Increase size of /dev/shm on fume

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

diff --git a/cookbooks/hardware/recipes/default.rb b/cookbooks/hardware/recipes/default.rb
index 564e530187d3e25c9e3034c779916ef71d89864c..c8898f405d2ef818bf34b603d972b75e3c37e918 100644 (file)
--- a/cookbooks/hardware/recipes/default.rb
+++ b/cookbooks/hardware/recipes/default.rb
@@ -65,7 +65,7 @@ when "HP"
     supports :status => true, :restart => true
   end
 
-  if product.end_with?("Gen8")
+  if product.end_with?("Gen8", "Gen9")
     package "hp-ams" do
       action :install
       notifies :restart, "service[hp-ams]"
@@ -84,7 +84,7 @@ when "TYAN Computer Corporation"
   units << "0"
 when "Supermicro"
   case product
-  when "H8DGU", "X9SCD", "X7DBU", "X7DW3", "X9DR7/E-(J)LN4F", "X9DR3-F", "X9DRW", "SYS-1028U-TN10RT+", "SYS-2028U-TN24R4T+", "Super Server"
+  when "H8DGU", "X9SCD", "X7DBU", "X7DW3", "X9DR7/E-(J)LN4F", "X9DR3-F", "X9DRW", "SYS-1028U-TN10RT+", "SYS-2028U-TN24R4T+", "SYS-1029P-WTRT", "Super Server"
     units << "1"
   else
     units << "0"
@@ -94,29 +94,8 @@ when "IBM"
 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
-
-    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
 
@@ -130,6 +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-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
@@ -198,24 +178,6 @@ service "lldpd" do
   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 = {}
 
@@ -285,6 +247,26 @@ else
   end
 end
 
+if status_packages.include?("cciss-vol-status")
+  template "/usr/local/bin/cciss-vol-statusd" do
+    source "cciss-vol-statusd.erb"
+    owner "root"
+    group "root"
+    mode 0o755
+    notifies :restart, "service[cciss-vol-statusd]"
+  end
+
+  systemd_service "cciss-vol-statusd" do
+    description "Check cciss_vol_status values in the background"
+    exec_start "/usr/local/bin/cciss-vol-statusd"
+    private_tmp true
+    protect_system "full"
+    protect_home true
+    no_new_privileges true
+    notifies :restart, "service[cciss-vol-statusd]"
+  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
@@ -332,26 +314,28 @@ intel_nvmes = nvmes.select { |pci| pci[:vendor_name] == "Intel Corporation" }
 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
 
-  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"
-    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
-    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|
-  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]]
@@ -369,6 +353,9 @@ disks = disks.map do |disk|
         munin = "#{device}-#{Regexp.last_match(1)}:#{Regexp.last_match(2)}"
       end
     end
+  elsif disk[:device] =~ %r{^/dev/(nvme\d+)n\d+$}
+    device = Regexp.last_match(1)
+    munin = device
   elsif disk[:device]
     device = disk[:device].sub("/dev/", "")
     munin = device
@@ -384,13 +371,7 @@ disks = disks.map do |disk|
   ]
 end
 
-smartd_service = if node[:lsb][:release].to_f >= 16.04
-                   "smartd"
-                 else
-                   "smartmontools"
-                 end
-
-disks = disks.compact
+disks = disks.compact.uniq
 
 if disks.count.positive?
   package "smartmontools"
@@ -417,7 +398,7 @@ if disks.count.positive?
     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]"
@@ -437,7 +418,7 @@ if disks.count.positive?
     end
   end
 else
-  service smartd_service do
+  service "smartd" do
     action [:stop, :disable]
   end
 end
@@ -460,6 +441,7 @@ plugins = Dir.glob("/etc/munin/plugins/smart_*").map { |p| File.basename(p) } -
 plugins.each do |plugin|
   munin_plugin plugin do
     action :delete
+    conf "munin.smart.erb"
   end
 end
 
@@ -553,3 +535,12 @@ unless Dir.glob("/sys/class/hwmon/hwmon*").empty?
     notifies :run, "execute[/etc/sensors.d/chef.conf]"
   end
 end
+
+if node[:hardware][:shm_size]
+  mount "/dev/shm" do
+    action [:mount, :enable]
+    device "tmpfs"
+    fstype "tmpfs"
+    options "rw,nosuid,nodev,size=#{node[:hardware][:shm_size]}"
+  end
+end