include_recipe "apt"
include_recipe "git"
include_recipe "munin"
+include_recipe "prometheus"
include_recipe "sysfs"
include_recipe "tools"
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
execute "update-grub" do
action :nothing
command "/usr/sbin/update-grub"
+ not_if { ENV["TEST_KITCHEN"] }
end
template "/etc/default/grub" 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"
+
+ prometheus_exporter "ipmi" do
+ port 9290
+ end
+end
package "irqbalance"
if !intel_ssds.empty? || !intel_nvmes.empty?
package "unzip"
- intel_ssd_tool_version = "3.0.25"
+ intel_mas_tool_version = "1.5"
+ intel_mas_package_version = "#{intel_mas_tool_version}.113-0"
- 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"
+ remote_file "#{Chef::Config[:file_cache_path]}/Intel_MAS_CLI_Tool_#{intel_mas_tool_version}_Linux.zip" do
+ source "https://downloadmirror.intel.com/30162/eng/Intel%C2%AE_MAS_CLI_Tool_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 Intel_SSD_Data_Center_Tool_#{intel_ssd_tool_version}_Linux/isdct_#{intel_ssd_tool_version}-1_amd64.deb"
+ 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"
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]}/intelmas_#{intel_mas_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"
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"
+ action :purge
end
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)
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"
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