hardware: do not fail if node[:hardware][:pci] is undefined (tests)

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

diff --git a/cookbooks/hardware/recipes/default.rb b/cookbooks/hardware/recipes/default.rb
index 3d1d8aa1986d2ea53842eb7a3c141bdda550883f..4e676f3879712720e3ac7211c2e47ae6dd187b59 100644 (file)
--- a/cookbooks/hardware/recipes/default.rb
+++ b/cookbooks/hardware/recipes/default.rb
@@ -17,6 +17,7 @@
 # limitations under the License.
 #
 
+include_recipe "apt"
 include_recipe "git"
 include_recipe "munin"
 include_recipe "prometheus"
@@ -27,11 +28,17 @@ ohai_plugin "hardware" do
   template "ohai.rb.erb"
 end
 
-case node[:cpu][:"0"][:vendor_id]
-when "GenuineIntel"
-  package "intel-microcode"
-when "AuthenticAMD"
-  package "amd64-microcode"
+if platform?("debian")
+  package "firmware-linux"
+end
+
+if node[:cpu] && node[:cpu][:"0"] && node[:cpu][:"0"][:vendor_id]
+  case node[:cpu][:"0"][:vendor_id]
+  when "GenuineIntel"
+    package "intel-microcode"
+  when "AuthenticAMD"
+    package "amd64-microcode"
+  end
 end
 
 if node[:dmi] && node[:dmi][:system]
@@ -77,13 +84,13 @@ when "HP", "HPE"
   package "hp-health" do
     action :install
     notifies :restart, "service[hp-health]"
-    only_if { node[:lsb][:release].to_f < 22.04 }
+    only_if { platform?("ubuntu") && node[:lsb][:release].to_f < 22.04 }
   end
 
   service "hp-health" do
     action [:enable, :start]
     supports :status => true, :restart => true
-    only_if { node[:lsb][:release].to_f < 22.04 }
+    only_if { platform?("ubuntu") && node[:lsb][:release].to_f < 22.04 }
   end
 
   if product.end_with?("Gen8", "Gen9")
@@ -186,6 +193,8 @@ if File.exist?("/etc/default/grub")
   end
 end
 
+package "initramfs-tools"
+
 execute "update-initramfs" do
   action :nothing
   command "update-initramfs -u -k all"
@@ -201,9 +210,24 @@ template "/etc/initramfs-tools/conf.d/mdadm" do
   notifies :run, "execute[update-initramfs]"
 end
 
-package "haveged"
-service "haveged" do
-  action [:enable, :start]
+# haveged is only required on older kernels
+# /dev/random is not blocking anymore in 5.15+
+if Chef::Util.compare_versions(node[:kernel][:release], [5, 15]).negative?
+  package "haveged"
+  service "haveged" do
+    action [:enable, :start]
+  end
+else
+  service "haveged" do
+    action [:stop, :disable]
+  end
+  package "haveged" do
+    action :remove
+  end
+end
+
+watchdog_module = %w[hpwdt sp5100_tco].find do |module_name|
+  node[:hardware][:pci]&.any? { |_, pci| pci[:modules]&.any?(module_name) }
 end
 
 if node[:kernel][:modules].include?("ipmi_si")
@@ -226,6 +250,8 @@ if node[:kernel][:modules].include?("ipmi_si")
     options "--config.file=/etc/prometheus/ipmi_local.yml"
     subscribes :restart, "template[/etc/prometheus/ipmi_local.yml]"
   end
+
+  watchdog_module ||= "ipmi_watchdog"
 end
 
 package "irqbalance"
@@ -336,63 +362,32 @@ end
 
 include_recipe "apt::hwraid" unless status_packages.empty?
 
-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 "755"
-    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"
-    nice 10
-    private_tmp true
-    protect_system "full"
-    protect_home true
-    no_new_privileges true
-    notifies :restart, "service[cciss-vol-statusd]"
-  end
-else
-  systemd_service "cciss-vol-statusd" do
-    action :delete
-  end
-
-  template "/usr/local/bin/cciss-vol-statusd" do
-    action :delete
-  end
-end
-
 %w[cciss-vol-status mpt-status sas2ircu-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 "644"
-      variables :devices => status_packages[status_package]
+    service "#{status_package}d" do
+      action [:stop, :disable]
     end
 
-    service "#{status_package}d" do
-      action [:start, :enable]
-      supports :status => false, :restart => true, :reload => false
-      subscribes :restart, "template[/etc/default/#{status_package}d]"
+    file "/etc/default/#{status_package}d" do
+      action :delete
     end
   else
     package status_package do
       action :purge
     end
-
-    file "/etc/default/#{status_package}d" do
-      action :delete
-    end
   end
 end
 
+systemd_service "cciss-vol-statusd" do
+  action :delete
+end
+
+template "/usr/local/bin/cciss-vol-statusd" do
+  action :delete
+end
+
 disks = if node[:hardware][:disk]
           node[:hardware][:disk][:disks]
         else
@@ -475,7 +470,7 @@ disks = disks.map do |disk|
     munin = device
   end
 
-  next if device.nil?
+  next if device.nil? || munin.nil?
 
   Hash[
     :device => device,
@@ -536,7 +531,7 @@ if disks.count.positive?
   prometheus_collector "smart" do
     interval "15m"
     user "root"
-    capability_bounding_set %w[CAP_SYS_ADMIN CAP_SYS_RAWIO]
+    capability_bounding_set %w[CAP_DAC_OVERRIDE CAP_SYS_ADMIN CAP_SYS_RAWIO]
     private_devices false
     private_users false
     protect_clock false
@@ -620,19 +615,30 @@ node[:hardware][:blacklisted_modules].each do |module_name|
   end
 end
 
-if node[:hardware][:watchdog]
-  package "watchdog"
+if watchdog_module
+  kernel_module watchdog_module do
+    action :install
+  end
+
+  execute "systemctl-reload" do
+    action :nothing
+    command "systemctl daemon-reload"
+    user "root"
+    group "root"
+  end
 
-  template "/etc/default/watchdog" do
-    source "watchdog.erb"
+  directory "/etc/systemd/system.conf.d" do
     owner "root"
     group "root"
-    mode "644"
-    variables :module => node[:hardware][:watchdog]
+    mode "755"
   end
 
-  service "watchdog" do
-    action [:enable, :start]
+  template "/etc/systemd/system.conf.d/watchdog.conf" do
+    source "watchdog.conf.erb"
+    owner "root"
+    group "root"
+    mode "644"
+    notifies :run, "execute[systemctl-reload]"
   end
 end
 
@@ -700,7 +706,7 @@ prometheus_collector "ohai" do
   interval "15m"
   user "root"
   proc_subset "all"
-  capability_bounding_set "CAP_SYS_ADMIN"
+  capability_bounding_set %w[CAP_DAC_OVERRIDE CAP_SYS_ADMIN CAP_SYS_RAWIO]
   private_devices false
   private_users false
   protect_clock false