Fix munin alert thresholds for G9 machines

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

diff --git a/cookbooks/hardware/recipes/default.rb b/cookbooks/hardware/recipes/default.rb
index 837c8ce23e8509440c65bcfc21573c18d3c78c29..2fa629adfa09c5a674a07aeae25e6c8466add21e 100644 (file)
--- a/cookbooks/hardware/recipes/default.rb
+++ b/cookbooks/hardware/recipes/default.rb
@@ -65,6 +65,18 @@ when "HP"
     supports :status => true, :restart => true
   end
 
     supports :status => true, :restart => true
   end
 

+  if product.end_with?("Gen8", "Gen9")
+    package "hp-ams" do
+      action :install
+      notifies :restart, "service[hp-ams]"
+    end
+
+    service "hp-ams" do
+      action [:enable, :start]
+      supports :status => true, :restart => true
+    end
+  end
+

   units << "1"
 when "TYAN"
   units << "0"
   units << "1"
 when "TYAN"
   units << "0"


@@ -72,7 +84,7 @@ when "TYAN Computer Corporation"
   units << "0"
 when "Supermicro"
   case product
   units << "0"
 when "Supermicro"
   case product

-  when "H8DGU", "X9SCD", "X7DBU", "X7DW3", "X9DR7/E-(J)LN4F", "X9DR3-F", "X9DRW", "SYS-2028U-TN24R4T+", "Super Server"
+  when "H8DGU", "X9SCD", "X7DBU", "X7DW3", "X9DR7/E-(J)LN4F", "X9DR3-F", "X9DRW", "SYS-1028U-TN10RT+", "SYS-2028U-TN24R4T+", "Super Server"

     units << "1"
   else
     units << "0"
     units << "1"
   else
     units << "0"


@@ -81,45 +93,9 @@ when "IBM"
   units << "0"
 end
 
   units << "0"
 end
 

-# 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
-      manage_symlink_source false
-    end
-  end
-
-  directory "/opt/hp/hp-legacy" do
-    action :delete
-    recursive true
-  end
-end
-

 units.sort.uniq.each do |unit|
 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
-
-    template "/etc/init/ttyS#{unit}.conf" do
-      source "tty.conf.erb"
-      owner "root"
-      group "root"
-      mode 0o644
-      variables :unit => unit
-    end
-
-    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
+  service "serial-getty@ttyS#{unit}" do
+    action [:enable, :start]

   end
 end
 
   end
 end
 


@@ -133,6 +109,7 @@ if node[:hardware][:grub][:kernel]
   package "linux-image-#{kernel_version}-generic"
   package "linux-image-extra-#{kernel_version}-generic"
   package "linux-headers-#{kernel_version}-generic"
   package "linux-image-#{kernel_version}-generic"
   package "linux-image-extra-#{kernel_version}-generic"
   package "linux-headers-#{kernel_version}-generic"

+  package "linux-tools-#{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
 
   boot_device = IO.popen(["df", "/boot"]).readlines.last.split.first
   boot_uuid = IO.popen(["blkid", "-o", "value", "-s", "UUID", boot_device]).readlines.first.chomp


@@ -201,24 +178,6 @@ service "lldpd" do
   supports :status => true, :restart => true, :reload => true
 end
 
   supports :status => true, :restart => true, :reload => true
 end
 

-if node[:hardware][:mcelog][:enabled]
-  package "mcelog"
-
-  %w[bus cache dimm iomca page socket-memory unknown].each do |trigger|
-    template "/etc/mcelog/#{trigger}-error-trigger.local" do
-      source "mcelog-trigger.erb"
-      owner "root"
-      group "root"
-      mode 0o755
-    end
-  end
-
-  service "mcelog" do
-    action [:start, :enable]
-    supports :status => true, :restart => true, :reload => false
-  end
-end
-

 tools_packages = []
 status_packages = {}
 
 tools_packages = []
 status_packages = {}
 


@@ -335,39 +294,44 @@ intel_nvmes = nvmes.select { |pci| pci[:vendor_name] == "Intel Corporation" }
 if !intel_ssds.empty? || !intel_nvmes.empty?
   package "unzip"
 
 if !intel_ssds.empty? || !intel_nvmes.empty?
   package "unzip"
 

-  remote_file "#{Chef::Config[:file_cache_path]}/Intel_SSD_Data_Center_Tool_3.0.13_Linux.zip" do
-    source "https://downloadmirror.intel.com/27863/eng/Intel_SSD_Data_Center_Tool_3.0.13_Linux.zip"
+  intel_ssd_tool_version = "3.0.19"
+
+  remote_file "#{Chef::Config[:file_cache_path]}/Intel_SSD_Data_Center_Tool_#{intel_ssd_tool_version}_Linux.zip" do
+    source "https://downloadmirror.intel.com/28639/eng/Intel_SSD_Data_Center_Tool_#{intel_ssd_tool_version}_Linux.zip"

   end
 
   end
 

-  execute "#{Chef::Config[:file_cache_path]}/Intel_SSD_Data_Center_Tool_3.0.13_Linux.zip" do
-    command "unzip Intel_SSD_Data_Center_Tool_3.0.13_Linux.zip isdct_3.0.13.400-17_amd64.deb"
+  execute "#{Chef::Config[:file_cache_path]}/Intel_SSD_Data_Center_Tool_#{intel_ssd_tool_version}_Linux.zip" do
+    command "unzip Intel_SSD_Data_Center_Tool_#{intel_ssd_tool_version}_Linux.zip isdct_#{intel_ssd_tool_version}-1_amd64.deb"

     cwd Chef::Config[:file_cache_path]
     user "root"
     group "root"
     cwd Chef::Config[:file_cache_path]
     user "root"
     group "root"

-    not_if { File.exist?("#{Chef::Config[:file_cache_path]}/isdct_3.0.13.400-17_amd64.deb") }
+    not_if { File.exist?("#{Chef::Config[:file_cache_path]}/isdct_#{intel_ssd_tool_version}-1_amd64.deb") }

   end
 
   dpkg_package "isdct" do
   end
 
   dpkg_package "isdct" do

-    version "3.0.13.400-17"
-    source "#{Chef::Config[:file_cache_path]}/isdct_3.0.13.400-17_amd64.deb"
+    version "#{intel_ssd_tool_version}-1"
+    source "#{Chef::Config[:file_cache_path]}/isdct_#{intel_ssd_tool_version}-1_amd64.deb"

   end
 end
 
 disks = disks.map do |disk|
   end
 end
 
 disks = disks.map do |disk|

-  next if disk[:state] == "spun_down"
+  next if disk[:state] == "spun_down" || %w[unconfigured failed].any?(disk[:status])

 
   if disk[:smart_device]
     controller = node[:hardware][:disk][:controllers][disk[:controller]]
 
   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+)$}
-      munin = "#{device}-#{Regexp.last_match(1)}:#{Regexp.last_match(2)}"
+
+    if controller && controller[:device]
+      device = controller[:device].sub("/dev/", "")
+      smart = disk[:smart_device]
+
+      if device.start_with?("cciss/") && smart =~ /^cciss,(\d+)$/
+        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

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


@@ -384,12 +348,6 @@ disks = disks.map do |disk|
   ]
 end
 
   ]
 end
 

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

 disks = disks.compact
 
 if disks.count.positive?
 disks = disks.compact
 
 if disks.count.positive?


@@ -417,7 +375,7 @@ if disks.count.positive?
     mode 0o644
   end
 
     mode 0o644
   end
 

-  service smartd_service do
+  service "smartd" do

     action [:enable, :start]
     subscribes :reload, "template[/etc/smartd.conf]"
     subscribes :restart, "template[/etc/default/smartmontools]"
     action [:enable, :start]
     subscribes :reload, "template[/etc/smartd.conf]"
     subscribes :restart, "template[/etc/default/smartmontools]"


@@ -437,7 +395,7 @@ if disks.count.positive?
     end
   end
 else
     end
   end
 else

-  service smartd_service do
+  service "smartd" do

     action [:stop, :disable]
   end
 end
     action [:stop, :disable]
   end
 end


@@ -460,6 +418,7 @@ plugins = Dir.glob("/etc/munin/plugins/smart_*").map { |p| File.basename(p) } -
 plugins.each do |plugin|
   munin_plugin plugin do
     action :delete
 plugins.each do |plugin|
   munin_plugin plugin do
     action :delete

+    conf "munin.smart.erb"

   end
 end
 
   end
 end