+package "haveged"
+service "haveged" do
+ action [:enable, :start]
+end
+
+package "ipmitool" if node[:kernel][:modules].include?("ipmi_si")
+
+package "irqbalance"
+
+template "/etc/default/irqbalance" do
+ source "irqbalance.erb"
+ owner "root"
+ group "root"
+ mode 0o644
+end
+
+service "irqbalance" do
+ action [:start, :enable]
+ supports :status => false, :restart => true, :reload => false
+ subscribes :restart, "template[/etc/default/irqbalance]"
+end
+
+# Link Layer Discovery Protocol Daemon
+package "lldpd"
+service "lldpd" do
+ action [:start, :enable]
+ supports :status => true, :restart => true, :reload => true
+end
+
+if node[:hardware][:mcelog][:enabled]
+ package "mcelog"
+
+ %w[bus cache dimm iomca page socket-memory unknown].each do |trigger|
+ template "/etc/mcelog/#{trigger}-error-trigger.local" do
+ source "mcelog-trigger.erb"
+ owner "root"
+ group "root"
+ mode 0o755
+ end
+ end
+
+ service "mcelog" do
+ action [:start, :enable]
+ supports :status => true, :restart => true, :reload => false
+ end
+end
+
+tools_packages = []
+status_packages = {}
+
+node[:kernel][:modules].each_key do |modname|
+ case modname
+ when "cciss"
+ tools_packages << "ssacli"
+ status_packages["cciss-vol-status"] ||= []
+ when "hpsa"
+ tools_packages << "ssacli"
+ status_packages["cciss-vol-status"] ||= []
+ when "mptsas"
+ tools_packages << "lsiutil"
+ status_packages["mpt-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 "aacraid"
+ tools_packages << "arcconf"
+ status_packages["aacraid-status"] ||= []
+ when "arcmsr"
+ tools_packages << "areca"
+ end
+end
+
+node[:block_device].each do |name, attributes|
+ next unless attributes[:vendor] == "HP" && attributes[:model] == "LOGICAL VOLUME"
+
+ if name =~ /^cciss!(c[0-9]+)d[0-9]+$/
+ status_packages["cciss-vol-status"] |= ["cciss/#{Regexp.last_match[1]}d0"]
+ else
+ Dir.glob("/sys/block/#{name}/device/scsi_generic/*").each do |sg|
+ status_packages["cciss-vol-status"] |= [File.basename(sg)]
+ end
+ end
+end
+
+%w[ssacli lsiutil sas2ircu megactl megacli arcconf].each do |tools_package|
+ if tools_packages.include?(tools_package)
+ package tools_package
+ else
+ package tools_package do
+ action :purge
+ end
+ end
+end
+
+if tools_packages.include?("areca")
+ include_recipe "git"
+
+ git "/opt/areca" do
+ action :sync
+ repository "https://chef.openstreetmap.org/areca.git"
+ user "root"
+ group "root"
+ end
+else
+ directory "/opt/areca" do
+ action :delete
+ recursive true
+ end
+end
+
+["cciss-vol-status", "mpt-status", "sas2ircu-status", "megaraid-status", "megaclisas-status", "aacraid-status"].each do |status_package|
+ if status_packages.include?(status_package)
+ package status_package
+
+ template "/etc/default/#{status_package}d" do
+ source "raid.default.erb"
+ owner "root"
+ group "root"
+ mode 0o644
+ variables :devices => status_packages[status_package]
+ end
+
+ service "#{status_package}d" do
+ action [:start, :enable]
+ supports :status => false, :restart => true, :reload => false
+ subscribes :restart, "template[/etc/default/#{status_package}d]"
+ end
+ else
+ package status_package do
+ action :purge
+ end
+
+ file "/etc/default/#{status_package}d" do
+ action :delete
+ end
+ end
+end
+
+disks = if node[:hardware][:disk]
+ node[:hardware][:disk][:disks]
+ else
+ []
+ 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"
+
+ remote_file "#{Chef::Config[:file_cache_path]}/Intel_SSD_Data_Center_Tool_3.0.13_Linux.zip" do
+ source "https://downloadmirror.intel.com/27863/eng/Intel_SSD_Data_Center_Tool_3.0.13_Linux.zip"
+ end
+
+ execute "#{Chef::Config[:file_cache_path]}/Intel_SSD_Data_Center_Tool_3.0.13_Linux.zip" do
+ command "unzip Intel_SSD_Data_Center_Tool_3.0.13_Linux.zip isdct_3.0.13.400-17_amd64.deb"
+ cwd Chef::Config[:file_cache_path]
+ user "root"
+ group "root"
+ not_if { File.exist?("#{Chef::Config[:file_cache_path]}/isdct_3.0.13.400-17_amd64.deb") }
+ end
+
+ dpkg_package "isdct" do
+ version "3.0.13.400-17"
+ source "#{Chef::Config[:file_cache_path]}/isdct_3.0.13.400-17_amd64.deb"
+ end
+end
+
+disks = disks.map do |disk|
+ next if disk[:state] == "spun_down"
+
+ if disk[:smart_device]
+ controller = node[:hardware][:disk][:controllers][disk[:controller]]
+ device = controller[:device].sub("/dev/", "")
+ smart = disk[:smart_device]
+
+ if device.start_with?("cciss/") && smart =~ /^cciss,(\d+)$/
+ array = node[:hardware][:disk][:arrays][disk[:arrays].first]
+ munin = "cciss-3#{array[:wwn]}-#{Regexp.last_match(1)}"
+ elsif smart =~ /^.*,(\d+)$/
+ munin = "#{device}-#{Regexp.last_match(1)}"
+ elsif smart =~ %r{^.*,(\d+)/(\d+)$}
+ munin = "#{device}-#{Regexp.last_match(1)}:#{Regexp.last_match(2)}"
+ end
+ elsif disk[:device]
+ device = disk[:device].sub("/dev/", "")
+ munin = device
+ end
+
+ next if device.nil?
+
+ Hash[
+ :device => device,
+ :smart => smart,
+ :munin => munin,
+ :hddtemp => munin.tr("-:", "_")
+ ]
+end
+
+smartd_service = if node[:lsb][:release].to_f >= 16.04
+ "smartd"
+ else
+ "smartmontools"
+ end
+
+disks = disks.compact
+
+if disks.count.positive?
+ package "smartmontools"
+
+ template "/usr/local/bin/smartd-mailer" do
+ source "smartd-mailer.erb"
+ owner "root"
+ group "root"
+ mode 0o755
+ end
+
+ template "/etc/smartd.conf" do
+ source "smartd.conf.erb"
+ owner "root"
+ group "root"
+ mode 0o644
+ variables :disks => disks
+ end
+
+ template "/etc/default/smartmontools" do
+ source "smartmontools.erb"
+ owner "root"
+ group "root"
+ mode 0o644
+ end
+
+ service smartd_service do
+ action [:enable, :start]
+ subscribes :reload, "template[/etc/smartd.conf]"
+ subscribes :restart, "template[/etc/default/smartmontools]"
+ 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
+ # throw errors if it clashes with munin
+ disks = disks.reject { |disk| disk[:smart]&.start_with?("areca,") }
+
+ disks.each do |disk|
+ munin_plugin "smart_#{disk[:munin]}" do
+ target "smart_"
+ conf "munin.smart.erb"
+ conf_variables :disk => disk
+ end
+ end
+else
+ service smartd_service do
+ action [:stop, :disable]
+ end
+end
+
+if disks.count.positive?
+ munin_plugin "hddtemp_smartctl" do
+ conf "munin.hddtemp.erb"
+ conf_variables :disks => disks
+ end
+else
+ munin_plugin "hddtemp_smartctl" do
+ action :delete
+ conf "munin.hddtemp.erb"
+ end
+end
+
+plugins = Dir.glob("/etc/munin/plugins/smart_*").map { |p| File.basename(p) } -
+ disks.map { |d| "smart_#{d[:munin]}" }
+
+plugins.each do |plugin|
+ munin_plugin plugin do
+ action :delete
+ end
+end
+
+if File.exist?("/etc/mdadm/mdadm.conf")
+ mdadm_conf = edit_file "/etc/mdadm/mdadm.conf" do |line|
+ line.gsub!(/^MAILADDR .*$/, "MAILADDR admins@openstreetmap.org")
+
+ line
+ end
+
+ file "/etc/mdadm/mdadm.conf" do
+ owner "root"
+ group "root"
+ mode 0o644
+ content mdadm_conf
+ end
+
+ service "mdadm" do
+ action :nothing
+ subscribes :restart, "file[/etc/mdadm/mdadm.conf]"
+ end
+end
+
+template "/etc/modules" do
+ source "modules.erb"
+ owner "root"
+ group "root"
+ mode 0o644
+end
+
+service "kmod" do
+ action :nothing
+ subscribes :start, "template[/etc/modules]"
+end
+
+if node[:hardware][:watchdog]
+ package "watchdog"
+
+ template "/etc/default/watchdog" do
+ source "watchdog.erb"
+ owner "root"
+ group "root"
+ mode 0o644
+ variables :module => node[:hardware][:watchdog]
+ end
+
+ service "watchdog" do
+ action [:enable, :start]
+ end
+end
+
+unless Dir.glob("/sys/class/hwmon/hwmon*").empty?
+ package "lm-sensors"
+
+ Dir.glob("/sys/devices/platform/coretemp.*").each do |coretemp|
+ cpu = File.basename(coretemp).sub("coretemp.", "").to_i
+ chip = format("coretemp-isa-%04d", cpu)
+
+ temps = if File.exist?("#{coretemp}/name")
+ Dir.glob("#{coretemp}/temp*_input").map do |temp|
+ File.basename(temp).sub("temp", "").sub("_input", "").to_i
+ end.sort
+ else
+ Dir.glob("#{coretemp}/hwmon/hwmon*/temp*_input").map do |temp|
+ File.basename(temp).sub("temp", "").sub("_input", "").to_i
+ end.sort
+ end
+
+ if temps.first == 1
+ node.default[:hardware][:sensors][chip][:temps][:temp1][:label] = "CPU #{cpu}"
+ temps.shift
+ end
+
+ temps.each_with_index do |temp, index|
+ node.default[:hardware][:sensors][chip][:temps]["temp#{temp}"][:label] = "CPU #{cpu} Core #{index}"
+ end
+ end
+
+ execute "/etc/sensors.d/chef.conf" do
+ action :nothing
+ command "/usr/bin/sensors -s"
+ user "root"
+ group "root"
+ end
+
+ template "/etc/sensors.d/chef.conf" do
+ source "sensors.conf.erb"
+ owner "root"
+ group "root"
+ mode 0o644
+ notifies :run, "execute[/etc/sensors.d/chef.conf]"
+ end
projects / chef.git / blobdiff