X-Git-Url: https://git.openstreetmap.org./chef.git/blobdiff_plain/695c1e5c1bb1dae68d7366f7e60991e65aec159a..2b69a518ed7853fa5b52c95b09bcf47d09144193:/cookbooks/hardware/recipes/default.rb

diff --git a/cookbooks/hardware/recipes/default.rb b/cookbooks/hardware/recipes/default.rb
index 05d5c0794..238d65690 100644
--- a/cookbooks/hardware/recipes/default.rb
+++ b/cookbooks/hardware/recipes/default.rb
@@ -17,6 +17,13 @@
 # limitations under the License.
 #
 
+include_recipe "tools"
+include_recipe "munin"
+
+ohai_plugin "hardware" do
+  template "ohai.rb.erb"
+end
+
 case node[:cpu][:"0"][:vendor_id]
 when "GenuineIntel"
   package "intel-microcode"
@@ -77,6 +84,7 @@ if manufacturer == "HP" && node[:lsb][:release].to_f > 11.10
     repository "git://chef.openstreetmap.org/hp-legacy.git"
     user "root"
     group "root"
+    ignore_failure true
   end
 
   link "/opt/hp/hp-health/bin/hpasmd" do
@@ -117,6 +125,24 @@ 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.
+if node[:hardware][:grub][:kernel]
+  kernel_version = node[:hardware][:grub][:kernel]
+
+  package "linux-image-#{kernel_version}-generic"
+  package "linux-image-extra-#{kernel_version}-generic"
+  package "linux-headers-#{kernel_version}-generic"
+
+  boot_device = IO.popen(["df", "/boot"]).readlines.last.split.first
+  boot_uuid = IO.popen(["blkid", "-o", "value", "-s", "UUID", boot_device]).readlines.first.chomp
+  grub_entry = "gnulinux-advanced-#{boot_uuid}>gnulinux-#{kernel_version}-advanced-#{boot_uuid}"
+else
+  grub_entry = "0"
+end
+
 if File.exist?("/etc/default/grub")
   execute "update-grub" do
     action :nothing
@@ -128,7 +154,7 @@ if File.exist?("/etc/default/grub")
     owner "root"
     group "root"
     mode 0644
-    variables :unit => unit, :speed => speed
+    variables :unit => unit, :speed => speed, :entry => grub_entry
     notifies :run, "execute[update-grub]"
   end
 end
@@ -180,15 +206,15 @@ status_packages = {}
 node[:kernel][:modules].each_key do |modname|
   case modname
   when "cciss"
-    tools_packages << "hpacucli"
+    tools_packages << "hpssacli"
     status_packages["cciss-vol-status"] ||= []
   when "hpsa"
-    tools_packages << "hpacucli"
+    tools_packages << "hpssacli"
     status_packages["cciss-vol-status"] ||= []
   when "mptsas"
     tools_packages << "lsiutil"
     # status_packages["mpt-status"] ||= []
-  when "mpt2sas"
+  when "mpt2sas", "mpt3sas"
     tools_packages << "sas2ircu"
     status_packages["sas2ircu-status"] ||= []
   when "megaraid_mm"
@@ -200,6 +226,8 @@ node[:kernel][:modules].each_key do |modname|
   when "aacraid"
     tools_packages << "arcconf"
     status_packages["aacraid-status"] ||= []
+  when "arcmsr"
+    tools_packages << "areca"
   end
 end
 
@@ -215,7 +243,7 @@ node[:block_device].each do |name, attributes|
   end
 end
 
-%w(hpacucli lsiutil sas2ircu megactl megacli arcconf).each do |tools_package|
+%w(hpssacli lsiutil sas2ircu megactl megacli arcconf).each do |tools_package|
   if tools_packages.include?(tools_package)
     package tools_package
   else
@@ -225,6 +253,22 @@ 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 +303,7 @@ 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 /
@@ -267,9 +311,116 @@ if status_packages["cciss-vol-status"]
   end
 end
 
+if status_packages["megaclisas-status"] && File.exist?("/usr/sbin/megacli")
+  controller = 0
+
+  Dir.glob("/sys/class/scsi_host/host*") do |host|
+    driver = File.new("#{host}/proc_name").read.chomp
+
+    next unless driver == "megaraid_sas"
+
+    bus = host.sub("/sys/class/scsi_host/host", "")
+    device = File.basename(Dir.glob("/sys/bus/scsi/devices/#{bus}:*/scsi_generic/*").first)
+
+    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
+
+    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
+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"
 
+  template "/usr/local/bin/smartd-mailer" do
+    source "smartd-mailer.erb"
+    owner "root"
+    group "root"
+    mode 0755
+  end
+
   template "/etc/smartd.conf" do
     source "smartd.conf.erb"
     owner "root"
@@ -289,6 +440,21 @@ 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
 else
   service "smartmontools" do
@@ -296,6 +462,27 @@ else
   end
 end
 
+if disks.count > 0
+  munin_plugin "hddtemp_smartctl" do
+    conf "munin.hddtemp.erb"
+    conf_variables :disks => disks
+  end
+else
+  munin_plugin "hddtemp_smartctl" do
+    action :delete
+    conf "munin.hddtemp.erb"
+  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")
   mdadm_conf = edit_file "/etc/mdadm/mdadm.conf" do |line|
     line.gsub!(/^MAILADDR .*$/, "MAILADDR admins@openstreetmap.org")
@@ -356,6 +543,30 @@ end
 unless Dir.glob("/sys/class/hwmon/hwmon*").empty?
   package "lm-sensors"
 
+  Dir.glob("/sys/devices/platform/coretemp.*").each do |coretemp|
+    cpu = File.basename(coretemp).sub("coretemp.", "").to_i
+    chip = format("coretemp-isa-%04d", cpu)
+
+    if File.exist?("#{coretemp}/name")
+      temps = Dir.glob("#{coretemp}/temp*_input").map do |temp|
+        File.basename(temp).sub("temp", "").sub("_input", "").to_i
+      end.sort
+    else
+      temps = Dir.glob("#{coretemp}/hwmon/hwmon*/temp*_input").map do |temp|
+        File.basename(temp).sub("temp", "").sub("_input", "").to_i
+      end.sort
+    end
+
+    if temps.first == 1
+      node.default[:hardware][:sensors][chip][:temps][:temp1][:label] = "CPU #{cpu}"
+      temps.shift
+    end
+
+    temps.each_with_index do |temp, index|
+      node.default[:hardware][:sensors][chip][:temps]["temp#{temp}"][:label] = "CPU #{cpu} Core #{index}"
+    end
+  end
+
   execute "/etc/sensors.d/chef.conf" do
     action :nothing
     command "/usr/bin/sensors -s"