X-Git-Url: https://git.openstreetmap.org/chef.git/blobdiff_plain/551ba1f4c9d24c1d374c8df04d6101474bd59e52..ea48ac6d563c698f43fd018f40ca84f7ec28f110:/cookbooks/hardware/recipes/default.rb

diff --git a/cookbooks/hardware/recipes/default.rb b/cookbooks/hardware/recipes/default.rb
index fe185248b..5609bd34b 100644
--- a/cookbooks/hardware/recipes/default.rb
+++ b/cookbooks/hardware/recipes/default.rb
@@ -17,9 +17,9 @@
 # limitations under the License.
 #
 
-include_recipe "apt"
 include_recipe "git"
 include_recipe "munin"
+include_recipe "prometheus"
 include_recipe "sysfs"
 include_recipe "tools"
 
@@ -50,25 +50,42 @@ end
 
 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
-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"
+  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]"
+    only_if { 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 }
   end
 
-  if product.end_with?("Gen8", "Gen9")
+  if product.end_with?("Gen8", "Gen9", "Gen10")
     package "hp-ams" do
       action :install
       notifies :restart, "service[hp-ams]"
@@ -140,13 +157,14 @@ if File.exist?("/etc/default/grub")
   execute "update-grub" do
     action :nothing
     command "/usr/sbin/update-grub"
+    not_if { kitchen? }
   end
 
   template "/etc/default/grub" do
     source "grub.erb"
     owner "root"
     group "root"
-    mode 0o644
+    mode "644"
     variables :units => units, :entry => grub_entry
     notifies :run, "execute[update-grub]"
   end
@@ -163,7 +181,7 @@ template "/etc/initramfs-tools/conf.d/mdadm" do
   source "initramfs-mdadm.erb"
   owner "root"
   group "root"
-  mode 0o644
+  mode "644"
   notifies :run, "execute[update-initramfs]"
 end
 
@@ -172,7 +190,23 @@ service "haveged" do
   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
+    options "--config.file=/etc/prometheus/ipmi_local.yml"
+    subscribes :restart, "template[/etc/prometheus/ipmi_local.yml]"
+  end
+end
 
 package "irqbalance"
 
@@ -181,13 +215,30 @@ service "irqbalance" do
   supports :status => false, :restart => true, :reload => false
 end
 
-# Link Layer Discovery Protocol Daemon
 package "lldpd"
+
 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
+end
+
 tools_packages = []
 status_packages = {}
 
@@ -210,9 +261,6 @@ if node[:virtualization][:role] != "guest" ||
     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"] ||= []
@@ -256,7 +304,7 @@ if tools_packages.include?("areca")
     depth 1
     user "root"
     group "root"
-    not_if { ENV["TEST_KITCHEN"] }
+    not_if { kitchen? }
   end
 else
   directory "/opt/areca" do
@@ -265,18 +313,21 @@ 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"
     owner "root"
     group "root"
-    mode 0o755
+    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
@@ -293,7 +344,7 @@ 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
 
@@ -301,7 +352,7 @@ end
       source "raid.default.erb"
       owner "root"
       group "root"
-      mode 0o644
+      mode "644"
       variables :devices => status_packages[status_package]
     end
 
@@ -335,28 +386,37 @@ nvmes = if node[:hardware][:pci]
           []
         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_ssd_tool_version = "3.0.24"
+  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/29399/eng/Intel_SSD_DCT_#{intel_ssd_tool_version}_Linux%20.zip"
+  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 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"
-    not_if { ::File.exist?("#{Chef::Config[:file_cache_path]}/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 "isdct" do
-    version "#{intel_ssd_tool_version}-1"
-    source "#{Chef::Config[:file_cache_path]}/isdct_#{intel_ssd_tool_version}-1_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 "intelmas" do
+    action :purge
   end
 end
 
@@ -378,6 +438,13 @@ disks = disks.map do |disk|
       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)
@@ -402,18 +469,25 @@ disks = disks.compact.uniq
 if disks.count.positive?
   package "smartmontools"
 
+  template "/etc/cron.daily/update-smart-drivedb" do
+    source "update-smart-drivedb.erb"
+    owner "root"
+    group "root"
+    mode "755"
+  end
+
   template "/usr/local/bin/smartd-mailer" do
     source "smartd-mailer.erb"
     owner "root"
     group "root"
-    mode 0o755
+    mode "755"
   end
 
   template "/etc/smartd.conf" do
     source "smartd.conf.erb"
     owner "root"
     group "root"
-    mode 0o644
+    mode "644"
     variables :disks => disks
   end
 
@@ -421,15 +495,27 @@ if disks.count.positive?
     source "smartmontools.erb"
     owner "root"
     group "root"
-    mode 0o644
+    mode "644"
   end
 
-  service "smartd" do
+  service "smartmontools" do
     action [:enable, :start]
     subscribes :reload, "template[/etc/smartd.conf]"
     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"
+  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
@@ -481,7 +567,7 @@ if File.exist?("/etc/mdadm/mdadm.conf")
   file "/etc/mdadm/mdadm.conf" do
     owner "root"
     group "root"
-    mode 0o644
+    mode "644"
     content mdadm_conf
   end
 
@@ -491,16 +577,21 @@ if File.exist?("/etc/mdadm/mdadm.conf")
   end
 end
 
-template "/etc/modules" do
-  source "modules.erb"
-  owner "root"
-  group "root"
-  mode 0o644
+file "/etc/modules" do
+  action :delete
 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]
@@ -510,7 +601,7 @@ if node[:hardware][:watchdog]
     source "watchdog.erb"
     owner "root"
     group "root"
-    mode 0o644
+    mode "644"
     variables :module => node[:hardware][:watchdog]
   end
 
@@ -557,7 +648,7 @@ unless Dir.glob("/sys/class/hwmon/hwmon*").empty?
     source "sensors.conf.erb"
     owner "root"
     group "root"
-    mode 0o644
+    mode "644"
     notifies :run, "execute[/etc/sensors.d/chef.conf]"
   end
 end
@@ -578,3 +669,7 @@ if node[:hardware][:shm_size]
     notifies :run, "execute[remount-dev-shm]"
   end
 end
+
+prometheus_collector "ohai" do
+  interval "15m"
+end