Enable prometheus SMART monitoring for Areca controllers

[chef.git] / cookbooks / hardware / recipes / default.rb

diff --git a/cookbooks/hardware/recipes/default.rb b/cookbooks/hardware/recipes/default.rb
index 711fa45712420f6a8d0302671f3d32dcc5acfbda..42d2e911709fd9d3b37e29b00b5044506943f759 100644 (file)
--- a/cookbooks/hardware/recipes/default.rb
+++ b/cookbooks/hardware/recipes/default.rb
@@ -51,7 +51,7 @@ end
 
 units = []
 
-if node[:roles].include?("bytemark") || node[:roles].include?("exonetric")
+if node[:roles].include?("bytemark") || node[:roles].include?("exonetric") || node[:roles].include?("prgmr")
   units << "0"
 end
 
@@ -174,7 +174,14 @@ service "haveged" do
   action [:enable, :start]
 end
 
-package "ipmitool" if node[:kernel][:modules].include?("ipmi_si")
+if node[:kernel][:modules].include?("ipmi_si")
+  package "ipmitool"
+  package "freeipmi-tools"
+
+  prometheus_exporter "ipmi" do
+    port 9290
+  end
+end
 
 package "irqbalance"
 
@@ -342,23 +349,23 @@ intel_nvmes = nvmes.select { |pci| pci[:vendor_name] == "Intel Corporation" }
 if !intel_ssds.empty? || !intel_nvmes.empty?
   package "unzip"
 
-  intel_ssd_tool_version = "3.0.25"
+  intel_ssd_tool_version = "3.0.26"
 
   remote_file "#{Chef::Config[:file_cache_path]}/Intel_SSD_Data_Center_Tool_#{intel_ssd_tool_version}_Linux.zip" do
-    source "https://downloadmirror.intel.com/29556/eng/Intel_SSD_Data_Center_Tool_#{intel_ssd_tool_version}_Linux.zip"
+    source "https://downloadmirror.intel.com/29720/eng/Intel_SSD_DCT_#{intel_ssd_tool_version}_Linux.zip"
   end
 
   execute "#{Chef::Config[:file_cache_path]}/Intel_SSD_Data_Center_Tool_#{intel_ssd_tool_version}_Linux.zip" do
-    command "unzip Intel_SSD_Data_Center_Tool_#{intel_ssd_tool_version}_Linux.zip Intel_SSD_Data_Center_Tool_#{intel_ssd_tool_version}_Linux/isdct_#{intel_ssd_tool_version}-1_amd64.deb"
+    command "unzip Intel_SSD_Data_Center_Tool_#{intel_ssd_tool_version}_Linux.zip isdct_#{intel_ssd_tool_version}.400-1_amd64.deb"
     cwd Chef::Config[:file_cache_path]
     user "root"
     group "root"
-    not_if { ::File.exist?("#{Chef::Config[:file_cache_path]}/Intel_SSD_Data_Center_Tool_#{intel_ssd_tool_version}_Linux/isdct_#{intel_ssd_tool_version}-1_amd64.deb") }
+    not_if { ::File.exist?("#{Chef::Config[:file_cache_path]}/isdct_#{intel_ssd_tool_version}.400-1_amd64.deb") }
   end
 
   dpkg_package "isdct" do
     version "#{intel_ssd_tool_version}-1"
-    source "#{Chef::Config[:file_cache_path]}/Intel_SSD_Data_Center_Tool_#{intel_ssd_tool_version}_Linux/isdct_#{intel_ssd_tool_version}-1_amd64.deb"
+    source "#{Chef::Config[:file_cache_path]}/isdct_#{intel_ssd_tool_version}.400-1_amd64.deb"
   end
 end
 
@@ -380,6 +387,13 @@ disks = disks.map do |disk|
       elsif smart =~ %r{^.*,(\d+)/(\d+)$}
         munin = "#{device}-#{Regexp.last_match(1)}:#{Regexp.last_match(2)}"
       end
+    elsif disk[:device]
+      device = disk[:device].sub("/dev/", "")
+      smart = disk[:smart_device]
+
+      if smart =~ /^.*,(\d+),(\d+),(\d+)$/
+        munin = "#{device}-#{Regexp.last_match(1)}:#{Regexp.last_match(2)}:#{Regexp.last_match(3)}"
+      end
     end
   elsif disk[:device] =~ %r{^/dev/(nvme\d+)n\d+$}
     device = Regexp.last_match(1)
@@ -439,6 +453,18 @@ if disks.count.positive?
     subscribes :restart, "template[/etc/default/smartmontools]"
   end
 
+  template "/etc/prometheus/collectors/smart.devices" do
+    source "smart.devices.erb"
+    owner "root"
+    group "root"
+    mode "644"
+    variables :disks => disks
+  end
+
+  prometheus_collector "smart" do
+    interval "15m"
+  end
+
   # Don't try and do munin monitoring of disks behind
   # an Areca controller as they only allow one thing to
   # talk to the controller at a time and smartd will
@@ -452,18 +478,6 @@ if disks.count.positive?
       conf_variables :disk => disk
     end
   end
-
-  template "/etc/prometheus/collectors/smart.devices" do
-    source "smart.devices.erb"
-    owner "root"
-    group "root"
-    mode "644"
-    variables :disks => disks
-  end
-
-  prometheus_collector "smart" do
-    interval "15m"
-  end
 else
   service "smartd" do
     action [:stop, :disable]