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 c5b5ac226ef6d42c51a9c717d5e347ea0660da7f..48e8f28889d3286069cada8c68ce92bd47a7e213 100644 (file)
--- a/cookbooks/hardware/recipes/default.rb
+++ b/cookbooks/hardware/recipes/default.rb
@@ -17,9 +17,9 @@
 # limitations under the License.
 #
 
 # limitations under the License.
 #
 

-include_recipe "apt"

 include_recipe "git"
 include_recipe "munin"
 include_recipe "git"
 include_recipe "munin"

+include_recipe "prometheus"

 include_recipe "sysfs"
 include_recipe "tools"
 
 include_recipe "sysfs"
 include_recipe "tools"
 


@@ -50,22 +50,40 @@ end
 
 units = []
 
 
 units = []
 

-if node[:roles].include?("bytemark") || node[:roles].include?("exonetric")
+if node[:roles].include?("bytemark") || node[:roles].include?("exonetric") || node[:roles].include?("prgmr")

   units << "0"
 end
 
 case manufacturer
   units << "0"
 end
 
 case manufacturer

-when "HP"
+when "HP", "HPE"
+  include_recipe "apt::management-component-pack"
+

   package "hponcfg"
 
   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]"
   package "hp-health" do
     action :install
     notifies :restart, "service[hp-health]"

+    only_if { node[:lsb][:release].to_f < 22.04 }

   end
 
   service "hp-health" do
     action [:enable, :start]
     supports :status => true, :restart => true
   end
 
   service "hp-health" do
     action [:enable, :start]
     supports :status => true, :restart => true

+    only_if { node[:lsb][:release].to_f < 22.04 }

   end
 
   if product.end_with?("Gen8", "Gen9")
   end
 
   if product.end_with?("Gen8", "Gen9")


@@ -78,9 +96,23 @@ when "HP"
       action [:enable, :start]
       supports :status => true, :restart => true
     end
       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
 
   end
 

-  units << "1"
+  units << if product.end_with?("Gen10")
+             "0"
+           else
+             "1"
+           end

 when "TYAN"
   units << "0"
 when "TYAN Computer Corporation"
 when "TYAN"
   units << "0"
 when "TYAN Computer Corporation"


@@ -114,6 +146,7 @@ end
 units.sort.uniq.each do |unit|
   service "serial-getty@ttyS#{unit}" do
     action [:enable, :start]
 units.sort.uniq.each do |unit|
   service "serial-getty@ttyS#{unit}" do
     action [:enable, :start]

+    not_if { kitchen? }

   end
 end
 
   end
 end
 


@@ -140,6 +173,7 @@ if File.exist?("/etc/default/grub")
   execute "update-grub" do
     action :nothing
     command "/usr/sbin/update-grub"
   execute "update-grub" do
     action :nothing
     command "/usr/sbin/update-grub"

+    not_if { kitchen? }

   end
 
   template "/etc/default/grub" do
   end
 
   template "/etc/default/grub" do


@@ -152,6 +186,8 @@ if File.exist?("/etc/default/grub")
   end
 end
 
   end
 end
 

+package "initramfs-tools"
+

 execute "update-initramfs" do
   action :nothing
   command "update-initramfs -u -k all"
 execute "update-initramfs" do
   action :nothing
   command "update-initramfs -u -k all"


@@ -172,7 +208,27 @@ service "haveged" do
   action [:enable, :start]
 end
 
   action [:enable, :start]
 end
 

-package "ipmitool" if node[:kernel][:modules].include?("ipmi_si")
+if node[:kernel][:modules].include?("ipmi_si")
+  package "ipmitool"
+  package "freeipmi-tools"
+
+  template "/etc/prometheus/ipmi_local.yml" do
+    source "ipmi_local.yml.erb"
+    owner "root"
+    group "root"
+    mode "644"
+  end
+
+  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
+end

 
 package "irqbalance"
 
 
 package "irqbalance"
 


@@ -181,13 +237,31 @@ service "irqbalance" do
   supports :status => false, :restart => true, :reload => false
 end
 
   supports :status => false, :restart => true, :reload => false
 end
 

-# Link Layer Discovery Protocol Daemon

 package "lldpd"
 package "lldpd"

+

 service "lldpd" do
   action [:start, :enable]
   supports :status => true, :restart => true, :reload => true
 end
 
 service "lldpd" do
   action [:start, :enable]
   supports :status => true, :restart => true, :reload => true
 end
 

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


@@ -210,9 +284,6 @@ if node[:virtualization][:role] != "guest" ||
     when "mpt2sas", "mpt3sas"
       tools_packages << "sas2ircu"
       status_packages["sas2ircu-status"] ||= []
     when "mpt2sas", "mpt3sas"
       tools_packages << "sas2ircu"
       status_packages["sas2ircu-status"] ||= []

-    when "megaraid_mm"
-      tools_packages << "megactl"
-      status_packages["megaraid-status"] ||= []

     when "megaraid_sas"
       tools_packages << "megacli"
       status_packages["megaclisas-status"] ||= []
     when "megaraid_sas"
       tools_packages << "megacli"
       status_packages["megaclisas-status"] ||= []


@@ -256,7 +327,7 @@ if tools_packages.include?("areca")
     depth 1
     user "root"
     group "root"
     depth 1
     user "root"
     group "root"

-    not_if { ENV["TEST_KITCHEN"] }
+    not_if { kitchen? }

   end
 else
   directory "/opt/areca" do
   end
 else
   directory "/opt/areca" do


@@ -265,6 +336,8 @@ else
   end
 end
 
   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"
 if status_packages.include?("cciss-vol-status")
   template "/usr/local/bin/cciss-vol-statusd" do
     source "cciss-vol-statusd.erb"


@@ -277,6 +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"
   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
     private_tmp true
     protect_system "full"
     protect_home true


@@ -293,7 +367,7 @@ else
   end
 end
 
   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
 
   if status_packages.include?(status_package)
     package status_package
 


@@ -335,28 +409,37 @@ nvmes = if node[:hardware][:pci]
           []
         end
 
           []
         end
 

+unless nvmes.empty?
+  package "nvme-cli"
+end
+

 intel_nvmes = nvmes.select { |pci| pci[:vendor_name] == "Intel Corporation" }
 
 if !intel_ssds.empty? || !intel_nvmes.empty?
   package "unzip"
 
 intel_nvmes = nvmes.select { |pci| pci[:vendor_name] == "Intel Corporation" }
 
 if !intel_ssds.empty? || !intel_nvmes.empty?
   package "unzip"
 

-  intel_ssd_tool_version = "3.0.25"
+  sst_tool_version = "1.3"
+  sst_package_version = "#{sst_tool_version}.208-0"

 
 

-  remote_file "#{Chef::Config[:file_cache_path]}/Intel_SSD_Data_Center_Tool_#{intel_ssd_tool_version}_Linux.zip" do
-    source "https://downloadmirror.intel.com/29556/eng/Intel_SSD_Data_Center_Tool_#{intel_ssd_tool_version}_Linux.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_SSD_Data_Center_Tool_#{intel_ssd_tool_version}_Linux.zip" do
-    command "unzip Intel_SSD_Data_Center_Tool_#{intel_ssd_tool_version}_Linux.zip Intel_SSD_Data_Center_Tool_#{intel_ssd_tool_version}_Linux/isdct_#{intel_ssd_tool_version}-1_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"
     cwd Chef::Config[:file_cache_path]
     user "root"
     group "root"

-    not_if { ::File.exist?("#{Chef::Config[:file_cache_path]}/Intel_SSD_Data_Center_Tool_#{intel_ssd_tool_version}_Linux/isdct_#{intel_ssd_tool_version}-1_amd64.deb") }
+    not_if { ::File.exist?("#{Chef::Config[:file_cache_path]}/sst_#{sst_package_version}_amd64.deb") }
+  end
+
+  dpkg_package "sst" do
+    version "#{sst_package_version}"
+    source "#{Chef::Config[:file_cache_path]}/sst_#{sst_package_version}_amd64.deb"

   end
 
   end
 

-  dpkg_package "isdct" do
-    version "#{intel_ssd_tool_version}-1"
-    source "#{Chef::Config[:file_cache_path]}/Intel_SSD_Data_Center_Tool_#{intel_ssd_tool_version}_Linux/isdct_#{intel_ssd_tool_version}-1_amd64.deb"
+  dpkg_package "intelmas" do
+    action :purge

   end
 end
 
   end
 end
 


@@ -378,6 +461,13 @@ disks = disks.map do |disk|
       elsif smart =~ %r{^.*,(\d+)/(\d+)$}
         munin = "#{device}-#{Regexp.last_match(1)}:#{Regexp.last_match(2)}"
       end
       elsif smart =~ %r{^.*,(\d+)/(\d+)$}
         munin = "#{device}-#{Regexp.last_match(1)}:#{Regexp.last_match(2)}"
       end

+    elsif disk[:device]
+      device = disk[:device].sub("/dev/", "")
+      smart = disk[:smart_device]
+
+      if smart =~ /^.*,(\d+),(\d+),(\d+)$/
+        munin = "#{device}-#{Regexp.last_match(1)}:#{Regexp.last_match(2)}:#{Regexp.last_match(3)}"
+      end

     end
   elsif disk[:device] =~ %r{^/dev/(nvme\d+)n\d+$}
     device = Regexp.last_match(1)
     end
   elsif disk[:device] =~ %r{^/dev/(nvme\d+)n\d+$}
     device = Regexp.last_match(1)


@@ -437,6 +527,23 @@ if disks.count.positive?
     subscribes :restart, "template[/etc/default/smartmontools]"
   end
 
     subscribes :restart, "template[/etc/default/smartmontools]"
   end
 

+  template "/etc/prometheus/collectors/smart.devices" do
+    source "smart.devices.erb"
+    owner "root"
+    group "root"
+    mode "644"
+    variables :disks => disks
+  end
+
+  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
   # an Areca controller as they only allow one thing to
   # talk to the controller at a time and smartd will
   # Don't try and do munin monitoring of disks behind
   # an Areca controller as they only allow one thing to
   # talk to the controller at a time and smartd will


@@ -492,22 +599,27 @@ if File.exist?("/etc/mdadm/mdadm.conf")
     content mdadm_conf
   end
 
     content mdadm_conf
   end
 

-  service "mdadm" do
+  service "mdmonitor" do

     action :nothing
     subscribes :restart, "file[/etc/mdadm/mdadm.conf]"
   end
 end
 
     action :nothing
     subscribes :restart, "file[/etc/mdadm/mdadm.conf]"
   end
 end
 

-template "/etc/modules" do
-  source "modules.erb"
-  owner "root"
-  group "root"
-  mode "644"
+file "/etc/modules" do
+  action :delete

 end
 
 end
 

-service "kmod" do
-  action :nothing
-  subscribes :start, "template[/etc/modules]"
+node[:hardware][:modules].each do |module_name|
+  kernel_module module_name do
+    action :install
+    not_if { kitchen? }
+  end
+end
+
+node[:hardware][:blacklisted_modules].each do |module_name|
+  kernel_module module_name do
+    action :blacklist
+  end

 end
 
 if node[:hardware][:watchdog]
 end
 
 if node[:hardware][:watchdog]


@@ -585,3 +697,14 @@ if node[:hardware][:shm_size]
     notifies :run, "execute[remount-dev-shm]"
   end
 end
     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