Allow pinning kernel versions to work around hardware / driver issues

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

diff --git a/cookbooks/hardware/recipes/default.rb b/cookbooks/hardware/recipes/default.rb
index d4c8d047037dc6650c7c52c1c7a90620cfd61dcb..620c7d58bb686507dd3321732037f5f2c311871c 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"
@@ -117,6 +120,16 @@ unless unit.nil?
   end
 end
 
+# if we need a different / special kernel version to make the hardware
+# work (e.g: https://github.com/openstreetmap/operations/issues/45) then
+# ensure that we have the package installed. the grub template will
+# make sure that this is the default on boot.
+unless node[:hardware][:grub][:kernel] == :latest
+  package "linux-image-#{node[:hardware][:grub][:kernel]}-generic"
+  package "linux-image-extra-#{node[:hardware][:grub][:kernel]}-generic"
+  package "linux-headers-#{node[:hardware][:grub][:kernel]}-generic"
+end
+
 if File.exist?("/etc/default/grub")
   execute "update-grub" do
     action :nothing
@@ -188,7 +201,7 @@ node[:kernel][:modules].each_key do |modname|
   when "mptsas"
     tools_packages << "lsiutil"
     # status_packages["mpt-status"] ||= []
-  when "mpt2sas"
+  when "mpt2sas", "mpt3sas"
     tools_packages << "sas2ircu"
     status_packages["sas2ircu-status"] ||= []
   when "megaraid_mm"
@@ -277,7 +290,7 @@ 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 /
@@ -324,11 +337,13 @@ if status_packages["sas2ircu-status"]
   Dir.glob("/sys/class/scsi_host/host*") do |host|
     driver = File.new("#{host}/proc_name").read.chomp
 
-    next unless driver == "mpt2sas"
+    next unless driver == "mpt2sas" || driver == "mpt3sas"
 
     bus = host.sub("/sys/class/scsi_host/host", "")
 
     Dir.glob("/sys/bus/scsi/devices/#{bus}:0:*/scsi_generic/*").each do |sg|
+      next if File.directory?("#{sg}/../../block")
+
       disks << { :device => File.basename(sg) }
     end
   end
@@ -348,7 +363,7 @@ if status_packages["aacraid-status"]
   end
 end
 
-if tools_packages.include?("areca")
+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
@@ -368,7 +383,7 @@ if tools_packages.include?("areca")
 end
 
 disks.each do |disk|
-  if disk[:device] =~ /^cciss\/(.*)$/
+  if disk[:device] =~ %r{^cciss/(.*)$}
     id = File.read("/sys/bus/cciss/devices/#{Regexp.last_match[1]}/unique_id").chomp
 
     disk[:munin] = "cciss-3#{id.downcase}"
@@ -405,8 +420,15 @@ 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_"
@@ -430,7 +452,7 @@ else
 end
 
 plugins = Dir.glob("/etc/munin/plugins/smart_*").map { |p| File.basename(p) } -
-          disks.map { |d| "smart_#{d[:munin_name]}" }
+          disks.map { |d| "smart_#{d[:munin]}" }
 
 plugins.each do |plugin|
   munin_plugin plugin do