From 87c958638df68fd6a45b59314bb10075d12275bd Mon Sep 17 00:00:00 2001
From: Roberto Rosario <roberto.rosario.gonzalez@gmail.com>
Date: Mon, 30 Jul 2012 23:51:05 -0400
Subject: [PATCH] Implement clustering housekeeping, deleting 'dead' nodes

---
 apps/clustering/__init__.py |  9 +++---
 apps/clustering/models.py   | 62 +++++++++++++++++++++++++++++++------
 apps/clustering/tasks.py    | 16 +++++++---
 3 files changed, 69 insertions(+), 18 deletions(-)

diff --git a/apps/clustering/__init__.py b/apps/clustering/__init__.py
index 4149c12347..f5b453f6d2 100644
--- a/apps/clustering/__init__.py
+++ b/apps/clustering/__init__.py
@@ -6,13 +6,12 @@ from scheduler.api import register_interval_job
 from navigation.api import bind_links
 from project_tools.api import register_tool
 
-from .tasks import refresh_node
+from .tasks import node_heartbeat, house_keeping
 from .links import tool_link, node_list
-from .models import Node
+from .models import Node, ClusteringConfig
 
-NODE_REFRESH_INTERVAL = 1
-
-register_interval_job('refresh_node', _(u'Update a node\'s properties.'), refresh_node, seconds=NODE_REFRESH_INTERVAL)
+register_interval_job('node_heartbeat', _(u'Update a node\'s properties.'), node_heartbeat, seconds=ClusteringConfig.get().node_heartbeat_interval)
+register_interval_job('house_keeping', _(u'Check for unresponsive nodes in the cluster list.'), house_keeping, seconds=1)
 
 register_tool(tool_link)
 bind_links([Node, 'node_list'], [node_list], menu_name='secondary_menu')
diff --git a/apps/clustering/models.py b/apps/clustering/models.py
index 67ded6cf04..764691aadc 100644
--- a/apps/clustering/models.py
+++ b/apps/clustering/models.py
@@ -11,30 +11,47 @@ from django.db import close_connection
 from django.utils.translation import ugettext_lazy as _
 from django.utils.translation import ugettext
 
+from common.models import Singleton
+
+DEFAULT_NODE_TTL = 5
+DEFAULT_NODE_HEARTBEAT_INTERVAL = 1
+
 
 class NodeManager(models.Manager):
     def myself(self):
-        node, created = self.model.objects.get_or_create(hostname=platform.node(), defaults={'memory_usage': 100})
+        node, created = self.model.objects.get_or_create(hostname=platform.node())
         node.refresh()
+        if created:
+            # Store the refresh data because is a new instance
+            node.save()
         return node
 
 
 class Node(models.Model):
     hostname = models.CharField(max_length=255, verbose_name=_(u'hostname'))
-    cpuload = models.PositiveIntegerField(blank=True, default=0, verbose_name=_(u'cpu load'))
+    cpuload = models.FloatField(blank=True, default=0.0, verbose_name=_(u'cpu load'))
     heartbeat = models.DateTimeField(blank=True, default=datetime.datetime.now(), verbose_name=_(u'last heartbeat check'))
-    memory_usage = models.FloatField(blank=True, verbose_name=_(u'memory usage'))
-    
-    objects = NodeManager()
+    memory_usage = models.FloatField(blank=True, default=0.0, verbose_name=_(u'memory usage'))
 
+    objects = NodeManager()
+    
+    @classmethod
+    def platform_info(cls):
+        return {
+            'cpuload': psutil.cpu_percent(),
+            'memory_usage': psutil.phymem_usage().percent
+        }
+    
     def __unicode__(self):
         return self.hostname
         
     def refresh(self):
-        self.cpuload = psutil.cpu_percent()
-        self.memory_usage = psutil.phymem_usage().percent
-        self.save()
-   
+        if self.hostname == platform.node():
+            # Make we can only update ourselves
+            info = Node.platform_info()
+            self.cpuload = info['cpuload']
+            self.memory_usage = info['memory_usage']
+        
     def save(self, *args, **kwargs):
         self.heartbeat = datetime.datetime.now()
         return super(Node, self).save(*args, **kwargs)
@@ -42,3 +59,30 @@ class Node(models.Model):
     class Meta:
         verbose_name = _(u'node')
         verbose_name_plural = _(u'nodes')
+
+
+class ClusteringConfigManager(models.Manager):
+    def dead_nodes(self):
+        return Node.objects.filter(heartbeat__lt=datetime.datetime.now() - datetime.timedelta(seconds=self.model.get().node_time_to_live))
+
+    def delete_dead_nodes(self):
+        self.dead_nodes().delete()
+
+    def zombiest_node(self):
+        try:
+            return self.dead_nodes().order_by('-heartbeat')[0]
+        except IndexError:
+            return None
+
+
+class ClusteringConfig(Singleton):
+    node_time_to_live = models.PositiveIntegerField(verbose_name=(u'time to live (in seconds)'), default=DEFAULT_NODE_TTL) #  After this time a worker is considered dead
+    node_heartbeat_interval = models.PositiveIntegerField(verbose_name=(u'heartbeat interval'), default=DEFAULT_NODE_HEARTBEAT_INTERVAL)
+
+    objects = ClusteringConfigManager()
+
+    def __unicode__(self):
+        return ugettext('clustering config')
+
+    class Meta:
+        verbose_name = verbose_name_plural = _(u'clustering config')
diff --git a/apps/clustering/tasks.py b/apps/clustering/tasks.py
index aa9f01b53e..c5938047d9 100644
--- a/apps/clustering/tasks.py
+++ b/apps/clustering/tasks.py
@@ -4,14 +4,22 @@ import logging
 
 from lock_manager.decorators import simple_locking
 
-from .models import Node
+from .models import Node, ClusteringConfig
 
 LOCK_EXPIRE = 10
 
 logger = logging.getLogger(__name__)
 
 
-@simple_locking('refresh_node', 10)
-def refresh_node():
+@simple_locking('node_heartbeat', 10)
+def node_heartbeat():
     logger.debug('starting')
-    node = Node.objects.myself()  # Automatically calls the refresh() method too
+    node = Node.objects.myself()
+    node.save()
+
+
+@simple_locking('house_keeping', 10)
+def house_keeping():
+    logger.debug('starting')
+    ClusteringConfig.objects.delete_dead_nodes()
+