Merge pull request #940 from DataDog/tristan/haproxy-count-per-status

HAProxy: gauge missing statuses with zero

checks.d/haproxy.py

@@ -1,9 +1,7 @@
-import urlparse
 import urllib2
-import socket
 
 from checks import AgentCheck
-from util import json, headers
+from util import headers
 
 import time
 
@@ -19,6 +17,9 @@ class Services(object):
     BACKEND = 'BACKEND'
     FRONTEND = 'FRONTEND'
     ALL = (BACKEND, FRONTEND)
+    ALL_STATUSES = (
+            'up', 'open', 'no_check', 'down', 'maint', 'nolb'
+        )
 
 class HAProxy(AgentCheck):
     def __init__(self, name, init_config, agentConfig):
@@ -119,16 +120,17 @@ def _process_data(
 
             self._update_data_dict(data_dict, back_or_front)
 
-            self._update_hosts_statuses_if_needed(
-                collect_status_metrics, collect_status_metrics_by_host,
-                data_dict, hosts_statuses
-            )
 
             if self._should_process(data_dict, collect_aggregates_only):
+                # update status
+                self._update_hosts_statuses_if_needed(
+                    collect_status_metrics, collect_status_metrics_by_host,
+                    data_dict, hosts_statuses
+                )
                 # Send the list of data to the metric and event callbacks
                 self._process_metrics(data_dict, url)
-            if process_events:
-                self._process_event(data_dict, url)
+                if process_events:
+                    self._process_event(data_dict, url)
 
         if collect_status_metrics:
             self._process_status_metric(hosts_statuses, collect_status_metrics_by_host)
@@ -182,7 +184,7 @@ def _should_process(self, data_dict, collect_aggregates_only):
             if self._is_aggregate(data_dict):
                 return True
             return False
-        elif data_dict['svname'] is Services.BACKEND:
+        elif data_dict['svname'] == Services.BACKEND:
             return False
         return True
 
@@ -195,12 +197,12 @@ def _process_status_metric(self, hosts_statuses, collect_status_metrics_by_host)
                 service, status = host_status
             status = status.lower()
 
-            tags = ['status:%s' % status, 'service:%s' % service]
+            tags = ['service:%s' % service]
             if collect_status_metrics_by_host:
                 tags.append('backend:%s' % hostname)
-            self.gauge("haproxy.count_per_status", count, tags=tags)
+            self._gauge_all_statuses("haproxy.count_per_status", count, status, tags=tags)
 
-            if 'up' in status:
+            if 'up' in status or 'open' in status:
                 agg_statuses[service]['available'] += count
             if 'down' in status or 'maint' in status or 'nolb' in status:
                 agg_statuses[service]['unavailable'] += count
@@ -210,6 +212,13 @@ def _process_status_metric(self, hosts_statuses, collect_status_metrics_by_host)
                 tags = ['status:%s' % status, 'service:%s' % service]
                 self.gauge("haproxy.count_per_status", count, tags=tags)
 
+    def _gauge_all_statuses(self, metric_name, count, status, tags):
+        self.gauge(metric_name, count, tags + ['status:%s' % status])
+        for state in Services.ALL_STATUSES:
+            if state != status:
+                self.gauge(metric_name, 0, tags + ['status:%s' % state])
+
+
     def _process_metrics(self, data, url):
         """
         Data is a dictionary related to one host
@@ -263,13 +272,13 @@ def _process_event(self, data, url):
     def _create_event(self, status, hostname, lastchg, service_name):
         if status == "DOWN":
             alert_type = "error"
-            title = "HAProxy %s front-end reported %s %s" % (service_name, hostname, status)
+            title = "%s reported %s %s" % (service_name, hostname, status)
         else:
             if status == "UP":
                 alert_type = "success"
             else:
                 alert_type = "info"
-            title = "HAProxy %s front-end reported %s back and %s" % (service_name, hostname, status)
+            title = "%s reported %s back and %s" % (service_name, hostname, status)
 
         return {
              'timestamp': int(time.time() - lastchg),

emitter.py

@@ -18,7 +18,7 @@ def get_http_library(proxy_settings, use_forwarder):
     else:
         # Python version < 2.6.3
         import urllib2proxy as urllib2
-    return urllib2 
+    return urllib2
 
 def post_headers(agentConfig, payload):
     return {
@@ -75,13 +75,13 @@ def get_opener(log, proxy_settings, use_forwarder, urllib2):
         log.debug("Not using proxy settings")
     else:
         proxy_url = '%s:%s' % (proxy_settings['host'], proxy_settings['port'])
-        
+
         if proxy_settings.get('user') is not None:
             proxy_auth = proxy_settings['user']
             if proxy_settings.get('password') is not None:
                 proxy_auth = '%s:%s' % (proxy_auth, proxy_settings['password'])
             proxy_url = '%s@%s' % (proxy_auth, proxy_url)
-        
+
         proxy = {'https': proxy_url}
         log.debug("Using proxy settings %s" % proxy)
 

tests/test_haproxy.py

@@ -89,7 +89,7 @@ def testCheck(self):
         self.assertTrue(len(metrics) > 0)
 
         self.assertEquals(len([t for t in metrics
-            if t[0] == "haproxy.backend.bytes.in_rate"]), 4, metrics)
+            if t[0] == "haproxy.backend.bytes.in_rate"]), 3, metrics)
         self.assertEquals(len([t for t in metrics
             if t[0] == "haproxy.frontend.session.current"]), 1, metrics)
 
@@ -151,7 +151,7 @@ def testOpenConfig(self):
         self.assertTrue(len(metrics) > 0)
 
         self.assertEquals(len([t for t in metrics
-            if t[0] == "haproxy.backend.bytes.in_rate"]), 4, metrics)
+            if t[0] == "haproxy.backend.bytes.in_rate"]), 3, metrics)
         self.assertEquals(len([t for t in metrics
             if t[0] == "haproxy.frontend.session.current"]), 1, metrics)