From 382eba220a12cb5fdd9b75322e5504c8c305e353 Mon Sep 17 00:00:00 2001
From: Brendan Shaklovitz <nyanshak@users.noreply.github.com>
Date: Wed, 3 Aug 2016 07:35:49 -0500
Subject: [PATCH] [gearmand] Add per-task gearman metrics (#2672)

* Adds gearman.{queued_by_task, running_by_task, workers_by_task}
  metrics to collect data on each individual task. This lets you see how
  many of each tasks are queued up to catch problems with any individual
  queue not being processed.
* Each new metric is tagged by task:<task_name>
* Limits the maximum number of tasks on which per-task metrics are
  collected
---
 checks.d/gearmand.py         | 43 ++++++++++++++++++++++++++++++------
 conf.d/gearmand.yaml.example |  8 +++++++
 2 files changed, 44 insertions(+), 7 deletions(-)

diff --git a/checks.d/gearmand.py b/checks.d/gearmand.py
index 823ce74b78..3f81344f00 100644
--- a/checks.d/gearmand.py
+++ b/checks.d/gearmand.py
@@ -9,6 +9,9 @@
 # project
 from checks import AgentCheck
 
+
+MAX_NUM_TASKS = 200
+
 class Gearman(AgentCheck):
     SERVICE_CHECK_NAME = 'gearman.can_connect'
 
@@ -20,18 +23,17 @@ def _get_client(self,host,port):
         return gearman.GearmanAdminClient(["%s:%s" %
             (host, port)])
 
-    def _get_metrics(self, client, tags):
-        data = client.get_status()
+    def _get_aggregate_metrics(self, tasks, tags):
         running = 0
         queued = 0
         workers = 0
 
-        for stat in data:
+        for stat in tasks:
             running += stat['running']
             queued += stat['queued']
             workers += stat['workers']
 
-        unique_tasks = len(data)
+        unique_tasks = len(tasks)
 
         self.gauge("gearman.unique_tasks", unique_tasks, tags=tags)
         self.gauge("gearman.running", running, tags=tags)
@@ -41,9 +43,34 @@ def _get_metrics(self, client, tags):
         self.log.debug("running %d, queued %d, unique tasks %d, workers: %d"
         % (running, queued, unique_tasks, workers))
 
+    def _get_per_task_metrics(self, tasks, task_filter, tags):
+        if len(task_filter) > MAX_NUM_TASKS:
+            self.warning(
+                "The maximum number of tasks you can specify is {}.".format(MAX_NUM_TASKS))
+
+        if not len(task_filter) == 0:
+            tasks = [t for t in tasks if t['task'] in task_filter]
+
+        if len(tasks) > MAX_NUM_TASKS:
+            # Display a warning in the info page
+            self.warning(
+                "Too many tasks to fetch. You must choose the tasks you are interested in by editing the gearmand.yaml configuration file or get in touch with Datadog Support")
+
+        for stat in tasks[:MAX_NUM_TASKS]:
+            running = stat['running']
+            queued = stat['queued']
+            workers = stat['workers']
+
+            task_tags = tags[:]
+            task_tags.append("task:{}".format(stat['task']))
+            self.gauge("gearman.running_by_task", running, tags=task_tags)
+            self.gauge("gearman.queued_by_task", queued, tags=task_tags)
+            self.gauge("gearman.workers_by_task", workers, tags=task_tags)
+
     def _get_conf(self, instance):
         host = instance.get('server', None)
         port = instance.get('port', None)
+        tasks = instance.get('tasks', [])
 
         if host is None:
             self.warning("Host not set, assuming 127.0.0.1")
@@ -55,12 +82,12 @@ def _get_conf(self, instance):
 
         tags = instance.get('tags', [])
 
-        return host, port, tags
+        return host, port, tasks, tags
 
     def check(self, instance):
         self.log.debug("Gearman check start")
 
-        host, port, tags = self._get_conf(instance)
+        host, port, task_filter, tags = self._get_conf(instance)
         service_check_tags = ["server:{0}".format(host),
             "port:{0}".format(port)]
 
@@ -70,7 +97,9 @@ def check(self, instance):
         tags += service_check_tags
 
         try:
-            self._get_metrics(client, tags)
+            tasks = client.get_status()
+            self._get_aggregate_metrics(tasks, tags)
+            self._get_per_task_metrics(tasks, task_filter, tags)
             self.service_check(self.SERVICE_CHECK_NAME, AgentCheck.OK,
                 message="Connection to %s:%s succeeded." % (host, port),
                 tags=service_check_tags)
diff --git a/conf.d/gearmand.yaml.example b/conf.d/gearmand.yaml.example
index 2abf945222..ab6a060672 100644
--- a/conf.d/gearmand.yaml.example
+++ b/conf.d/gearmand.yaml.example
@@ -3,6 +3,14 @@ init_config:
 instances:
   - server: localhost
     port: 4730
+    # Use the `tasks` parameter to specify the tasks you'd like to
+    # collect metrics on (up to 200 tasks).
+    #
+    # If you have fewer than 200 tasks, you don't have to set this parameter,
+    # the metrics will be collected on all the tasks by default.
+    # tasks:
+    #   - task1 
+    #   - task2
     # tags:
     #   - optional_tag_1
     #   - optional_tag_2