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 7799058e1d8875a4f4270a619ed2a63c6592de4e..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
 


@@ -140,6 +141,7 @@ 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
   end
 
   template "/etc/default/grub" do


@@ -172,7 +174,23 @@ 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"
+
+  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
+    options "--config.file=/etc/prometheus/ipmi_local.yml"
+    subscribes :restart, "template[/etc/prometheus/ipmi_local.yml]"
+  end
+end

 
 package "irqbalance"
 
 
 package "irqbalance"
 


@@ -181,13 +199,17 @@ service "irqbalance" do
   supports :status => false, :restart => true, :reload => false
 end
 
   supports :status => false, :restart => true, :reload => false
 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 = {}
 


@@ -256,7 +278,7 @@ if tools_packages.include?("areca")
     depth 1
     user "root"
     group "root"
     depth 1
     user "root"
     group "root"

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

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


@@ -335,28 +357,37 @@ nvmes = if node[:hardware][:pci]
           []
         end
 
           []
         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_nvmes = nvmes.select { |pci| pci[:vendor_name] == "Intel Corporation" }
 
 if !intel_ssds.empty? || !intel_nvmes.empty?
   package "unzip"
 

-  intel_ssd_tool_version = "3.0.25"
+  intel_mas_tool_version = "1.10"
+  intel_mas_package_version = "#{intel_mas_tool_version}.155-0"

 
 

-  remote_file "#{Chef::Config[:file_cache_path]}/Intel_SSD_Data_Center_Tool_#{intel_ssd_tool_version}_Linux.zip" do
-    source "https://downloadmirror.intel.com/29556/eng/Intel_SSD_Data_Center_Tool_#{intel_ssd_tool_version}_Linux.zip"
+  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
 
   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 Intel_SSD_Data_Center_Tool_#{intel_ssd_tool_version}_Linux/isdct_#{intel_ssd_tool_version}-1_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"
     cwd Chef::Config[:file_cache_path]
     user "root"
     group "root"

-    not_if { ::File.exist?("#{Chef::Config[:file_cache_path]}/Intel_SSD_Data_Center_Tool_#{intel_ssd_tool_version}_Linux/isdct_#{intel_ssd_tool_version}-1_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
   end
 
   dpkg_package "isdct" do

-    version "#{intel_ssd_tool_version}-1"
-    source "#{Chef::Config[:file_cache_path]}/Intel_SSD_Data_Center_Tool_#{intel_ssd_tool_version}_Linux/isdct_#{intel_ssd_tool_version}-1_amd64.deb"
+    action :purge

   end
 end
 
   end
 end
 


@@ -378,6 +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)


@@ -402,6 +440,13 @@ 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"
   template "/usr/local/bin/smartd-mailer" do
     source "smartd-mailer.erb"
     owner "root"


@@ -430,6 +475,18 @@ if disks.count.positive?
     subscribes :restart, "template[/etc/default/smartmontools]"
   end
 
     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


@@ -491,16 +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 "644"
+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]