Add alerts for planet dumps and replication feeds

[chef.git] / cookbooks / hardware / recipes / default.rb

diff --git a/cookbooks/hardware/recipes/default.rb b/cookbooks/hardware/recipes/default.rb
index fa5f43c648f33f2ec2dd3e7cd554573a619dcc8e..751810a3e9faa5f63c98efafb6d6450230f7c8a0 100644 (file)
--- a/cookbooks/hardware/recipes/default.rb
+++ b/cookbooks/hardware/recipes/default.rb
@@ -20,6 +20,7 @@
 include_recipe "apt"
 include_recipe "git"
 include_recipe "munin"
 include_recipe "apt"
 include_recipe "git"
 include_recipe "munin"

+include_recipe "prometheus"

 include_recipe "sysfs"
 include_recipe "tools"
 
 include_recipe "sysfs"
 include_recipe "tools"
 


@@ -50,7 +51,7 @@ end
 
 units = []
 
 
 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
 
   units << "0"
 end
 


@@ -86,12 +87,7 @@ when "TYAN"
 when "TYAN Computer Corporation"
   units << "0"
 when "Supermicro"
 when "TYAN Computer Corporation"
   units << "0"
 when "Supermicro"

-  case product
-  when "H8DGU", "X9SCD", "X7DBU", "X7DW3", "X9DR7/E-(J)LN4F", "X9DR3-F", "X9DRW", "SYS-1028U-TN10RT+", "SYS-2028U-TN24R4T+", "SYS-1029P-WTRT", "Super Server"
-    units << "1"
-  else
-    units << "0"
-  end
+  units << "1"

 when "IBM"
   units << "0"
 when "VMware, Inc."
 when "IBM"
   units << "0"
 when "VMware, Inc."


@@ -145,13 +141,14 @@ if File.exist?("/etc/default/grub")
   execute "update-grub" do
     action :nothing
     command "/usr/sbin/update-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"
   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
     variables :units => units, :entry => grub_entry
     notifies :run, "execute[update-grub]"
   end


@@ -168,7 +165,7 @@ template "/etc/initramfs-tools/conf.d/mdadm" do
   source "initramfs-mdadm.erb"
   owner "root"
   group "root"
   source "initramfs-mdadm.erb"
   owner "root"
   group "root"

-  mode 0o644
+  mode "644"

   notifies :run, "execute[update-initramfs]"
 end
 
   notifies :run, "execute[update-initramfs]"
 end
 


@@ -177,69 +174,87 @@ service "haveged" do
   action [:enable, :start]
 end
 
   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
 
 end
 

+package "irqbalance"
+

 service "irqbalance" do
   action [:start, :enable]
   supports :status => false, :restart => true, :reload => false
 service "irqbalance" do
   action [:start, :enable]
   supports :status => false, :restart => true, :reload => false

-  subscribes :restart, "template[/etc/default/irqbalance]"

 end
 
 end
 

-# Link Layer Discovery Protocol Daemon

 package "lldpd"
 package "lldpd"

+

 service "lldpd" do
   action [:start, :enable]
   supports :status => true, :restart => true, :reload => true
 end
 
 service "lldpd" do
   action [:start, :enable]
   supports :status => true, :restart => true, :reload => true
 end
 

+ohai_plugin "lldp" do
+  template "lldp.rb.erb"
+end
+

 tools_packages = []
 status_packages = {}
 
 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]+$/
-    status_packages["cciss-vol-status"] |= ["cciss/#{Regexp.last_match[1]}d0"]
-  else
-    Dir.glob("/sys/block/#{name}/device/scsi_generic/*").each do |sg|
-      status_packages["cciss-vol-status"] |= [File.basename(sg)]
+if node[:virtualization][:role] != "guest" ||
+   (node[:virtualization][:system] != "lxc" &&
+    node[:virtualization][:system] != "lxd" &&
+    node[:virtualization][:system] != "openvz")
+
+  node[:kernel][:modules].each_key do |modname|
+    case modname
+    when "cciss"
+      tools_packages << "ssacli"
+      status_packages["cciss-vol-status"] ||= []
+    when "hpsa"
+      tools_packages << "ssacli"
+      status_packages["cciss-vol-status"] ||= []
+    when "mptsas"
+      tools_packages << "lsiutil"
+      status_packages["mpt-status"] ||= []
+    when "mpt2sas", "mpt3sas"
+      tools_packages << "sas2ircu"
+      status_packages["sas2ircu-status"] ||= []
+    when "megaraid_mm"
+      tools_packages << "megactl"
+      status_packages["megaraid-status"] ||= []
+    when "megaraid_sas"
+      tools_packages << "megacli"
+      status_packages["megaclisas-status"] ||= []
+    when "aacraid"
+      tools_packages << "arcconf"
+      status_packages["aacraid-status"] ||= []
+    when "arcmsr"
+      tools_packages << "areca"
+    end
+  end
+
+  node[:block_device].each do |name, attributes|
+    next unless attributes[:vendor] == "HP" && attributes[:model] == "LOGICAL VOLUME"
+
+    if name =~ /^cciss!(c[0-9]+)d[0-9]+$/
+      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
 end
     end
   end
 end


@@ -260,9 +275,10 @@ if tools_packages.include?("areca")
   git "/opt/areca" do
     action :sync
     repository "https://git.openstreetmap.org/private/areca.git"
   git "/opt/areca" do
     action :sync
     repository "https://git.openstreetmap.org/private/areca.git"

+    depth 1

     user "root"
     group "root"
     user "root"
     group "root"

-    not_if { ENV["TEST_KITCHEN"] }
+    not_if { kitchen? }

   end
 else
   directory "/opt/areca" do
   end
 else
   directory "/opt/areca" do


@@ -276,7 +292,7 @@ if status_packages.include?("cciss-vol-status")
     source "cciss-vol-statusd.erb"
     owner "root"
     group "root"
     source "cciss-vol-statusd.erb"
     owner "root"
     group "root"

-    mode 0o755
+    mode "755"

     notifies :restart, "service[cciss-vol-statusd]"
   end
 
     notifies :restart, "service[cciss-vol-statusd]"
   end
 


@@ -289,6 +305,14 @@ if status_packages.include?("cciss-vol-status")
     no_new_privileges true
     notifies :restart, "service[cciss-vol-statusd]"
   end
     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 megaraid-status megaclisas-status aacraid-status].each do |status_package|
 end
 
 %w[cciss-vol-status mpt-status sas2ircu-status megaraid-status megaclisas-status aacraid-status].each do |status_package|


@@ -299,7 +323,7 @@ end
       source "raid.default.erb"
       owner "root"
       group "root"
       source "raid.default.erb"
       owner "root"
       group "root"

-      mode 0o644
+      mode "644"

       variables :devices => status_packages[status_package]
     end
 
       variables :devices => status_packages[status_package]
     end
 


@@ -325,38 +349,47 @@ disks = if node[:hardware][:disk]
           []
         end
 
           []
         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"
-#
-#   intel_ssd_tool_version = "3.0.21"
-#
-#   remote_file "#{Chef::Config[:file_cache_path]}/Intel_SSD_Data_Center_Tool_#{intel_ssd_tool_version}_Linux.zip" do
-#     source "https://downloadmirror.intel.com/29115/eng/Intel_SSD_Data_Center_Tool_#{intel_ssd_tool_version}_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 isdct_#{intel_ssd_tool_version}-1_amd64.deb"
-#     cwd Chef::Config[:file_cache_path]
-#     user "root"
-#     group "root"
-#     not_if { File.exist?("#{Chef::Config[:file_cache_path]}/isdct_#{intel_ssd_tool_version}-1_amd64.deb") }
-#   end
-#
-#   dpkg_package "isdct" do
-#     version "#{intel_ssd_tool_version}-1"
-#     source "#{Chef::Config[:file_cache_path]}/isdct_#{intel_ssd_tool_version}-1_amd64.deb"
-#   end
-# 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"
+
+  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_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]}/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
+    action :purge
+  end
+end

 
 disks = disks.map do |disk|
   next if disk[:state] == "spun_down" || %w[unconfigured failed].any?(disk[:status])
 
 disks = disks.map do |disk|
   next if disk[:state] == "spun_down" || %w[unconfigured failed].any?(disk[:status])


@@ -376,6 +409,13 @@ disks = disks.map do |disk|
       elsif smart =~ %r{^.*,(\d+)/(\d+)$}
         munin = "#{device}-#{Regexp.last_match(1)}:#{Regexp.last_match(2)}"
       end
       elsif smart =~ %r{^.*,(\d+)/(\d+)$}
         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+$}
     device = Regexp.last_match(1)
     end
   elsif disk[:device] =~ %r{^/dev/(nvme\d+)n\d+$}
     device = Regexp.last_match(1)


@@ -400,18 +440,25 @@ disks = disks.compact.uniq
 if disks.count.positive?
   package "smartmontools"
 
 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"
   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"
   end
 
   template "/etc/smartd.conf" do
     source "smartd.conf.erb"
     owner "root"
     group "root"

-    mode 0o644
+    mode "644"

     variables :disks => disks
   end
 
     variables :disks => disks
   end
 


@@ -419,15 +466,27 @@ if disks.count.positive?
     source "smartmontools.erb"
     owner "root"
     group "root"
     source "smartmontools.erb"
     owner "root"
     group "root"

-    mode 0o644
+    mode "644"

   end
 
   end
 

-  service "smartd" do
+  service "smartmontools" do

     action [:enable, :start]
     subscribes :reload, "template[/etc/smartd.conf]"
     subscribes :restart, "template[/etc/default/smartmontools]"
   end
 
     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
   # 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


@@ -479,7 +538,7 @@ if File.exist?("/etc/mdadm/mdadm.conf")
   file "/etc/mdadm/mdadm.conf" do
     owner "root"
     group "root"
   file "/etc/mdadm/mdadm.conf" do
     owner "root"
     group "root"

-    mode 0o644
+    mode "644"

     content mdadm_conf
   end
 
     content mdadm_conf
   end
 


@@ -489,16 +548,21 @@ if File.exist?("/etc/mdadm/mdadm.conf")
   end
 end
 
   end
 end
 

-template "/etc/modules" do
-  source "modules.erb"
-  owner "root"
-  group "root"
-  mode 0o644
+file "/etc/modules" do
+  action :delete

 end
 
 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]
 end
 
 if node[:hardware][:watchdog]


@@ -508,7 +572,7 @@ if node[:hardware][:watchdog]
     source "watchdog.erb"
     owner "root"
     group "root"
     source "watchdog.erb"
     owner "root"
     group "root"

-    mode 0o644
+    mode "644"

     variables :module => node[:hardware][:watchdog]
   end
 
     variables :module => node[:hardware][:watchdog]
   end
 


@@ -555,7 +619,7 @@ unless Dir.glob("/sys/class/hwmon/hwmon*").empty?
     source "sensors.conf.erb"
     owner "root"
     group "root"
     source "sensors.conf.erb"
     owner "root"
     group "root"

-    mode 0o644
+    mode "644"

     notifies :run, "execute[/etc/sensors.d/chef.conf]"
   end
 end
     notifies :run, "execute[/etc/sensors.d/chef.conf]"
   end
 end