Allow CAP_DAC_OVERRIDE for the smart collector

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

diff --git a/cookbooks/hardware/recipes/default.rb b/cookbooks/hardware/recipes/default.rb
index 611ebb74d27a5e9b23af89d89a4b43a8cb68443d..48e8f28889d3286069cada8c68ce92bd47a7e213 100644 (file)
--- a/cookbooks/hardware/recipes/default.rb
+++ b/cookbooks/hardware/recipes/default.rb
@@ -17,7 +17,6 @@
 # limitations under the License.
 #
 
-include_recipe "apt"
 include_recipe "git"
 include_recipe "munin"
 include_recipe "prometheus"
@@ -56,9 +55,25 @@ if node[:roles].include?("bytemark") || node[:roles].include?("exonetric") || no
 end
 
 case manufacturer
-when "HP"
+when "HP", "HPE"
+  include_recipe "apt::management-component-pack"
+
   package "hponcfg"
 
+  execute "update-ilo" do
+    action :nothing
+    command "/usr/sbin/hponcfg -f /etc/ilo-defaults.xml"
+    not_if { kitchen? }
+  end
+
+  template "/etc/ilo-defaults.xml" do
+    source "ilo-defaults.xml.erb"
+    owner "root"
+    group "root"
+    mode "644"
+    notifies :run, "execute[update-ilo]"
+  end
+
   package "hp-health" do
     action :install
     notifies :restart, "service[hp-health]"
@@ -81,9 +96,23 @@ when "HP"
       action [:enable, :start]
       supports :status => true, :restart => true
     end
+  elsif product.end_with?("Gen10")
+    package "amsd" do
+      action :install
+      notifies :restart, "service[amsd]"
+    end
+
+    service "amsd" do
+      action [:enable, :start]
+      supports :status => true, :restart => true
+    end
   end
 
-  units << "1"
+  units << if product.end_with?("Gen10")
+             "0"
+           else
+             "1"
+           end
 when "TYAN"
   units << "0"
 when "TYAN Computer Corporation"
@@ -117,6 +146,7 @@ end
 units.sort.uniq.each do |unit|
   service "serial-getty@ttyS#{unit}" do
     action [:enable, :start]
+    not_if { kitchen? }
   end
 end
 
@@ -156,6 +186,8 @@ if File.exist?("/etc/default/grub")
   end
 end
 
+package "initramfs-tools"
+
 execute "update-initramfs" do
   action :nothing
   command "update-initramfs -u -k all"
@@ -189,6 +221,10 @@ if node[:kernel][:modules].include?("ipmi_si")
 
   prometheus_exporter "ipmi" do
     port 9290
+    user "root"
+    private_devices false
+    protect_clock false
+    system_call_filter ["@system-service", "@raw-io"]
     options "--config.file=/etc/prometheus/ipmi_local.yml"
     subscribes :restart, "template[/etc/prometheus/ipmi_local.yml]"
   end
@@ -212,6 +248,20 @@ ohai_plugin "lldp" do
   template "lldp.rb.erb"
 end
 
+package %w[
+  rasdaemon
+  ruby-sqlite3
+]
+
+service "rasdaemon" do
+  action [:enable, :start]
+end
+
+prometheus_exporter "rasdaemon" do
+  port 9797
+  user "root"
+end
+
 tools_packages = []
 status_packages = {}
 
@@ -233,10 +283,7 @@ if node[:virtualization][:role] != "guest" ||
       status_packages["mpt-status"] ||= []
     when "mpt2sas", "mpt3sas"
       tools_packages << "sas2ircu"
-      status_packages["sas2ircu-status"] ||= [] if node[:lsb][:release].to_f < 22.04
-    when "megaraid_mm"
-      tools_packages << "megactl"
-      status_packages["megaraid-status"] ||= []
+      status_packages["sas2ircu-status"] ||= []
     when "megaraid_sas"
       tools_packages << "megacli"
       status_packages["megaclisas-status"] ||= []
@@ -289,6 +336,8 @@ else
   end
 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"
@@ -301,6 +350,7 @@ if status_packages.include?("cciss-vol-status")
   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
@@ -317,7 +367,7 @@ else
   end
 end
 
-%w[cciss-vol-status mpt-status sas2ircu-status megaraid-status megaclisas-status aacraid-status].each do |status_package|
+%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
 
@@ -368,27 +418,27 @@ intel_nvmes = nvmes.select { |pci| pci[:vendor_name] == "Intel Corporation" }
 if !intel_ssds.empty? || !intel_nvmes.empty?
   package "unzip"
 
-  intel_mas_tool_version = "1.10"
-  intel_mas_package_version = "#{intel_mas_tool_version}.155-0"
+  sst_tool_version = "1.3"
+  sst_package_version = "#{sst_tool_version}.208-0"
 
-  remote_file "#{Chef::Config[:file_cache_path]}/Intel_MAS_CLI_Tool_#{intel_mas_tool_version}_Linux.zip" do
-    source "https://downloadmirror.intel.com/646992/Intel_MAS_CLI_Tool_Linux_#{intel_mas_tool_version}-v2.zip"
-  end
+  # remote_file "#{Chef::Config[:file_cache_path]}/SST_CLI_Linux_#{sst_tool_version}.zip" do
+  #   source "https://downloadmirror.intel.com/743764/SST_CLI_Linux_#{sst_tool_version}.zip"
+  # end
 
-  execute "#{Chef::Config[:file_cache_path]}/Intel_MAS_CLI_Tool_#{intel_mas_tool_version}_Linux.zip" do
-    command "unzip Intel_MAS_CLI_Tool_#{intel_mas_tool_version}_Linux.zip intelmas_#{intel_mas_package_version}_amd64.deb"
+  execute "#{Chef::Config[:file_cache_path]}/SST_CLI_Linux_#{sst_tool_version}.zip" do
+    command "unzip SST_CLI_Linux_#{sst_tool_version}.zip sst_#{sst_package_version}_amd64.deb"
     cwd Chef::Config[:file_cache_path]
     user "root"
     group "root"
-    not_if { ::File.exist?("#{Chef::Config[:file_cache_path]}/intelmas_#{intel_mas_package_version}_amd64.deb") }
+    not_if { ::File.exist?("#{Chef::Config[:file_cache_path]}/sst_#{sst_package_version}_amd64.deb") }
   end
 
-  dpkg_package "intelmas" do
-    version "#{intel_mas_package_version}"
-    source "#{Chef::Config[:file_cache_path]}/intelmas_#{intel_mas_package_version}_amd64.deb"
+  dpkg_package "sst" do
+    version "#{sst_package_version}"
+    source "#{Chef::Config[:file_cache_path]}/sst_#{sst_package_version}_amd64.deb"
   end
 
-  dpkg_package "isdct" do
+  dpkg_package "intelmas" do
     action :purge
   end
 end
@@ -487,6 +537,11 @@ if disks.count.positive?
 
   prometheus_collector "smart" do
     interval "15m"
+    user "root"
+    capability_bounding_set %w[CAP_DAC_OVERRIDE CAP_SYS_ADMIN CAP_SYS_RAWIO]
+    private_devices false
+    private_users false
+    protect_clock false
   end
 
   # Don't try and do munin monitoring of disks behind
@@ -544,7 +599,7 @@ if File.exist?("/etc/mdadm/mdadm.conf")
     content mdadm_conf
   end
 
-  service "mdadm" do
+  service "mdmonitor" do
     action :nothing
     subscribes :restart, "file[/etc/mdadm/mdadm.conf]"
   end
@@ -642,3 +697,14 @@ if node[:hardware][:shm_size]
     notifies :run, "execute[remount-dev-shm]"
   end
 end
+
+prometheus_collector "ohai" do
+  interval "15m"
+  user "root"
+  proc_subset "all"
+  capability_bounding_set %w[CAP_DAC_OVERRIDE CAP_SYS_ADMIN]
+  private_devices false
+  private_users false
+  protect_clock false
+  protect_kernel_modules false
+end