Fix monitoring of devices behind adaptec RAID cards

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

diff --git a/cookbooks/hardware/recipes/default.rb b/cookbooks/hardware/recipes/default.rb
index 66fedff1a02c2f950e67df308f1349a9998ce0c6..e8f903f64cc8a255aedcc127fae666f85a42b89d 100644 (file)
--- a/cookbooks/hardware/recipes/default.rb
+++ b/cookbooks/hardware/recipes/default.rb
@@ -20,6 +20,7 @@
 include_recipe "apt"
 include_recipe "git"
 include_recipe "munin"
+include_recipe "prometheus"
 include_recipe "sysfs"
 include_recipe "tools"
 
@@ -86,12 +87,7 @@ when "TYAN"
 when "TYAN Computer Corporation"
   units << "0"
 when "Supermicro"
-  case product
-  when "H8DGU", "X9SCD", "X7DBU", "X7DW3", "X9DR7/E-(J)LN4F", "X9DR3-F", "X9DRW", "SYS-1028U-TN10RT+", "SYS-2028U-TN24R4T+", "SYS-1029P-WTRT", "Super Server"
-    units << "1"
-  else
-    units << "0"
-  end
+  units << "1"
 when "IBM"
   units << "0"
 when "VMware, Inc."
@@ -145,13 +141,14 @@ if File.exist?("/etc/default/grub")
   execute "update-grub" do
     action :nothing
     command "/usr/sbin/update-grub"
+    not_if { ENV["TEST_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
@@ -168,7 +165,7 @@ 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
 
@@ -197,7 +194,9 @@ tools_packages = []
 status_packages = {}
 
 if node[:virtualization][:role] != "guest" ||
-   node[:virtualization][:system] != "lxd"
+   (node[:virtualization][:system] != "lxc" &&
+    node[:virtualization][:system] != "lxd" &&
+    node[:virtualization][:system] != "openvz")
 
   node[:kernel][:modules].each_key do |modname|
     case modname
@@ -273,7 +272,7 @@ if status_packages.include?("cciss-vol-status")
     source "cciss-vol-statusd.erb"
     owner "root"
     group "root"
-    mode 0o755
+    mode "755"
     notifies :restart, "service[cciss-vol-statusd]"
   end
 
@@ -304,7 +303,7 @@ end
       source "raid.default.erb"
       owner "root"
       group "root"
-      mode 0o644
+      mode "644"
       variables :devices => status_packages[status_package]
     end
 
@@ -330,38 +329,38 @@ 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"
-#
-#   intel_ssd_tool_version = "3.0.21"
-#
-#   remote_file "#{Chef::Config[:file_cache_path]}/Intel_SSD_Data_Center_Tool_#{intel_ssd_tool_version}_Linux.zip" do
-#     source "https://downloadmirror.intel.com/29115/eng/Intel_SSD_Data_Center_Tool_#{intel_ssd_tool_version}_Linux.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"
-#     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") }
-#   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"
-#   end
-# 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"
+
+  intel_ssd_tool_version = "3.0.26"
+
+  remote_file "#{Chef::Config[:file_cache_path]}/Intel_SSD_Data_Center_Tool_#{intel_ssd_tool_version}_Linux.zip" do
+    source "https://downloadmirror.intel.com/29720/eng/Intel_SSD_DCT_#{intel_ssd_tool_version}_Linux.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}.400-1_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}.400-1_amd64.deb") }
+  end
+
+  dpkg_package "isdct" do
+    version "#{intel_ssd_tool_version}-1"
+    source "#{Chef::Config[:file_cache_path]}/isdct_#{intel_ssd_tool_version}.400-1_amd64.deb"
+  end
+end
 
 disks = disks.map do |disk|
   next if disk[:state] == "spun_down" || %w[unconfigured failed].any?(disk[:status])
@@ -381,6 +380,13 @@ 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)
@@ -405,18 +411,25 @@ 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
 
@@ -424,10 +437,10 @@ 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]"
@@ -446,6 +459,18 @@ if disks.count.positive?
       conf_variables :disk => disk
     end
   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
 else
   service "smartd" do
     action [:stop, :disable]
@@ -484,7 +509,7 @@ 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
 
@@ -498,7 +523,7 @@ template "/etc/modules" do
   source "modules.erb"
   owner "root"
   group "root"
-  mode 0o644
+  mode "644"
 end
 
 service "kmod" do
@@ -513,7 +538,7 @@ if node[:hardware][:watchdog]
     source "watchdog.erb"
     owner "root"
     group "root"
-    mode 0o644
+    mode "644"
     variables :module => node[:hardware][:watchdog]
   end
 
@@ -560,7 +585,7 @@ 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