if len(states) > 0:
@@ -285,15 +312,33 @@ class CapacityUsage(CheckResult):
     https://vdc-download.vmware.com/vmwb-repository/dcr-public/787988e9-6348-4b2a-8617-e6d672c690ee/a187360c-77d5-4c0c-92a8-8e07aa161a27/api_includes/method_GetProtonCapacityUsage.html
     """
 
-    def __init__(self, data, max_age):
+    def __init__(self, data, max_age, excludes):
         super().__init__()
         self.data = data
         self.max_age = max_age
+        self.excludes = excludes
+        if excludes is None:
+            self.excludes = []
+
+    def _is_excluded(self, usage):
+        # to exclude via --exclude
+        identifier = "%s %s" % (
+            usage['severity'],
+            usage['display_name'])
+
+        for exclude in self.excludes:
+            regexp = re.compile(exclude)
+            if bool(regexp.search(identifier)):
+                return True
+        return False
 
     def build_output(self):
         states = {}
 
         for usage in self.data['capacity_usage']:
+            if self._is_excluded(usage):
+                continue
+
             severity = usage['severity']  # INFO, WARNING, CRITICAL, ERROR
 
             if severity in states:
@@ -341,6 +386,9 @@ def build_status(self):
             self.summary.append("last update older than %s minutes" % (self.max_age))
 
         for usage in self.data['capacity_usage']:
+            if self._is_excluded(usage):
+                continue
+
             severity = usage['severity']  # INFO, WARNING, CRITICAL, ERROR
 
             if severity == "INFO":
@@ -398,6 +446,8 @@ def commandline(args):
                         help='Password for Basic Auth', required=True)
     parser.add_argument('--mode', '-m', choices=['cluster-status', 'alarms', 'capacity-usage'],
                         help='Check mode to exectue. Hint: alarms will only include open alarms.', required=True)
+    parser.add_argument('--exclude', nargs='*', action='extend', type=str,
+                        help="Exclude alarms or usage from the check results. Can be used multiple times and supports regular expressions.")
     parser.add_argument('--max-age', '-M', type=int,
                         help='Max age in minutes for capacity usage updates. Defaults to 5', default=5, required=False)
     parser.add_argument('--insecure',
@@ -421,11 +471,11 @@ def main(args):
     client = Client(args.api, args.username, args.password, verify=(not args.insecure), max_age=args.max_age)
 
     if args.mode == 'cluster-status':
-        return client.get_cluster_status().print_and_return()
+        return client.get_cluster_status(args.exclude).print_and_return()
     if args.mode == 'alarms':
-        return client.get_alarms().print_and_return()
+        return client.get_alarms(args.exclude).print_and_return()
     if args.mode == 'capacity-usage':
-        return client.get_capacity_usage().print_and_return()
+        return client.get_capacity_usage(args.exclude).print_and_return()
 
     print("[UNKNOWN] unknown mode %s" % args.mode)
     return UNKNOWN

diff --git a/test_check_vmware_nsxt.py b/test_check_vmware_nsxt.py
@@ -153,6 +153,26 @@ def test_alarms_ok(self, mock_req, mock_print):
         self.assertEqual(actual, expected)
         mock_print.assert_called_with('[WARNING] 1 alarms - 1 medium\n\n[MEDIUM] (2021-04-26 15:25:18) (node1) Intelligence Health/Storage Latency High - Intelligence node storage latency is high.\n| alarms=1;;;0 alarms.medium=1;;;0')
 
+    @mock.patch('builtins.print')
+    @mock.patch('requests.request')
+    def test_alarms_exclude(self, mock_req, mock_print):
+
+        with open('testdata/fixtures/alarms.json') as f:
+            testdata = json.load(f)
+
+        m = mock.MagicMock()
+        m.status_code = 200
+        m.json.return_value = testdata
+        mock_req.return_value = m
+
+        c = Client('api', 'username', 'password', logger=None, verify=True, max_age=5)
+
+        actual = c.get_alarms(excludes=["M[A-Z]+M"]).print_and_return()
+        expected = 0
+
+        self.assertEqual(actual, expected)
+        mock_print.assert_called_with('[OK] 1 alarms\n| alarms=1;;;0')
+
     @mock.patch('builtins.print')
     @mock.patch('requests.request')
     def test_capacity_usage_ok(self, mock_req, mock_print):
@@ -172,3 +192,20 @@ def test_capacity_usage_ok(self, mock_req, mock_print):
 
         self.assertEqual(actual, expected)
         mock_print.assert_called_with('[WARNING] 28 info - last update: 2021-04-30 09:17:40 - last update older than 5 minutes\n\n[OK] [INFO] System-wide NAT rules: 0 of 25000 (0%)\n[OK] [INFO] Network Introspection Rules: 1 of 10000 (0.01%)\n[OK] [INFO] System-wide Endpoint Protection Enabled Hosts: 0 of 256 (0%)\n[OK] [INFO] Hypervisor Hosts: 18 of 1024 (1.75%)\n[OK] [INFO] System-wide Firewall Rules: 81 of 100000 (0.08%)\n[OK] [INFO] System-wide DHCP Pools: 0 of 10000 (0%)\n[OK] [INFO] System-wide Edge Nodes: 10 of 320 (3.12%)\n[OK] [INFO] Active Directory Domains (Identity Firewall): 0 of 4 (0%)\n[OK] [INFO] vSphere Clusters Prepared for NSX: 4 of 128 (3.12%)\n[OK] [INFO] Prefix-lists: 20 of 500 (4%)\n[OK] [INFO] Logical Switches: 12 of 10000 (0.12%)\n[OK] [INFO] System-wide Logical Switch Ports: 145 of 25000 (0.58%)\n[OK] [INFO] Active Directory Groups (Identity Firewall): 0 of 100000 (0%)\n[OK] [INFO] Distributed Firewall Rules: 75 of 100000 (0.07%)\n[OK] [INFO] System-wide Endpoint Protection Enabled Virtual Machines: 0 of 7500 (0%)\n[OK] [INFO] Distributed Firewall Sections: 23 of 10000 (0.23%)\n[OK] [INFO] Groups Based on IP Sets: 37 of 10000 (0.37%)\n[OK] [INFO] Edge Clusters: 3 of 160 (1.87%)\n[OK] [INFO] Tier-1 Logical Routers with NAT Enabled: 0 of 4000 (0%)\n[OK] [INFO] System-wide Firewall Sections: 29 of 10000 (0.29%)\n[OK] [INFO] Network Introspection Sections: 1 of 500 (0.2%)\n[OK] [INFO] Groups: 74 of 20000 (0.37%)\n[OK] [INFO] Tier-1 Logical Routers: 4 of 4000 (0.1%)\n[OK] [INFO] IP Sets: 37 of 10000 (0.37%)\n[OK] [INFO] Network Introspection Service Chains: 0 of 24 (0%)\n[OK] [INFO] Network Introspection Service Paths: 0 of 4000 (0%)\n[OK] [INFO] Tier-0 Logical Routers: 2 of 160 (1.25%)\n[OK] [INFO] DHCP Server Instances: 0 of 10000 (0%)\n| number_of_nat_rules=0%;70;100;0;100 number_of_si_rules=0.01%;70;100;0;100 number_of_gi_protected_hosts=0%;70;100;0;100 number_of_prepared_hosts=1.75%;70;100;0;100 number_of_firewall_rules=0.08%;70;100;0;100 number_of_dhcp_ip_pools=0%;70;100;0;100 number_of_edge_nodes=3.12%;70;100;0;100 number_of_active_directory_domains=0%;70;100;0;100 number_of_vcenter_clusters=3.12%;70;100;0;100 number_of_prefix_list=4%;70;100;0;100 number_of_logical_switches=0.12%;70;100;0;100 number_of_logical_ports=0.58%;70;100;0;100 number_of_active_directory_groups=0%;70;100;0;100 number_of_dfw_rules=0.07%;70;100;0;100 number_of_gi_protected_vms=0%;70;100;0;100 number_of_dfw_sections=0.23%;70;100;0;100 number_of_groups_based_on_ip_sets=0.37%;70;100;0;100 number_of_edge_clusters=1.87%;70;100;0;100 number_of_tier1_with_nat_rule=0%;70;100;0;100 number_of_firewall_sections=0.29%;70;100;0;100 number_of_si_sections=0.2%;70;100;0;100 number_of_nsgroup=0.37%;70;100;0;100 number_of_tier1_routers=0.1%;70;100;0;100 number_of_ipsets=0.37%;70;100;0;100 number_of_si_service_chains=0%;70;100;0;100 number_of_si_service_paths=0%;70;100;0;100 number_of_tier0_routers=1.25%;70;100;0;100 number_of_dhcp_servers=0%;70;100;0;100')
+
+    @mock.patch('builtins.print')
+    @mock.patch('requests.request')
+    def test_capacity_usage_exclude(self, mock_req, mock_print):
+
+        with open('testdata/fixtures/capacity-usage.json') as f:
+            testdata = json.load(f)
+
+        m = mock.MagicMock()
+        m.status_code = 200
+        m.json.return_value = testdata
+        mock_req.return_value = m
+
+        c = Client('api', 'username', 'password', logger=None, verify=True, max_age=5)
+
+        actual = c.get_capacity_usage(".*").print_and_return()
+        expected = 0