Add authentication and timeout options (#3214)

DataDog · Feb 27, 2019 · 295b5b2 · 295b5b2
1 parent f5e64b3
commit 295b5b2
Show file tree

Hide file tree

Showing 3 changed files with 46 additions and 9 deletions.
diff --git a/aerospike/README.md b/aerospike/README.md
@@ -7,8 +7,6 @@ Get metrics from Aerospike Database in real time to:
 * Visualize and monitor Aerospike states
 * Be notified about Aerospike failovers and events.
 
-Note: Authentication and TLS are not supported.
-
 ## Setup
 
 ### Installation
@@ -34,7 +32,8 @@ See [metadata.csv][5] for a list of metrics provided by this integration.
 
 ### Service Checks
 
-Aerospike does not include any service checks.
+- `aerospike.can_connect`
+- `aerospike.cluster_up`
 
 ### Events
 

diff --git a/aerospike/datadog_checks/aerospike/aerospike.py b/aerospike/datadog_checks/aerospike/aerospike.py
@@ -64,10 +64,20 @@ class AerospikeCheck(AgentCheck):
     def __init__(self, name, init_config, instances):
         super(AerospikeCheck, self).__init__(name, init_config, instances)
 
+        # https://www.aerospike.com/apidocs/python/aerospike.html#aerospike.client
         host = self.instance.get('host', 'localhost')
         port = int(self.instance.get('port', 3000))
+        tls_name = self.instance.get('tls_name')
+        self._host = (host, port, tls_name) if tls_name else (host, port)
+
+        # https://www.aerospike.com/apidocs/python/client.html#aerospike.Client.connect
+        self._username = self.instance.get('username')
+        self._password = self.instance.get('password')
+
+        # In milliseconds, see https://www.aerospike.com/apidocs/python/client.html#aerospike-info-policies
+        timeout = int(self.instance.get('timeout', 10)) * 1000
+        self._info_policies = {'timeout': timeout}
 
-        self._host = (host, port)
         self._metrics = set(self.instance.get('metrics', []))
         self._namespace_metrics = set(self.instance.get('namespace_metrics', []))
         self._required_namespaces = self.instance.get('namespaces')
@@ -85,33 +95,39 @@ def check(self, instance):
 
             self._client = client
 
+        # https://www.aerospike.com/docs/reference/info/#statistics
         self.collect_info('statistics', CLUSTER_METRIC_TYPE, required_keys=self._metrics, tags=self._tags)
 
+        # https://www.aerospike.com/docs/reference/info/#namespaces
         namespaces = self.get_namespaces()
 
         for ns in namespaces:
             namespace_tags = ['namespace:{}'.format(ns)]
             namespace_tags.extend(self._tags)
 
+            # https://www.aerospike.com/docs/reference/info/#namespace
             self.collect_info(
                 'namespace/{}'.format(ns),
                 NAMESPACE_METRIC_TYPE,
                 required_keys=self._namespace_metrics,
                 tags=namespace_tags
             )
 
+            # https://www.aerospike.com/docs/reference/info/#sindex
             sindex = self.get_info('sindex/{}'.format(ns))
             for idx in parse_namespace(sindex[:-1], ns, 'indexname'):
                 sindex_tags = ['sindex:{}'.format(idx)]
                 sindex_tags.extend(namespace_tags)
                 self.collect_info('sindex/{}/{}'.format(ns, idx), SINDEX_METRIC_TYPE, tags=sindex_tags)
 
+            # https://www.aerospike.com/docs/reference/info/#sets
             sets = self.get_info('sets/{}'.format(ns))
             for s in parse_namespace(sets, ns, 'set'):
                 set_tags = ['set:{}'.format(s)]
                 set_tags.extend(namespace_tags)
                 self.collect_info('sets/{}/{}'.format(ns, s), SET_METRIC_TYPE, separator=':', tags=set_tags)
 
+        # https://www.aerospike.com/docs/reference/info/#throughput
         self.collect_throughput(namespaces)
 
         self.service_check(SERVICE_CHECK_UP, self.OK, tags=self._tags)
@@ -149,7 +165,7 @@ def get_namespaces(self):
 
     def get_client(self):
         try:
-            client = aerospike.client({'hosts': [self._host]}).connect()
+            client = aerospike.client({'hosts': [self._host]}).connect(self._username, self._password)
         except Exception as e:
             self.log.error('Unable to connect to database: {}'.format(e))
             self.service_check(SERVICE_CHECK_CONNECT, self.CRITICAL, tags=self._tags)
@@ -158,8 +174,9 @@ def get_client(self):
             return client
 
     def get_info(self, command, separator=';'):
-        # command\tKEY=VALUE;KEY=VALUE;...
-        data = self._client.info_node(command, self._host)
+        # See https://www.aerospike.com/docs/reference/info/
+        # Example output: command\tKEY=VALUE;KEY=VALUE;...
+        data = self._client.info_node(command, self._host, self._info_policies)
 
         # Get rid of command and whitespace
         data = data[len(command):].strip()

diff --git a/aerospike/datadog_checks/aerospike/data/conf.yaml.example b/aerospike/datadog_checks/aerospike/data/conf.yaml.example
@@ -10,8 +10,29 @@ instances:
   #
     port: 3000
 
-  ## @param tags  - list of key:value elements - optional
-  ## List of tags to attach to every metric, event and service check emitted by this integration.
+  ## @param username - string - optional
+  ## The database user to authenticate as.
+  #
+  #  username: <USERNAME>
+
+  ## @param password - string - optional
+  ## The password of `username`.
+  #
+  #  password: <PASSWORD>
+
+  ## @param tls_name - string - optional
+  ## This must match the tls-name specified in the node’s server
+  ## configuration file and match the server’s CA certificate.
+  #
+  #  tls_name: <TLS_NAME>
+
+  ## @param timeout - integer - optional - default: 10
+  ## The read timeout in seconds.
+  #
+  #  timeout: 10
+
+  ## @param tags  - list of key:value strings - optional
+  ## List of tags to attach to every metric and service check emitted by this integration.
   ##
   ## Learn more about tagging: https://docs.datadoghq.com/tagging/
   #