X-Git-Url: https://git.openstreetmap.org./chef.git/blobdiff_plain/fdd1bdfdaf34a4915e79aad85ad87c0f783e425a..5a753e9b0b3f168314666812681dee6056aa3996:/cookbooks/hardware/recipes/default.rb diff --git a/cookbooks/hardware/recipes/default.rb b/cookbooks/hardware/recipes/default.rb index 92abdf65d..517794141 100644 --- a/cookbooks/hardware/recipes/default.rb +++ b/cookbooks/hardware/recipes/default.rb @@ -17,6 +17,13 @@ # limitations under the License. # +include_recipe "tools" +include_recipe "munin" + +ohai_plugin "hardware" do + template "ohai.rb.erb" +end + case node[:cpu][:"0"][:vendor_id] when "GenuineIntel" package "intel-microcode" @@ -69,33 +76,6 @@ when "IBM" speed = "115200" end -if manufacturer == "HP" && node[:lsb][:release].to_f > 11.10 - include_recipe "git" - - git "/opt/hp/hp-legacy" do - action :sync - repository "git://chef.openstreetmap.org/hp-legacy.git" - user "root" - group "root" - end - - link "/opt/hp/hp-health/bin/hpasmd" do - to "/opt/hp/hp-legacy/hpasmd" - end - - link "/usr/lib/libhpasmintrfc.so.3.0" do - to "/opt/hp/hp-legacy/libhpasmintrfc.so.3.0" - end - - link "/usr/lib/libhpasmintrfc.so.3" do - to "libhpasmintrfc.so.3.0" - end - - link "/usr/lib/libhpasmintrfc.so" do - to "libhpasmintrfc.so.3.0" - end -end - unless unit.nil? file "/etc/init/ttySttyS#{unit}.conf" do action :delete @@ -117,6 +97,24 @@ unless unit.nil? end end +# if we need a different / special kernel version to make the hardware +# work (e.g: https://github.com/openstreetmap/operations/issues/45) then +# ensure that we have the package installed. the grub template will +# make sure that this is the default on boot. +if node[:hardware][:grub][:kernel] + kernel_version = node[:hardware][:grub][:kernel] + + package "linux-image-#{kernel_version}-generic" + package "linux-image-extra-#{kernel_version}-generic" + package "linux-headers-#{kernel_version}-generic" + + boot_device = IO.popen(["df", "/boot"]).readlines.last.split.first + boot_uuid = IO.popen(["blkid", "-o", "value", "-s", "UUID", boot_device]).readlines.first.chomp + grub_entry = "gnulinux-advanced-#{boot_uuid}>gnulinux-#{kernel_version}-advanced-#{boot_uuid}" +else + grub_entry = "0" +end + if File.exist?("/etc/default/grub") execute "update-grub" do action :nothing @@ -128,7 +126,7 @@ if File.exist?("/etc/default/grub") owner "root" group "root" mode 0644 - variables :unit => unit, :speed => speed + variables :unit => unit, :speed => speed, :entry => grub_entry notifies :run, "execute[update-grub]" end end @@ -180,15 +178,15 @@ status_packages = {} node[:kernel][:modules].each_key do |modname| case modname when "cciss" - tools_packages << "hpacucli" + tools_packages << "hpssacli" status_packages["cciss-vol-status"] ||= [] when "hpsa" - tools_packages << "hpacucli" + tools_packages << "hpssacli" status_packages["cciss-vol-status"] ||= [] when "mptsas" tools_packages << "lsiutil" # status_packages["mpt-status"] ||= [] - when "mpt2sas" + when "mpt2sas", "mpt3sas" tools_packages << "sas2ircu" status_packages["sas2ircu-status"] ||= [] when "megaraid_mm" @@ -200,6 +198,8 @@ node[:kernel][:modules].each_key do |modname| when "aacraid" tools_packages << "arcconf" status_packages["aacraid-status"] ||= [] + when "arcmsr" + tools_packages << "areca" end end @@ -215,7 +215,7 @@ node[:block_device].each do |name, attributes| end end -%w(hpacucli lsiutil sas2ircu megactl megacli arcconf).each do |tools_package| +%w(hpssacli lsiutil sas2ircu megactl megacli arcconf).each do |tools_package| if tools_packages.include?(tools_package) package tools_package else @@ -225,6 +225,22 @@ end end end +if tools_packages.include?("areca") + include_recipe "git" + + git "/opt/areca" do + action :sync + repository "git://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 @@ -253,72 +269,55 @@ end end end -disks = [] - -node[:block_device].each do |name, attributes| - disks << { :device => name } if attributes[:vendor] == "ATA" -end - -if status_packages["cciss-vol-status"] - status_packages["cciss-vol-status"].each do |device| - IO.popen(["cciss_vol_status", "-V", "/dev/#{device}"]).each do |line| - disks << { :device => device, :driver => "cciss", :id => Regexp.last_match[1].to_i - 1 } if line =~ / bay ([0-9]+) +HP / - end - end -end - -if status_packages["megaclisas-status"] - controller = 0 - - Dir.glob("/sys/class/scsi_host/host*") do |host| - driver = File.new("#{host}/proc_name").read.chomp - - next unless driver == "megaraid_sas" - - bus = host.sub("/sys/class/scsi_host/host", "") - device = File.basename(Dir.glob("/sys/bus/scsi/devices/#{bus}:*/scsi_generic/*").first) - - IO.popen(["megacli", "-PDList", "-a#{controller}", "-NoLog"]).each do |line| - disks << { :device => device, :driver => "megaraid", :id => Regexp.last_match[1] } if line =~ /^Device Id: ([0-9]+)$/ - - disks.pop if line =~ /^Firmware state: Hotspare, Spun down$/ +disks = if node[:hardware][:disk] + node[:hardware][:disk][:disks] + else + [] + 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 - - controller += 1 + elsif disk[:device] + device = disk[:device].sub("/dev/", "") + munin = device end -end - -if status_packages["sas2ircu-status"] - Dir.glob("/sys/class/scsi_host/host*") do |host| - driver = File.new("#{host}/proc_name").read.chomp - - next unless driver == "mpt2sas" - bus = host.sub("/sys/class/scsi_host/host", "") + next if device.nil? - Dir.glob("/sys/bus/scsi/devices/#{bus}:0:*/scsi_generic/*").each do |sg| - disks << { :device => File.basename(sg) } - end - end + Hash[ + :device => device, + :smart => smart, + :munin => munin, + :hddtemp => munin.tr("-:", "_") + ] end -if status_packages["aacraid-status"] - Dir.glob("/sys/class/scsi_host/host*") do |host| - driver = File.new("#{host}/proc_name").read.chomp - - next unless driver == "aacraid" - - bus = host.sub("/sys/class/scsi_host/host", "") - - Dir.glob("/sys/bus/scsi/devices/#{bus}:1:*/scsi_generic/*").each do |sg| - disks << { :device => File.basename(sg) } - end - end -end +disks = disks.compact if disks.count > 0 package "smartmontools" + template "/usr/local/bin/smartd-mailer" do + source "smartd-mailer.erb" + owner "root" + group "root" + mode 0755 + end + template "/etc/smartd.conf" do source "smartd.conf.erb" owner "root" @@ -338,6 +337,21 @@ if disks.count > 0 service "smartmontools" do action [:enable, :start] + supports :status => true, :restart => true, :reload => true + 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] && 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 "smartmontools" do @@ -345,6 +359,27 @@ else end end +if disks.count > 0 + 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") @@ -380,7 +415,11 @@ if node[:lsb][:release].to_f <= 12.10 end else service "kmod" do - provider Chef::Provider::Service::Upstart + if node[:lsb][:release].to_f >= 15.10 + provider Chef::Provider::Service::Systemd + else + provider Chef::Provider::Service::Upstart + end action :nothing subscribes :start, "template[/etc/modules]" end @@ -405,6 +444,30 @@ 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"