Porting of V5 YT-CC-1328 engineyard/ey-cookbooks-stable-v5#384

cookbooks/collectd/attributes/defaults.rb

@@ -4,3 +4,6 @@
 default['swap_crit_threshold'] = "0.70"
 
 default['collectd']['version'] = "5.7.2-2ubuntu1"
+
+# Enable monitoring of EC2/EBS credit balances only on T instances (T2, T3)
+default['collectd']['enable_credit_balances_monitoring'] = (size =~ /^t[a-z0-9]+\./)

cookbooks/collectd/files/default/ec2credits.types.db

@@ -0,0 +1,3 @@
+cpu_credits             balance:GAUGE:0:U, usage:GAUGE:0:U
+cpu_surplus_credits     balance:GAUGE:0:U, charged:GAUGE:0:U
+ebs_iops_burst          balance:GAUGE:0:100.1

cookbooks/collectd/libraries/ec2_credit_thresholds.rb

@@ -0,0 +1,36 @@
+class EC2CreditThresholds
+
+  def initialize(apps_data)
+    all_env_vars = merged_app_env_vars(apps_data)
+    set_ec2_credit_thresholds(all_env_vars)
+  end
+
+  attr_reader :cpu_credit_ok, :cpu_credit_warn, :cpu_credit_alert,
+    :vol_burst_ok, :vol_burst_warn, :vol_burst_alert
+
+  private
+
+  def merged_app_env_vars(apps_data)
+    apps_data
+      .map do |app_data|
+        metadata = app_data['components'].find {|component| component['key'] == 'app_metadata'}
+        return {} unless metadata && metadata['environment_variables']
+        Hash[
+          metadata['environment_variables'].collect do |var_hash|
+            [ var_hash['name'], ::Base64.strict_decode64(var_hash['value']) ]
+          end
+        ]
+      end
+      .reduce({}, :merge)
+  end
+
+  def set_ec2_credit_thresholds(all_env_vars)
+    @cpu_credit_ok = all_env_vars.fetch('EY_EC2_CPU_CREDITS_THRESHOLD_OK', 40)
+    @cpu_credit_warn = all_env_vars.fetch('EY_EC2_CPU_CREDITS_THRESHOLD_WARN', 30)
+    @cpu_credit_alert = all_env_vars.fetch('EY_EC2_CPU_CREDITS_THRESHOLD_ALERT', 15)
+    @vol_burst_ok = all_env_vars.fetch('EY_IOPS_BURST_BALANCE_THRESHOLD_OK', 40)
+    @vol_burst_warn = all_env_vars.fetch('EY_IOPS_BURST_BALANCE_THRESHOLD_WARN', 30)
+    @vol_burst_alert = all_env_vars.fetch('EY_IOPS_BURST_BALANCE_THRESHOLD_ALERT', 15)
+  end
+
+end

cookbooks/collectd/recipes/default.rb

@@ -12,6 +12,7 @@
 end
 
 include_recipe 'collectd::httpd'
+include_recipe 'collectd::ec2_credit_balances'
 
 template "/engineyard/bin/ey-alert.rb" do
   owner 'root'
@@ -91,7 +92,8 @@
         :load_failure => node['collectd']['load']['failure'],
         :swap_thresholds => SwapThresholds.new,
         :short_version => short_version,
-        :disk_thresholds => DiskThresholds.new
+        :disk_thresholds => DiskThresholds.new,
+        :enable_credit_balances_monitoring => node['collectd']['enable_credit_balances_monitoring']
       }
     }
   )

cookbooks/collectd/recipes/ec2_credit_balances.rb

@@ -0,0 +1,46 @@
+if node['collectd']['enable_credit_balances_monitoring'] then
+
+  # Install dependencies
+  enable_package 'dev-python/boto3' do
+    version '1.3.0'
+  end
+  enable_package 'dev-python/botocore' do
+    version '1.4.19'
+  end
+  enable_package 'dev-python/jmespath' do
+    version '0.9.0'
+  end
+  package 'dev-python/boto3' do
+    action :install
+  end
+  package 'dev-python/requests' do
+    action :install
+  end
+
+  managed_template "/engineyard/bin/check_for_ec2_credit_balances.py" do
+    source "check_for_ec2_credit_balances.py.erb"
+    owner node["owner_name"]
+    group node["owner_name"]
+    mode 0755
+    variables({
+      :ec2_thresholds => EC2CreditThresholds.new(node['dna']['engineyard']['environment']['apps'])
+    })
+  end
+
+  # The script stores internal state here
+  directory "/tmp/check_ec2_credit_balances" do
+    owner node["owner_name"]
+    group node["owner_name"]
+    mode 0755
+    recursive true
+  end
+
+  cookbook_file "/etc/engineyard/ec2credits.types.db" do
+    source "ec2credits.types.db"
+    owner node["owner_name"]
+    group node["owner_name"]
+    backup 0
+    mode 0644
+  end
+
+end # node['collectd']['enable_credit_balances_monitoring']