X-Git-Url: https://git.openstreetmap.org/chef.git/blobdiff_plain/b5fb6f1ea5350f2d792a6f882b47631ac67ee407..f6b63c0bdcfde35f28573ea4a3444daff2c1fe80:/cookbooks/hardware/recipes/default.rb

diff --git a/cookbooks/hardware/recipes/default.rb b/cookbooks/hardware/recipes/default.rb
index e9fe54e05..5cbb13eb0 100644
--- a/cookbooks/hardware/recipes/default.rb
+++ b/cookbooks/hardware/recipes/default.rb
@@ -17,12 +17,26 @@
 # 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"
 end
 
-if node[:dmi] and node[:dmi][:system]
+case node[:cpu][:"0"][:vendor_id]
+when "AuthenticAMD"
+  if node[:lsb][:release].to_f >= 14.04
+    package "amd64-microcode"
+  end
+end
+
+if node[:dmi] && node[:dmi][:system]
   case node[:dmi][:system][:manufacturer]
   when "empty"
     manufacturer = node[:dmi][:base_board][:manufacturer]
@@ -36,79 +50,75 @@ else
   product = "Unknown"
 end
 
+units = []
+
+if node[:roles].include?("bytemark") || node[:roles].include?("exonetric")
+  units << "0"
+end
+
 case manufacturer
 when "HP"
   package "hponcfg"
   package "hp-health"
-  package "hpacucli"
-  unit = "1"
-  speed = "115200"
+  units << "1"
 when "TYAN"
-  unit = "0"
-  speed = "115200"
+  units << "0"
 when "TYAN Computer Corporation"
-  unit = "0"
-  speed = "115200"
+  units << "0"
 when "Supermicro"
   case product
-  when "H8DGU", "X9SCD", "X7DBU", "X7DW3", "X9DR7/E-(J)LN4F", "X9DR3-F"
-    unit = "1"
-    speed = "115200"
+  when "H8DGU", "X9SCD", "X7DBU", "X7DW3", "X9DR7/E-(J)LN4F", "X9DR3-F", "X9DRW", "SYS-2028U-TN24R4T+"
+    units << "1"
   else
-    unit = "0"
-    speed = "115200"
+    units << "0"
   end
 when "IBM"
-  unit = "0"
-  speed = "115200"
+  units << "0"
 end
 
-if manufacturer == "HP" and node[:lsb][:release].to_f > 11.10
-  include_recipe "git"
-
-  git "/opt/hp/hp-legacy" do
-    action :sync
-    repository "git://chef.openstreetmap.org/hp-legacy.git"
-    user "root"
-    group "root"
-  end
+units.sort.uniq.each do |unit|
+  if node[:lsb][:release].to_f >= 16.04
+    service "serial-getty@ttyS#{unit}" do
+      action [:enable, :start]
+    end
+  else
+    file "/etc/init/ttySttyS#{unit}.conf" do
+      action :delete
+    end
 
-  link "/opt/hp/hp-health/bin/hpasmd" do
-    to "/opt/hp/hp-legacy/hpasmd"
-  end
+    template "/etc/init/ttyS#{unit}.conf" do
+      source "tty.conf.erb"
+      owner "root"
+      group "root"
+      mode 0644
+      variables :unit => unit
+    end
 
-  link "/usr/lib/libhpasmintrfc.so.3.0" do
-    to "/opt/hp/hp-legacy/libhpasmintrfc.so.3.0"
-  end
-
-  link "/usr/lib/libhpasmintrfc.so.3" do
-    to "libhpasmintrfc.so.3.0"
-  end
-
-  link "/usr/lib/libhpasmintrfc.so" do
-    to "libhpasmintrfc.so.3.0"
+    service "ttyS#{unit}" do
+      provider Chef::Provider::Service::Upstart
+      action [:enable, :start]
+      supports :status => true, :restart => true, :reload => false
+      subscribes :restart, "template[/etc/init/ttyS#{unit}.conf]"
+    end
   end
 end
 
-unless unit.nil?
-  file "/etc/init/ttySttyS#{unit}.conf" do
-    action :delete
-  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]
 
-  template "/etc/init/ttyS#{unit}.conf" do
-    source "tty.conf.erb"
-    owner "root"
-    group "root"
-    mode 0644
-    variables :unit => unit, :speed => speed
-  end
+  package "linux-image-#{kernel_version}-generic"
+  package "linux-image-extra-#{kernel_version}-generic"
+  package "linux-headers-#{kernel_version}-generic"
 
-  service "ttyS#{unit}" do
-    provider Chef::Provider::Service::Upstart
-    action [ :enable, :start ]
-    supports :status => true, :restart => true, :reload => false
-    subscribes :restart, resources(:template => "/etc/init/ttyS#{unit}.conf")
-  end
+  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")
@@ -122,8 +132,8 @@ if File.exist?("/etc/default/grub")
     owner "root"
     group "root"
     mode 0644
-    variables :unit => unit, :speed => speed
-    notifies :run, resources(:execute => "update-grub")
+    variables :units => units, :entry => grub_entry
+    notifies :run, "execute[update-grub]"
   end
 end
 
@@ -142,17 +152,340 @@ template "/etc/initramfs-tools/conf.d/mdadm" do
   notifies :run, "execute[update-initramfs]"
 end
 
-if node[:kernel][:modules].include?("mpt2sas")
-  package "sas2ircu"
-  package "sas2ircu-status"
+package "haveged"
+service "haveged" do
+  action [:enable, :start]
+end
+
+if node[:kernel][:modules].include?("ipmi_si")
+  package "ipmitool"
+end
+
+if node[:lsb][:release].to_f >= 12.10
+  package "irqbalance"
+
+  template "/etc/default/irqbalance" do
+    source "irqbalance.erb"
+    owner "root"
+    group "root"
+    mode 0644
+  end
+
+  service "irqbalance" do
+    action [:start, :enable]
+    supports :status => false, :restart => true, :reload => false
+    subscribes :restart, "template[/etc/default/irqbalance]"
+  end
+end
+
+tools_packages = []
+status_packages = {}
+
+node[:kernel][:modules].each_key do |modname|
+  case modname
+  when "cciss"
+    tools_packages << "hpssacli"
+    status_packages["cciss-vol-status"] ||= []
+  when "hpsa"
+    tools_packages << "hpssacli"
+    status_packages["cciss-vol-status"] ||= []
+  when "mptsas"
+    tools_packages << "lsiutil"
+    # status_packages["mpt-status"] ||= []
+  when "mpt2sas", "mpt3sas"
+    tools_packages << "sas2ircu"
+    status_packages["sas2ircu-status"] ||= []
+  when "megaraid_mm"
+    tools_packages << "megactl"
+    status_packages["megaraid-status"] ||= []
+  when "megaraid_sas"
+    tools_packages << "megacli"
+    status_packages["megaclisas-status"] ||= []
+  when "aacraid"
+    tools_packages << "arcconf"
+    status_packages["aacraid-status"] ||= []
+  when "arcmsr"
+    tools_packages << "areca"
+  end
 end
 
-if node[:kernel][:modules].include?("megaraid_sas")
-  package "megacli"
-  package "megaclisas-status"
+node[:block_device].each do |name, attributes|
+  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
 
-if node[:kernel][:modules].include?("aacraid")
-  package "arcconf"
-  package "aacraid-status"
+%w(hpssacli lsiutil sas2ircu megactl megacli arcconf).each do |tools_package|
+  if tools_packages.include?(tools_package)
+    package tools_package
+  else
+    package tools_package do
+      action :purge
+    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
+
+    template "/etc/default/#{status_package}d" do
+      source "raid.default.erb"
+      owner "root"
+      group "root"
+      mode 0644
+      variables :devices => status_packages[status_package]
+    end
+
+    service "#{status_package}d" do
+      action [:start, :enable]
+      supports :status => false, :restart => true, :reload => false
+      subscribes :restart, "template[/etc/default/#{status_package}d]"
+    end
+  else
+    package status_package do
+      action :purge
+    end
+
+    file "/etc/default/#{status_package}d" do
+      action :delete
+    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")
+
+    line
+  end
+
+  file "/etc/mdadm/mdadm.conf" do
+    owner "root"
+    group "root"
+    mode 0644
+    content mdadm_conf
+  end
+
+  service "mdadm" do
+    action :nothing
+    subscribes :restart, "file[/etc/mdadm/mdadm.conf]"
+  end
+end
+
+template "/etc/modules" do
+  source "modules.erb"
+  owner "root"
+  group "root"
+  mode 0644
+end
+
+if node[:lsb][:release].to_f <= 12.10
+  service "module-init-tools" do
+    provider Chef::Provider::Service::Upstart
+    action :nothing
+    subscribes :start, "template[/etc/modules]"
+  end
+else
+  service "kmod" do
+    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
+end
+
+if node[:hardware][:watchdog]
+  package "watchdog"
+
+  template "/etc/default/watchdog" do
+    source "watchdog.erb"
+    owner "root"
+    group "root"
+    mode 0644
+    variables :module => node[:hardware][:watchdog]
+  end
+
+  service "watchdog" do
+    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"
+    user "root"
+    group "root"
+  end
+
+  template "/etc/sensors.d/chef.conf" do
+    source "sensors.conf.erb"
+    owner "root"
+    group "root"
+    mode 0644
+    notifies :run, "execute[/etc/sensors.d/chef.conf]"
+  end
 end