Add option to force hp-health downgrade

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

diff --git a/cookbooks/hardware/recipes/default.rb b/cookbooks/hardware/recipes/default.rb
index 3169c94a9de9645b3ec426fd4aa8a095d2ca655d..13c1739b5a75de1d3ff5459aaac385e880e5c74c 100644 (file)
--- a/cookbooks/hardware/recipes/default.rb
+++ b/cookbooks/hardware/recipes/default.rb
@@ -20,6 +20,10 @@
 include_recipe "tools"
 include_recipe "munin"
 
 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"
 case node[:cpu][:"0"][:vendor_id]
 when "GenuineIntel"
   package "intel-microcode"


@@ -46,78 +50,102 @@ else
   product = "Unknown"
 end
 
   product = "Unknown"
 end
 

+units = []
+
+if node[:roles].include?("bytemark") || node[:roles].include?("exonetric")
+  units << "0"
+end
+

 case manufacturer
 when "HP"
   package "hponcfg"
 case manufacturer
 when "HP"
   package "hponcfg"

-  package "hp-health"
-  unit = "1"
-  speed = "115200"
+
+  # Downgrade hp-health to 10.0.0.1.3-4. as 10.40-1815.49 has issues with reliable startup
+  package "hp-health" do
+    action :install
+    version "10.0.0.1.3-4."
+    options "--allow-downgrades"
+    notifies :restart, "service[hp-health]"
+  end
+
+  service "hp-health" do
+    action [:enable, :start]
+    supports :status => true, :restart => true, :reload => true
+  end
+
+  units << "1"

 when "TYAN"
 when "TYAN"

-  unit = "0"
-  speed = "115200"
+  units << "0"

 when "TYAN Computer Corporation"
 when "TYAN Computer Corporation"

-  unit = "0"
-  speed = "115200"
+  units << "0"

 when "Supermicro"
   case product
 when "Supermicro"
   case product

-  when "H8DGU", "X9SCD", "X7DBU", "X7DW3", "X9DR7/E-(J)LN4F", "X9DR3-F", "X9DRW"
-    unit = "1"
-    speed = "115200"
+  when "H8DGU", "X9SCD", "X7DBU", "X7DW3", "X9DR7/E-(J)LN4F", "X9DR3-F", "X9DRW", "SYS-2028U-TN24R4T+"
+    units << "1"

   else
   else

-    unit = "0"
-    speed = "115200"
+    units << "0"

   end
 when "IBM"
   end
 when "IBM"

-  unit = "0"
-  speed = "115200"
+  units << "0"

 end
 
 end
 

-if manufacturer == "HP" && 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"
+# Remove legacy HP G4 support which breaks modern hp-health 10.4
+if manufacturer == "HP"
+  %w(/opt/hp/hp-health/bin/hpasmd /usr/lib/libhpasmintrfc.so.3.0 %/usr/lib/libhpasmintrfc.so.3 /usr/lib/libhpasmintrfc.so).each do |filename|
+    file filename do
+      action :delete
+    end

   end
 
   end
 

-  link "/opt/hp/hp-health/bin/hpasmd" do
-    to "/opt/hp/hp-legacy/hpasmd"
+  directory "/opt/hp/hp-legacy" do
+    action :delete
+    recursive true

   end
   end

+end

 
 

-  link "/usr/lib/libhpasmintrfc.so.3.0" do
-    to "/opt/hp/hp-legacy/libhpasmintrfc.so.3.0"
-  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 "/usr/lib/libhpasmintrfc.so.3" do
-    to "libhpasmintrfc.so.3.0"
-  end
+    template "/etc/init/ttyS#{unit}.conf" do
+      source "tty.conf.erb"
+      owner "root"
+      group "root"
+      mode 0o644
+      variables :unit => unit
+    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
 
   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, "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")
 end
 
 if File.exist?("/etc/default/grub")


@@ -130,8 +158,8 @@ if File.exist?("/etc/default/grub")
     source "grub.erb"
     owner "root"
     group "root"
     source "grub.erb"
     owner "root"
     group "root"

-    mode 0644
-    variables :unit => unit, :speed => speed
+    mode 0o644
+    variables :units => units, :entry => grub_entry

     notifies :run, "execute[update-grub]"
   end
 end
     notifies :run, "execute[update-grub]"
   end
 end


@@ -147,7 +175,7 @@ template "/etc/initramfs-tools/conf.d/mdadm" do
   source "initramfs-mdadm.erb"
   owner "root"
   group "root"
   source "initramfs-mdadm.erb"
   owner "root"
   group "root"

-  mode 0644
+  mode 0o644

   notifies :run, "execute[update-initramfs]"
 end
 
   notifies :run, "execute[update-initramfs]"
 end
 


@@ -167,7 +195,7 @@ if node[:lsb][:release].to_f >= 12.10
     source "irqbalance.erb"
     owner "root"
     group "root"
     source "irqbalance.erb"
     owner "root"
     group "root"

-    mode 0644
+    mode 0o644

   end
 
   service "irqbalance" do
   end
 
   service "irqbalance" do


@@ -183,15 +211,15 @@ status_packages = {}
 node[:kernel][:modules].each_key do |modname|
   case modname
   when "cciss"
 node[:kernel][:modules].each_key do |modname|
   case modname
   when "cciss"

-    tools_packages << "hpacucli"
+    tools_packages << "hpssacli"

     status_packages["cciss-vol-status"] ||= []
   when "hpsa"
     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"] ||= []
     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"
     tools_packages << "sas2ircu"
     status_packages["sas2ircu-status"] ||= []
   when "megaraid_mm"


@@ -220,7 +248,7 @@ node[:block_device].each do |name, attributes|
   end
 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
   if tools_packages.include?(tools_package)
     package tools_package
   else


@@ -254,7 +282,7 @@ end
       source "raid.default.erb"
       owner "root"
       group "root"
       source "raid.default.erb"
       owner "root"
       group "root"

-      mode 0644
+      mode 0o644

       variables :devices => status_packages[status_package]
     end
 
       variables :devices => status_packages[status_package]
     end
 


@@ -274,143 +302,127 @@ end
   end
 end
 
   end
 end
 

-disks = []
-
-node[:block_device].each do |name, attributes|
-  disks << { :device => name } if attributes[:vendor] == "ATA"
-end
-
-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 /
-    end
-  end
-end
-
-if status_packages["megaclisas-status"]
-  controller = 0
+disks = if node[:hardware][:disk]
+          node[:hardware][:disk][:disks]
+        else
+          []
+        end

 
 

-  Dir.glob("/sys/class/scsi_host/host*") do |host|
-    driver = File.new("#{host}/proc_name").read.chomp
+intel_ssds = disks.select { |d| d[:vendor] == "INTEL" && d[:model] =~ /^SSD/ }

 
 

-    next unless driver == "megaraid_sas"
+nvmes = if node[:hardware][:pci]
+          node[:hardware][:pci].values.select { |pci| pci[:driver] == "nvme" }
+        else
+          []
+        end

 
 

-    bus = host.sub("/sys/class/scsi_host/host", "")
-    device = File.basename(Dir.glob("/sys/bus/scsi/devices/#{bus}:*/scsi_generic/*").first)
+intel_nvmes = nvmes.select { |pci| pci[:vendor_name] == "Intel Corporation" }

 
 

-    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]+)$/
+if !intel_ssds.empty? || !intel_nvmes.empty?
+  package "unzip"
+  package "alien"

 
 

-      disks.pop if line =~ /^Firmware state: Hotspare, Spun down$/
-    end
-
-    controller += 1
+  remote_file "#{Chef::Config[:file_cache_path]}/DataCenterTool_3_0_0_Linux.zip" do
+    source "https://downloadmirror.intel.com/23931/eng/DataCenterTool_3_0_0_Linux.zip"

   end
   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
+  execute "unzip-DataCenterTool" do
+    command "unzip DataCenterTool_3_0_0_Linux.zip isdct-3.0.0.400-15.x86_64.rpm"
+    cwd Chef::Config[:file_cache_path]
+    user "root"
+    group "root"
+    not_if { File.exist?("#{Chef::Config[:file_cache_path]}/isdct-3.0.0.400-15.x86_64.rpm") }

   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"
-
-    bus = host.sub("/sys/class/scsi_host/host", "")
+  execute "alien-isdct" do
+    command "alien --to-deb isdct-3.0.0.400-15.x86_64.rpm"
+    cwd Chef::Config[:file_cache_path]
+    user "root"
+    group "root"
+    not_if { File.exist?("#{Chef::Config[:file_cache_path]}/isdct_3.0.0.400-16_amd64.deb") }
+  end

 
 

-    Dir.glob("/sys/bus/scsi/devices/#{bus}:0:*/scsi_generic/*").each do |sg|
-      disks << { :device => File.basename(sg) }
-    end
+  dpkg_package "isdct" do
+    source "#{Chef::Config[:file_cache_path]}/isdct_3.0.0.400-16_amd64.deb"

   end
 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"
+disks = disks.map do |disk|
+  next if disk[:state] == "spun_down"

 
 

-    bus = host.sub("/sys/class/scsi_host/host", "")
+  if disk[:smart_device]
+    controller = node[:hardware][:disk][:controllers][disk[:controller]]
+    device = controller[:device].sub("/dev/", "")
+    smart = disk[:smart_device]

 
 

-    Dir.glob("/sys/bus/scsi/devices/#{bus}:1:*/scsi_generic/*").each do |sg|
-      disks << { :device => File.basename(sg) }
+    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
     end

+  elsif disk[:device]
+    device = disk[:device].sub("/dev/", "")
+    munin = device

   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]
+  next if device.nil?

 
 

-      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
+  Hash[
+    :device => device,
+    :smart => smart,
+    :munin => munin,
+    :hddtemp => munin.tr("-:", "_")
+  ]

 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
+smartd_service = if node[:lsb][:release].to_f >= 16.04
+                   "smartd"
+                 else
+                   "smartmontools"
+                 end

 
 

-  if disk[:id]
-    disk[:munin] = "#{disk[:munin]}-#{disk[:id].to_s.tr('/', ':')}"
-  end
-
-  disk[:hddtemp] = disk[:munin].tr("-:", "_")
-end
+disks = disks.compact

 
 if disks.count > 0
   package "smartmontools"
 
 
 if disks.count > 0
   package "smartmontools"
 

+  template "/usr/local/bin/smartd-mailer" do
+    source "smartd-mailer.erb"
+    owner "root"
+    group "root"
+    mode 0o755
+  end
+

   template "/etc/smartd.conf" do
     source "smartd.conf.erb"
     owner "root"
     group "root"
   template "/etc/smartd.conf" do
     source "smartd.conf.erb"
     owner "root"
     group "root"

-    mode 0644
+    mode 0o644

     variables :disks => disks
     variables :disks => disks

-    notifies :reload, "service[smartmontools]"

   end
 
   template "/etc/default/smartmontools" do
     source "smartmontools.erb"
     owner "root"
     group "root"
   end
 
   template "/etc/default/smartmontools" do
     source "smartmontools.erb"
     owner "root"
     group "root"

-    mode 0644
-    notifies :restart, "service[smartmontools]"
+    mode 0o644

   end
 
   end
 

-  service "smartmontools" do
+  service smartd_service do

     action [:enable, :start]
     action [:enable, :start]

-    supports :status => true, :restart => true, :reload => true
+    subscribes :reload, "template[/etc/smartd.conf]"
+    subscribes :restart, "template[/etc/default/smartmontools]"

   end
 
   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_"
   disks.each do |disk|
     munin_plugin "smart_#{disk[:munin]}" do
       target "smart_"


@@ -418,18 +430,21 @@ if disks.count > 0
       conf_variables :disk => disk
     end
   end
       conf_variables :disk => disk
     end
   end

+else
+  service smartd_service 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
     conf "munin.hddtemp.erb"
     conf_variables :disks => disks
   end
 else

-  service "smartmontools" do
-    action [:stop, :disable]
-  end
-

   munin_plugin "hddtemp_smartctl" do
     action :delete
   munin_plugin "hddtemp_smartctl" do
     action :delete

+    conf "munin.hddtemp.erb"

   end
 end
 
   end
 end
 


@@ -452,7 +467,7 @@ if File.exist?("/etc/mdadm/mdadm.conf")
   file "/etc/mdadm/mdadm.conf" do
     owner "root"
     group "root"
   file "/etc/mdadm/mdadm.conf" do
     owner "root"
     group "root"

-    mode 0644
+    mode 0o644

     content mdadm_conf
   end
 
     content mdadm_conf
   end
 


@@ -466,7 +481,7 @@ template "/etc/modules" do
   source "modules.erb"
   owner "root"
   group "root"
   source "modules.erb"
   owner "root"
   group "root"

-  mode 0644
+  mode 0o644

 end
 
 if node[:lsb][:release].to_f <= 12.10
 end
 
 if node[:lsb][:release].to_f <= 12.10


@@ -477,7 +492,11 @@ if node[:lsb][:release].to_f <= 12.10
   end
 else
   service "kmod" do
   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
     action :nothing
     subscribes :start, "template[/etc/modules]"
   end


@@ -490,7 +509,7 @@ if node[:hardware][:watchdog]
     source "watchdog.erb"
     owner "root"
     group "root"
     source "watchdog.erb"
     owner "root"
     group "root"

-    mode 0644
+    mode 0o644

     variables :module => node[:hardware][:watchdog]
   end
 
     variables :module => node[:hardware][:watchdog]
   end
 


@@ -502,6 +521,30 @@ end
 unless Dir.glob("/sys/class/hwmon/hwmon*").empty?
   package "lm-sensors"
 
 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"
   execute "/etc/sensors.d/chef.conf" do
     action :nothing
     command "/usr/bin/sensors -s"


@@ -513,7 +556,7 @@ unless Dir.glob("/sys/class/hwmon/hwmon*").empty?
     source "sensors.conf.erb"
     owner "root"
     group "root"
     source "sensors.conf.erb"
     owner "root"
     group "root"

-    mode 0644
+    mode 0o644

     notifies :run, "execute[/etc/sensors.d/chef.conf]"
   end
 end
     notifies :run, "execute[/etc/sensors.d/chef.conf]"
   end
 end