X-Git-Url: https://git.openstreetmap.org/chef.git/blobdiff_plain/3a4d52bd4c0f8e6ae7679145846b7bc7845d8ace..adc8d923c01cee17a5818caf1b59e251b97540c5:/cookbooks/hardware/recipes/default.rb

diff --git a/cookbooks/hardware/recipes/default.rb b/cookbooks/hardware/recipes/default.rb
index 4f0ec7df6..8342a9ac3 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
@@ -111,12 +119,30 @@ unless unit.nil?
 
   service "ttyS#{unit}" do
     provider Chef::Provider::Service::Upstart
-    action [ :enable, :start ]
+    action [:enable, :start]
     supports :status => true, :restart => true, :reload => false
     subscribes :restart, "template[/etc/init/ttyS#{unit}.conf]"
   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
@@ -168,7 +194,7 @@ if node[:lsb][:release].to_f >= 12.10
   end
 
   service "irqbalance" do
-    action [ :start, :enable ]
+    action [:start, :enable]
     supports :status => false, :restart => true, :reload => false
     subscribes :restart, "template[/etc/default/irqbalance]"
   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"
@@ -200,22 +226,24 @@ node[:kernel][:modules].each_key do |modname|
   when "aacraid"
     tools_packages << "arcconf"
     status_packages["aacraid-status"] ||= []
+  when "arcmsr"
+    tools_packages << "areca"
   end
 end
 
 node[:block_device].each do |name, attributes|
-  if attributes[:vendor] == "HP" && attributes[:model] == "LOGICAL VOLUME"
-    if name =~ /^cciss!(c[0-9]+)d[0-9]+$/
-      status_packages["cciss-vol-status"] |= [ "cciss/#{Regexp.last_match[1]}d0" ]
-    else
-      Dir.glob("/sys/block/#{name}/device/scsi_generic/*").each do |sg|
-        status_packages["cciss-vol-status"] |= [ File.basename(sg) ]
-      end
+  next unless attributes[:vendor] == "HP" && attributes[:model] == "LOGICAL VOLUME"
+
+  if name =~ /^cciss!(c[0-9]+)d[0-9]+$/
+    status_packages["cciss-vol-status"] |= ["cciss/#{Regexp.last_match[1]}d0"]
+  else
+    Dir.glob("/sys/block/#{name}/device/scsi_generic/*").each do |sg|
+      status_packages["cciss-vol-status"] |= [File.basename(sg)]
     end
   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
@@ -238,7 +282,7 @@ end
     end
 
     service "#{status_package}d" do
-      action [ :start, :enable ]
+      action [:start, :enable]
       supports :status => false, :restart => true, :reload => false
       subscribes :restart, "template[/etc/default/#{status_package}d]"
     end
@@ -253,6 +297,117 @@ end
   end
 end
 
+disks = if node[:hardware][:disk]
+          node[:hardware][:disk][:disks]
+        else
+          []
+        end
+
+disks = disks.map do |disk|
+  next if disk[:state] == "spun_down"
+
+  if disk[:smart_device]
+    controller = node[:hardware][:disk][:controllers][disk[:controller]]
+    device = controller[:device].sub("/dev/", "")
+    smart = disk[:smart_device]
+
+    if device.start_with?("cciss/") && smart =~ /^cciss,(\d+)$/
+      array = node[:hardware][:disk][:arrays][disk[:arrays].first]
+      munin = "cciss-3#{array[:wwn]}-#{Regexp.last_match(1)}"
+    elsif smart =~ /^.*,(\d+)$/
+      munin = "#{device}-#{Regexp.last_match(1)}"
+    elsif smart =~ %r{^.*,(\d+)/(\d+)$}
+      munin = "#{device}-#{Regexp.last_match(1)}:#{Regexp.last_match(2)}"
+    end
+  elsif disk[:device]
+    device = disk[:device].sub("/dev/", "")
+    munin = device
+  end
+
+  next if device.nil?
+
+  Hash[
+    :device => device,
+    :smart => smart,
+    :munin => munin,
+    :hddtemp => munin.tr("-:", "_")
+  ]
+end
+
+disks = disks.compact
+
+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"
+    group "root"
+    mode 0644
+    variables :disks => disks
+    notifies :reload, "service[smartmontools]"
+  end
+
+  template "/etc/default/smartmontools" do
+    source "smartmontools.erb"
+    owner "root"
+    group "root"
+    mode 0644
+    notifies :restart, "service[smartmontools]"
+  end
+
+  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[:smart] && disk[:smart].start_with?("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
+    action [:stop, :disable]
+  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")
@@ -288,7 +443,11 @@ if node[:lsb][:release].to_f <= 12.10
   end
 else
   service "kmod" do
-    provider Chef::Provider::Service::Upstart
+    if node[:lsb][:release].to_f >= 15.10
+      provider Chef::Provider::Service::Systemd
+    else
+      provider Chef::Provider::Service::Upstart
+    end
     action :nothing
     subscribes :start, "template[/etc/modules]"
   end
@@ -306,13 +465,37 @@ if node[:hardware][:watchdog]
   end
 
   service "watchdog" do
-    action [ :enable, :start ]
+    action [:enable, :start]
   end
 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)
+
+    temps = if File.exist?("#{coretemp}/name")
+              Dir.glob("#{coretemp}/temp*_input").map do |temp|
+                File.basename(temp).sub("temp", "").sub("_input", "").to_i
+              end.sort
+            else
+              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"