X-Git-Url: https://git.openstreetmap.org./chef.git/blobdiff_plain/902f43c89759db08e4e4d35911280f958701faa5..6d4344a0b4300fd8210ee0b1e2af90d1ddfa9072:/cookbooks/hardware/recipes/default.rb

diff --git a/cookbooks/hardware/recipes/default.rb b/cookbooks/hardware/recipes/default.rb
index 936d0d0db..a77908cf4 100644
--- a/cookbooks/hardware/recipes/default.rb
+++ b/cookbooks/hardware/recipes/default.rb
@@ -1,8 +1,8 @@
 #
-# Cookbook Name:: hardware
+# Cookbook:: hardware
 # Recipe:: default
 #
-# Copyright 2012, OpenStreetMap Foundation
+# Copyright:: 2012, OpenStreetMap Foundation
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -17,8 +17,12 @@
 # limitations under the License.
 #
 
-include_recipe "tools"
+include_recipe "apt"
+include_recipe "git"
 include_recipe "munin"
+include_recipe "prometheus"
+include_recipe "sysfs"
+include_recipe "tools"
 
 ohai_plugin "hardware" do
   template "ohai.rb.erb"
@@ -47,7 +51,7 @@ end
 
 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
 
@@ -55,14 +59,41 @@ case manufacturer
 when "HP"
   package "hponcfg"
 
+  execute "update-ilo" do
+    action :nothing
+    command "/usr/sbin/hponcfg -f /etc/ilo-defaults.xml"
+  end
+
+  template "/etc/ilo-defaults.xml" do
+    source "ilo-defaults.xml.erb"
+    owner "root"
+    group "root"
+    mode "644"
+    notifies :run, "execute[update-ilo]"
+  end
+
   package "hp-health" do
     action :install
     notifies :restart, "service[hp-health]"
+    only_if { node[:lsb][:release].to_f < 22.04 }
   end
 
   service "hp-health" do
     action [:enable, :start]
     supports :status => true, :restart => true
+    only_if { node[:lsb][:release].to_f < 22.04 }
+  end
+
+  if product.end_with?("Gen8", "Gen9")
+    package "hp-ams" do
+      action :install
+      notifies :restart, "service[hp-ams]"
+    end
+
+    service "hp-ams" do
+      action [:enable, :start]
+      supports :status => true, :restart => true
+    end
   end
 
   units << "1"
@@ -71,55 +102,34 @@ 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-2028U-TN24R4T+", "Super Server"
-    units << "1"
-  else
-    units << "0"
-  end
+  units << "1"
 when "IBM"
   units << "0"
-end
+when "VMware, Inc."
+  package "open-vm-tools"
 
-# Remove legacy HP G4 support which breaks modern hp-health 10.4
-if manufacturer == "HP"
-  %w[/opt/hp/hp-health/bin/hpasmd /usr/lib/libhpasmintrfc.so.3.0 /usr/lib/libhpasmintrfc.so.3 /usr/lib/libhpasmintrfc.so].each do |filename|
-    file filename do
-      action :delete
-      manage_symlink_source false
-    end
+  # Remove timeSync plugin completely
+  # https://github.com/vmware/open-vm-tools/issues/302
+  file "/usr/lib/open-vm-tools/plugins/vmsvc/libtimeSync.so" do
+    action :delete
+    notifies :restart, "service[open-vm-tools]"
   end
 
-  directory "/opt/hp/hp-legacy" do
-    action :delete
-    recursive true
+  # Attempt to tell Host we are not interested in timeSync
+  execute "vmware-toolbox-cmd-timesync-disable" do
+    command "/usr/bin/vmware-toolbox-cmd timesync disable"
+    ignore_failure true
+  end
+
+  service "open-vm-tools" do
+    action [:enable, :start]
+    supports :status => true, :restart => true
   end
 end
 
 units.sort.uniq.each do |unit|
-  if node[:lsb][:release].to_f >= 16.04
-    service "serial-getty@ttyS#{unit}" do
-      action [:enable, :start]
-    end
-  else
-    file "/etc/init/ttySttyS#{unit}.conf" do
-      action :delete
-    end
-
-    template "/etc/init/ttyS#{unit}.conf" do
-      source "tty.conf.erb"
-      owner "root"
-      group "root"
-      mode 0o644
-      variables :unit => unit
-    end
-
-    service "ttyS#{unit}" do
-      provider Chef::Provider::Service::Upstart
-      action [:enable, :start]
-      supports :status => true, :restart => true, :reload => false
-      subscribes :restart, "template[/etc/init/ttyS#{unit}.conf]"
-    end
+  service "serial-getty@ttyS#{unit}" do
+    action [:enable, :start]
   end
 end
 
@@ -133,6 +143,7 @@ if node[:hardware][:grub][:kernel]
   package "linux-image-#{kernel_version}-generic"
   package "linux-image-extra-#{kernel_version}-generic"
   package "linux-headers-#{kernel_version}-generic"
+  package "linux-tools-#{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
@@ -145,13 +156,14 @@ if File.exist?("/etc/default/grub")
   execute "update-grub" do
     action :nothing
     command "/usr/sbin/update-grub"
+    not_if { 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 +180,7 @@ 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
 
@@ -177,87 +189,97 @@ service "haveged" 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"
 
-package "irqbalance"
+  template "/etc/prometheus/ipmi_local.yml" do
+    source "ipmi_local.yml.erb"
+    owner "root"
+    group "root"
+    mode "644"
+  end
 
-template "/etc/default/irqbalance" do
-  source "irqbalance.erb"
-  owner "root"
-  group "root"
-  mode 0o644
+  prometheus_exporter "ipmi" do
+    port 9290
+    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
-  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"
+ohai_plugin "lldp" do
+  template "lldp.rb.erb"
+end
 
-  %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
+package %w[
+  rasdaemon
+  ruby-sqlite3
+]
 
-  service "mcelog" do
-    action [:start, :enable]
-    supports :status => true, :restart => true, :reload => false
-  end
+service "rasdaemon" do
+  action [:enable, :start]
+end
+
+prometheus_exporter "rasdaemon" do
+  port 9797
 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]+$/
-    status_packages["cciss-vol-status"] |= ["cciss/#{Regexp.last_match[1]}d0"]
-  else
-    Dir.glob("/sys/block/#{name}/device/scsi_generic/*").each do |sg|
-      status_packages["cciss-vol-status"] |= [File.basename(sg)]
+if node[:virtualization][:role] != "guest" ||
+   (node[:virtualization][:system] != "lxc" &&
+    node[:virtualization][:system] != "lxd" &&
+    node[:virtualization][:system] != "openvz")
+
+  node[:kernel][:modules].each_key do |modname|
+    case modname
+    when "cciss"
+      tools_packages << "ssacli"
+      status_packages["cciss-vol-status"] ||= []
+    when "hpsa"
+      tools_packages << "ssacli"
+      status_packages["cciss-vol-status"] ||= []
+    when "mptsas"
+      tools_packages << "lsiutil"
+      status_packages["mpt-status"] ||= []
+    when "mpt2sas", "mpt3sas"
+      tools_packages << "sas2ircu"
+      status_packages["sas2ircu-status"] ||= []
+    when "megaraid_sas"
+      tools_packages << "megacli"
+      status_packages["megaclisas-status"] ||= []
+    when "aacraid"
+      tools_packages << "arcconf"
+      status_packages["aacraid-status"] ||= []
+    when "arcmsr"
+      tools_packages << "areca"
+    end
+  end
+
+  node[:block_device].each do |name, attributes|
+    next unless attributes[:vendor] == "HP" && attributes[:model] == "LOGICAL VOLUME"
+
+    if name =~ /^cciss!(c[0-9]+)d[0-9]+$/
+      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
@@ -277,9 +299,11 @@ if tools_packages.include?("areca")
 
   git "/opt/areca" do
     action :sync
-    repository "https://chef.openstreetmap.org/areca.git"
+    repository "https://git.openstreetmap.org/private/areca.git"
+    depth 1
     user "root"
     group "root"
+    not_if { kitchen? }
   end
 else
   directory "/opt/areca" do
@@ -288,7 +312,36 @@ else
   end
 end
 
-["cciss-vol-status", "mpt-status", "sas2ircu-status", "megaraid-status", "megaclisas-status", "aacraid-status"].each do |status_package|
+if status_packages.include?("cciss-vol-status")
+  template "/usr/local/bin/cciss-vol-statusd" do
+    source "cciss-vol-statusd.erb"
+    owner "root"
+    group "root"
+    mode "755"
+    notifies :restart, "service[cciss-vol-statusd]"
+  end
+
+  systemd_service "cciss-vol-statusd" do
+    description "Check cciss_vol_status values in the background"
+    exec_start "/usr/local/bin/cciss-vol-statusd"
+    nice 10
+    private_tmp true
+    protect_system "full"
+    protect_home true
+    no_new_privileges true
+    notifies :restart, "service[cciss-vol-statusd]"
+  end
+else
+  systemd_service "cciss-vol-statusd" do
+    action :delete
+  end
+
+  template "/usr/local/bin/cciss-vol-statusd" do
+    action :delete
+  end
+end
+
+%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
 
@@ -296,7 +349,7 @@ end
       source "raid.default.erb"
       owner "root"
       group "root"
-      mode 0o644
+      mode "644"
       variables :devices => status_packages[status_package]
     end
 
@@ -330,45 +383,69 @@ nvmes = if node[:hardware][:pci]
           []
         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"
 
-  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"
+  intel_mas_tool_version = "1.10"
+  intel_mas_package_version = "#{intel_mas_tool_version}.155-0"
+
+  remote_file "#{Chef::Config[:file_cache_path]}/Intel_MAS_CLI_Tool_#{intel_mas_tool_version}_Linux.zip" do
+    source "https://downloadmirror.intel.com/646992/Intel_MAS_CLI_Tool_Linux_#{intel_mas_tool_version}-v2.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"
+  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]}/isdct_3.0.13.400-17_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 "3.0.13.400-17"
-    source "#{Chef::Config[:file_cache_path]}/isdct_3.0.13.400-17_amd64.deb"
+    action :purge
   end
 end
 
 disks = disks.map do |disk|
-  next if disk[:state] == "spun_down"
+  next if disk[:state] == "spun_down" || %w[unconfigured failed].any?(disk[:status])
 
   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+)$}
-      munin = "#{device}-#{Regexp.last_match(1)}:#{Regexp.last_match(2)}"
+
+    if controller && controller[:device]
+      device = controller[:device].sub("/dev/", "")
+      smart = disk[:smart_device]
+
+      if device.start_with?("cciss/") && smart =~ /^cciss,(\d+)$/
+        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
@@ -384,29 +461,30 @@ disks = disks.map do |disk|
   ]
 end
 
-smartd_service = if node[:lsb][:release].to_f >= 16.04
-                   "smartd"
-                 else
-                   "smartmontools"
-                 end
-
-disks = disks.compact
+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
 
@@ -414,15 +492,27 @@ if disks.count.positive?
     source "smartmontools.erb"
     owner "root"
     group "root"
-    mode 0o644
+    mode "644"
   end
 
-  service smartd_service do
+  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"
+  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
@@ -437,7 +527,7 @@ if disks.count.positive?
     end
   end
 else
-  service smartd_service do
+  service "smartd" do
     action [:stop, :disable]
   end
 end
@@ -460,6 +550,7 @@ plugins = Dir.glob("/etc/munin/plugins/smart_*").map { |p| File.basename(p) } -
 plugins.each do |plugin|
   munin_plugin plugin do
     action :delete
+    conf "munin.smart.erb"
   end
 end
 
@@ -473,7 +564,7 @@ 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
 
@@ -483,16 +574,21 @@ if File.exist?("/etc/mdadm/mdadm.conf")
   end
 end
 
-template "/etc/modules" do
-  source "modules.erb"
-  owner "root"
-  group "root"
-  mode 0o644
+file "/etc/modules" do
+  action :delete
 end
 
-service "kmod" do
-  action :nothing
-  subscribes :start, "template[/etc/modules]"
+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]
@@ -502,7 +598,7 @@ if node[:hardware][:watchdog]
     source "watchdog.erb"
     owner "root"
     group "root"
-    mode 0o644
+    mode "644"
     variables :module => node[:hardware][:watchdog]
   end
 
@@ -549,7 +645,28 @@ 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
+
+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"
+end