+# haveged is only required on older kernels
+# /dev/random is not blocking anymore in 5.15+
+if Chef::Util.compare_versions(node[:kernel][:release], [5, 15]).negative?
+ package "haveged"
+ service "haveged" do
+ action [:enable, :start]
+ end
+else
+ service "haveged" do
+ action [:stop, :disable]
+ end
+ package "haveged" do
+ action :remove
+ end
+end
+
+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
+ user "root"
+ private_devices false
+ protect_clock false
+ system_call_filter ["@system-service", "@raw-io"]
+ options "--config.file=/etc/prometheus/ipmi_local.yml"
+ subscribes :restart, "template[/etc/prometheus/ipmi_local.yml]"
+ end
+end
+
+package "irqbalance"
+
+service "irqbalance" do
+ action [:start, :enable]
+ supports :status => false, :restart => true, :reload => false
+end
+
+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
+ user "root"
+end
+
+tools_packages = []
+status_packages = {}
+
+if node[:virtualization][:role] != "guest" ||
+ (node[:virtualization][:system] != "lxc" &&
+ node[:virtualization][:system] != "lxd" &&
+ node[:virtualization][:system] != "openvz")
+
+ 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_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
+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://git.openstreetmap.org/private/areca.git"
+ depth 1
+ user "root"
+ group "root"
+ not_if { kitchen? }
+ end
+else
+ directory "/opt/areca" do
+ action :delete
+ recursive true
+ end
+end
+
+include_recipe "apt::hwraid" unless status_packages.empty?
+
+%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
+
+ service "#{status_package}d" do
+ action [:stop, :disable]
+ end
+
+ file "/etc/default/#{status_package}d" do
+ action :delete
+ end
+ else
+ package status_package do
+ action :purge
+ end
+ end
+end
+
+systemd_service "cciss-vol-statusd" do
+ action :delete
+end
+
+template "/usr/local/bin/cciss-vol-statusd" do
+ action :delete
+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
+
+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"
+
+ sst_tool_version = "1.3"
+ sst_package_version = "#{sst_tool_version}.208-0"
+
+ # 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]}/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]}/sst_#{sst_package_version}_amd64.deb") }
+ end
+
+ 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
+
+disks = disks.map do |disk|
+ next if disk[:state] == "spun_down" || %w[unconfigured failed].any?(disk[:status])
+
+ if disk[:smart_device]
+ controller = node[:hardware][:disk][:controllers][disk[:controller]]
+
+ if controller && controller[:device]
+ 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/", "")
+ 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)
+ munin = device
+ elsif disk[:device]
+ device = disk[:device].sub("/dev/", "")
+ munin = device
+ end
+
+ next if device.nil? || munin.nil?
+
+ Hash[
+ :device => device,
+ :smart => smart,
+ :munin => munin,
+ :hddtemp => munin.tr("-:", "_")
+ ]
+end
+
+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 "755"
+ end
+
+ template "/etc/smartd.conf" do
+ source "smartd.conf.erb"
+ owner "root"
+ group "root"
+ mode "644"
+ variables :disks => disks
+ end
+
+ template "/etc/default/smartmontools" do
+ source "smartmontools.erb"
+ owner "root"
+ group "root"
+ mode "644"
+ end
+
+ 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"
+ user "root"
+ capability_bounding_set %w[CAP_DAC_OVERRIDE CAP_SYS_ADMIN CAP_SYS_RAWIO]
+ private_devices false
+ private_users false
+ protect_clock false
+ 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" 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
+ conf "munin.smart.erb"
+ 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 "644"
+ content mdadm_conf
+ end
+
+ service "mdmonitor" do
+ action :nothing
+ subscribes :restart, "file[/etc/mdadm/mdadm.conf]"
+ end
+end
+
+file "/etc/modules" do
+ action :delete
+end
+
+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]
+ package "watchdog"
+
+ template "/etc/default/watchdog" do
+ source "watchdog.erb"
+ owner "root"
+ group "root"
+ mode "644"
+ 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 "644"
+ notifies :run, "execute[/etc/sensors.d/chef.conf]"
+ end
+end
+
+if node[:hardware][:shm_size]
+ execute "remount-dev-shm" do
+ action :nothing
+ command "/bin/mount -o remount /dev/shm"
+ user "root"
+ group "root"
+ end
+
+ mount "/dev/shm" do
+ action :enable
+ device "tmpfs"
+ fstype "tmpfs"
+ options "rw,nosuid,nodev,size=#{node[:hardware][:shm_size]}"
+ notifies :run, "execute[remount-dev-shm]"
+ end
+end
+
+prometheus_collector "ohai" do
+ interval "15m"
+ user "root"
+ proc_subset "all"
+ capability_bounding_set %w[CAP_DAC_OVERRIDE CAP_SYS_ADMIN CAP_SYS_RAWIO]
+ private_devices false
+ private_users false
+ protect_clock false
+ protect_kernel_modules false
projects / chef.git / blobdiff