Bring web site back online using karm

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

diff --git a/cookbooks/hardware/recipes/default.rb b/cookbooks/hardware/recipes/default.rb
index aa0c889915af16eae7c9a006dcda70a221e34e47..9dd3c0a6a7e18f197dbb83b57f347be5f7f92e76 100644 (file)
--- a/cookbooks/hardware/recipes/default.rb
+++ b/cookbooks/hardware/recipes/default.rb
@@ -31,9 +31,7 @@ end
 
 case node[:cpu][:"0"][:vendor_id]
 when "AuthenticAMD"
-  if node[:lsb][:release].to_f >= 14.04
-    package "amd64-microcode"
-  end
+  package "amd64-microcode" if node[:lsb][:release].to_f >= 14.04
 end
 
 if node[:dmi] && node[:dmi][:system]
@@ -59,7 +57,20 @@ end
 case manufacturer
 when "HP"
   package "hponcfg"
-  package "hp-health"
+
+  # 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 "--force-yes"
+    notifies :restart, "service[hp-health]"
+  end
+
+  service "hp-health" do
+    action [:enable, :start]
+    supports :status => true, :restart => true
+  end
+
   units << "1"
 when "TYAN"
   units << "0"
@@ -67,7 +78,7 @@ when "TYAN Computer Corporation"
   units << "0"
 when "Supermicro"
   case product
-  when "H8DGU", "X9SCD", "X7DBU", "X7DW3", "X9DR7/E-(J)LN4F", "X9DR3-F", "X9DRW"
+  when "H8DGU", "X9SCD", "X7DBU", "X7DW3", "X9DR7/E-(J)LN4F", "X9DR3-F", "X9DRW", "SYS-2028U-TN24R4T+", "Super Server"
     units << "1"
   else
     units << "0"
@@ -76,6 +87,20 @@ when "IBM"
   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
+    end
+  end
+
+  directory "/opt/hp/hp-legacy" do
+    action :delete
+    recursive true
+  end
+end
+
 units.sort.uniq.each do |unit|
   if node[:lsb][:release].to_f >= 16.04
     service "serial-getty@ttyS#{unit}" do
@@ -90,7 +115,7 @@ units.sort.uniq.each do |unit|
       source "tty.conf.erb"
       owner "root"
       group "root"
-      mode 0644
+      mode 0o644
       variables :unit => unit
     end
 
@@ -131,7 +156,7 @@ if File.exist?("/etc/default/grub")
     source "grub.erb"
     owner "root"
     group "root"
-    mode 0644
+    mode 0o644
     variables :units => units, :entry => grub_entry
     notifies :run, "execute[update-grub]"
   end
@@ -148,7 +173,7 @@ template "/etc/initramfs-tools/conf.d/mdadm" do
   source "initramfs-mdadm.erb"
   owner "root"
   group "root"
-  mode 0644
+  mode 0o644
   notifies :run, "execute[update-initramfs]"
 end
 
@@ -157,9 +182,7 @@ service "haveged" do
   action [:enable, :start]
 end
 
-if node[:kernel][:modules].include?("ipmi_si")
-  package "ipmitool"
-end
+package "ipmitool" if node[:kernel][:modules].include?("ipmi_si")
 
 if node[:lsb][:release].to_f >= 12.10
   package "irqbalance"
@@ -168,7 +191,7 @@ if node[:lsb][:release].to_f >= 12.10
     source "irqbalance.erb"
     owner "root"
     group "root"
-    mode 0644
+    mode 0o644
   end
 
   service "irqbalance" do
@@ -178,6 +201,13 @@ if node[:lsb][:release].to_f >= 12.10
   end
 end
 
+# Link Layer Discovery Protocol Daemon
+package "lldpd"
+service "lldpd" do
+  action [:start, :enable]
+  supports :status => true, :restart => true, :reload => true
+end
+
 tools_packages = []
 status_packages = {}
 
@@ -255,7 +285,7 @@ end
       source "raid.default.erb"
       owner "root"
       group "root"
-      mode 0644
+      mode 0o644
       variables :devices => status_packages[status_package]
     end
 
@@ -281,6 +311,45 @@ disks = if node[:hardware][:disk]
           []
         end
 
+intel_ssds = disks.select { |d| d[:vendor] == "INTEL" && d[:model] =~ /^SSD/ }
+
+nvmes = if node[:hardware][:pci]
+          node[:hardware][:pci].values.select { |pci| pci[:driver] == "nvme" }
+        else
+          []
+        end
+
+intel_nvmes = nvmes.select { |pci| pci[:vendor_name] == "Intel Corporation" }
+
+if !intel_ssds.empty? || !intel_nvmes.empty?
+  package "unzip"
+  package "alien"
+
+  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
+
+  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
+
+  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
+
+  dpkg_package "isdct" do
+    source "#{Chef::Config[:file_cache_path]}/isdct_3.0.0.400-16_amd64.deb"
+  end
+end
+
 disks = disks.map do |disk|
   next if disk[:state] == "spun_down"
 
@@ -312,6 +381,12 @@ disks = disks.map do |disk|
   ]
 end
 
+smartd_service = if node[:lsb][:release].to_f >= 16.04
+                   "smartd"
+                 else
+                   "smartmontools"
+                 end
+
 disks = disks.compact
 
 if disks.count > 0
@@ -321,29 +396,28 @@ if disks.count > 0
     source "smartd-mailer.erb"
     owner "root"
     group "root"
-    mode 0755
+    mode 0o755
   end
 
   template "/etc/smartd.conf" do
     source "smartd.conf.erb"
     owner "root"
     group "root"
-    mode 0644
+    mode 0o644
     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]"
+    mode 0o644
   end
 
-  service "smartmontools" do
+  service smartd_service do
     action [:enable, :start]
-    supports :status => true, :restart => true, :reload => true
+    subscribes :reload, "template[/etc/smartd.conf]"
+    subscribes :restart, "template[/etc/default/smartmontools]"
   end
 
   # Don't try and do munin monitoring of disks behind
@@ -360,7 +434,7 @@ if disks.count > 0
     end
   end
 else
-  service "smartmontools" do
+  service smartd_service do
     action [:stop, :disable]
   end
 end
@@ -396,7 +470,7 @@ if File.exist?("/etc/mdadm/mdadm.conf")
   file "/etc/mdadm/mdadm.conf" do
     owner "root"
     group "root"
-    mode 0644
+    mode 0o644
     content mdadm_conf
   end
 
@@ -410,7 +484,7 @@ template "/etc/modules" do
   source "modules.erb"
   owner "root"
   group "root"
-  mode 0644
+  mode 0o644
 end
 
 if node[:lsb][:release].to_f <= 12.10
@@ -438,7 +512,7 @@ if node[:hardware][:watchdog]
     source "watchdog.erb"
     owner "root"
     group "root"
-    mode 0644
+    mode 0o644
     variables :module => node[:hardware][:watchdog]
   end
 
@@ -485,7 +559,7 @@ unless Dir.glob("/sys/class/hwmon/hwmon*").empty?
     source "sensors.conf.erb"
     owner "root"
     group "root"
-    mode 0644
+    mode 0o644
     notifies :run, "execute[/etc/sensors.d/chef.conf]"
   end
 end