Don't do munin monitoring for disks behind an Areca controller

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

diff --git a/cookbooks/hardware/recipes/default.rb b/cookbooks/hardware/recipes/default.rb
index c9da86168f4fe00eebcc0aabaa02dd967c1d33cc..6a324a36e4d30846e3ff4acb4330fe4b7f67d150 100644 (file)
--- a/cookbooks/hardware/recipes/default.rb
+++ b/cookbooks/hardware/recipes/default.rb
@@ -17,6 +17,9 @@
 # limitations under the License.
 #
 
+include_recipe "tools"
+include_recipe "munin"
+
 case node[:cpu][:"0"][:vendor_id]
 when "GenuineIntel"
   package "intel-microcode"
@@ -200,6 +203,8 @@ node[:kernel][:modules].each_key do |modname|
   when "aacraid"
     tools_packages << "arcconf"
     status_packages["aacraid-status"] ||= []
+  when "arcmsr"
+    tools_packages << "areca"
   end
 end
 
@@ -225,6 +230,22 @@ end
   end
 end
 
+if tools_packages.include?("areca")
+  include_recipe "git"
+
+  git "/opt/areca" do
+    action :sync
+    repository "git://chef.openstreetmap.org/areca.git"
+    user "root"
+    group "root"
+  end
+else
+  directory "/opt/areca" do
+    action :delete
+    recursive true
+  end
+end
+
 ["cciss-vol-status", "mpt-status", "sas2ircu-status", "megaraid-status", "megaclisas-status", "aacraid-status"].each do |status_package|
   if status_packages.include?(status_package)
     package status_package
@@ -259,7 +280,7 @@ node[:block_device].each do |name, attributes|
   disks << { :device => name } if attributes[:vendor] == "ATA"
 end
 
-if status_packages["cciss-vol-status"]
+if status_packages["cciss-vol-status"] && File.exist?("/usr/sbin/cciss_vol_status")
   status_packages["cciss-vol-status"].each do |device|
     IO.popen(["cciss_vol_status", "-V", "/dev/#{device}"]).each do |line|
       disks << { :device => device, :driver => "cciss", :id => Regexp.last_match[1].to_i - 1 } if line =~ / bay ([0-9]+) +HP /
@@ -280,12 +301,28 @@ if status_packages["megaclisas-status"]
 
     IO.popen(["megacli", "-PDList", "-a#{controller}", "-NoLog"]).each do |line|
       disks << { :device => device, :driver => "megaraid",  :id => Regexp.last_match[1] } if line =~ /^Device Id: ([0-9]+)$/
+
+      disks.pop if line =~ /^Firmware state: Hotspare, Spun down$/
     end
 
     controller += 1
   end
 end
 
+if tools_packages.include?("lsiutil")
+  Dir.glob("/sys/class/scsi_host/host*") do |host|
+    driver = File.new("#{host}/proc_name").read.chomp
+
+    next unless driver == "mptsas"
+
+    bus = host.sub("/sys/class/scsi_host/host", "")
+
+    Dir.glob("/sys/bus/scsi/devices/#{bus}:0:*/scsi_generic/*").each do |sg|
+      disks << { :device => File.basename(sg) }
+    end
+  end
+end
+
 if status_packages["sas2ircu-status"]
   Dir.glob("/sys/class/scsi_host/host*") do |host|
     driver = File.new("#{host}/proc_name").read.chomp
@@ -300,6 +337,55 @@ if status_packages["sas2ircu-status"]
   end
 end
 
+if status_packages["aacraid-status"]
+  Dir.glob("/sys/class/scsi_host/host*") do |host|
+    driver = File.new("#{host}/proc_name").read.chomp
+
+    next unless driver == "aacraid"
+
+    bus = host.sub("/sys/class/scsi_host/host", "")
+
+    Dir.glob("/sys/bus/scsi/devices/#{bus}:1:*/scsi_generic/*").each do |sg|
+      disks << { :device => File.basename(sg) }
+    end
+  end
+end
+
+if tools_packages.include?("areca") && File.exist?("/opt/areca/x86_64/cli64")
+  device = IO.popen(["lsscsi", "-g"]).grep(%r{Areca +RAID controller .*/dev/(sg[0-9]+)}) do
+    Regexp.last_match[1]
+  end.first
+
+  IO.popen(["/opt/areca/x86_64/cli64", "disk", "info"]).each do |line|
+    next if line =~ /N\.A\./
+
+    if line =~ /^ +[0-9]+ +0*([0-9]+) +(?:Slot#|SLOT )0*([0-9]+) +/
+      enc = Regexp.last_match[1]
+      slot = Regexp.last_match[2]
+
+      disks << { :device => device, :driver => "areca", :id => "#{slot}/#{enc}" }
+    elsif line =~ /^ +([0-9]+) +[0-9]+ +/
+      disks << { :device => device, :driver => "areca", :id => Regexp.last_match[1] }
+    end
+  end
+end
+
+disks.each do |disk|
+  if disk[:device] =~ %r{^cciss/(.*)$}
+    id = File.read("/sys/bus/cciss/devices/#{Regexp.last_match[1]}/unique_id").chomp
+
+    disk[:munin] = "cciss-3#{id.downcase}"
+  else
+    disk[:munin] = disk[:device]
+  end
+
+  if disk[:id]
+    disk[:munin] = "#{disk[:munin]}-#{disk[:id].to_s.tr('/', ':')}"
+  end
+
+  disk[:hddtemp] = disk[:munin].tr("-:", "_")
+end
+
 if disks.count > 0
   package "smartmontools"
 
@@ -322,11 +408,44 @@ if disks.count > 0
 
   service "smartmontools" do
     action [:enable, :start]
+    supports :status => true, :restart => true, :reload => true
+  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
+  # throw errors if it clashes with munin
+  disks = disks.reject { |disk| disk[:driver] == "areca" }
+
+  disks.each do |disk|
+    munin_plugin "smart_#{disk[:munin]}" do
+      target "smart_"
+      conf "munin.smart.erb"
+      conf_variables :disk => disk
+    end
+  end
+
+  munin_plugin "hddtemp_smartctl" do
+    conf "munin.hddtemp.erb"
+    conf_variables :disks => disks
   end
 else
   service "smartmontools" do
     action [:stop, :disable]
   end
+
+  munin_plugin "hddtemp_smartctl" do
+    action :delete
+  end
+end
+
+plugins = Dir.glob("/etc/munin/plugins/smart_*").map { |p| File.basename(p) } -
+          disks.map { |d| "smart_#{d[:munin]}" }
+
+plugins.each do |plugin|
+  munin_plugin plugin do
+    action :delete
+  end
 end
 
 if File.exist?("/etc/mdadm/mdadm.conf")