]> git.openstreetmap.org Git - chef.git/blobdiff - cookbooks/hardware/templates/default/ohai.rb.erb
Detect unconfigured and failed disks on megaraid controllers
[chef.git] / cookbooks / hardware / templates / default / ohai.rb.erb
index 132d5bcacdb015bb697a958a35958363d7562cf0..074f5cbf6e35c624ecda8fe39e7dc3a922df02e2 100644 (file)
@@ -148,16 +148,17 @@ Ohai.plugin(:Hardware) do
     find_direct_disks(disk)
     find_nvme_disks(disk)
 
-    find_hp_disks(disk) if File.exist?("/usr/sbin/hpssacli")
+    find_hp_disks(disk) if File.exist?("/usr/sbin/ssacli")
     find_megaraid_disks(disk) if File.exist?("/usr/sbin/megacli")
-    find_mpt_disks(disk) if File.exist?("/usr/sbin/sas2ircu")
+    find_mpt1_disks(disk) if File.exist?("/usr/sbin/lsiutil")
+    find_mpt2_disks(disk) if File.exist?("/usr/sbin/sas2ircu")
     find_adaptec_disks(disk) if File.exist?("/usr/sbin/arcconf")
     find_areca_disks(disk) if File.exist?("/opt/areca/x86_64/cli64")
 
     find_md_arrays(disk)
 
     disk[:disks].each do |disk|
-      if disk[:vendor] =~ /^CVPR/ && disk[:model] == "INTEL"
+      if disk[:vendor] =~ /^(BTWA|CVPR|PHDV)/ && disk[:model] == "INTEL"
         disk[:model] = disk[:serial_number]
         disk[:serial_number] = disk[:vendor]
         disk[:vendor] = "INTEL"
@@ -285,7 +286,7 @@ Ohai.plugin(:Hardware) do
     array = nil
     disk = nil
 
-    IO.popen(%w(hpssacli controller all show config detail)).each do |line|
+    IO.popen(%w(ssacli controller all show config detail)).each do |line|
       if line =~ /^Smart Array (\S+) /
         controller = {
           :id => devices[:controllers].count,
@@ -358,7 +359,7 @@ Ohai.plugin(:Hardware) do
         controller[:device] = File.basename(device).sub(/^cciss(\d+)$/, "/dev/cciss/c\\1d0")
       elsif device = Dir.glob("/sys/bus/pci/devices/#{controller[:pci_slot]}/host*/target*:3:0/*:3:0:0/scsi_generic/sg*").first
         controller[:device] = "/dev/#{File.basename(device)}"
-      elsif device = Dir.glob("/sys/bus/pci/devices/#{controller[:pci_slot]}/host*/target*:1:0/*:1:0:0/scsi_generic/sg*").first
+      elsif device = Dir.glob("/sys/bus/pci/devices/#{controller[:pci_slot]}/host*/target*:1:0/*:1:0:*/scsi_generic/sg*").first
         controller[:device] = "/dev/#{File.basename(device)}"
       end
     end
@@ -439,8 +440,10 @@ Ohai.plugin(:Hardware) do
       elsif disk && line =~ /^Firmware state:\s+(.*\S)\s*$/
         Regexp.last_match(1).split(/,\s*/).each do |state|
           case state
+          when "Unconfigured(bad)" then disk[:status] = "unconfigured"
           when "Online" then disk[:status] = "online"
           when "Hotspare" then disk[:status] = "hotspare"
+          when "Failed" then disk[:status] = "failed"
           when "Spun Up" then disk[:spun_down] = false
           when "Spun down" then disk[:spun_down] = true
           end
@@ -478,8 +481,10 @@ Ohai.plugin(:Hardware) do
       elsif disk && line =~ /^Firmware state:\s+(.*\S)\s*$/
         Regexp.last_match(1).split(/,\s*/).each do |state|
           case state
+          when "Unconfigured(bad)" then disk[:status] = "unconfigured"
           when "Online" then disk[:status] = "online"
           when "Hotspare" then disk[:status] = "hotspare"
+          when "Failed" then disk[:status] = "failed"
           when "Spun Up" then disk[:state] = "spun_up"
           when "Spun down" then disk[:state] = "spun_down"
           end
@@ -501,7 +506,86 @@ Ohai.plugin(:Hardware) do
     end
   end
 
-  def find_mpt_disks(devices)
+  def find_mpt1_disks(devices)
+    controllers = []
+    disks = []
+
+    controller = nil
+
+    IO.popen(%w(lsiutil -s)).each do |line|
+      if line =~ /^\/proc\/mpt\/ioc(\d+)\s+LSI Logic\s+(\S+)\s+/
+        controller = {
+          :id => devices[:controllers].count,
+          :model => Regexp.last_match(1),
+          :arrays => [],
+          :disks => []
+        }
+
+        controllers << controller
+        devices[:controllers] << controller
+      elsif line =~ /^\s+(\d+)\s+(\d+)\s+PhysDisk (\d+)\s+(\S+)\s+(\S+)\s+\d+\s+(\S+)\s+/
+        disks[Regexp.last_match(3).to_i] = {
+          :id => devices[:disks].count,
+          :controller => controller[:id],
+          :vendor => Regexp.last_match(4),
+          :model => Regexp.last_match(5),
+          :sas_address => Regexp.last_match(6),
+          :arrays => []
+        }
+
+        controller[:disks] << devices[:disks].count
+        devices[:disks] << disks[Regexp.last_match(3).to_i]
+      end
+    end
+
+    controllers.each_with_index do |controller, index|
+      port = index + 1
+      array = nil
+
+      IO.popen(["lsiutil", "-p", port.to_s, "-a", "69,0"]).each do |line|
+        if line =~ /^ (\d+)\s+(\d+)\s+(\d+)\s+(\d+)\s+RAID/
+          seg = Regexp.last_match(1).to_i
+          bus = Regexp.last_match(2).to_i
+          dev = Regexp.last_match(3).to_i
+          fun = Regexp.last_match(4).to_i
+
+          controller[:pci_slot] = sprintf("%04x:%02x:%02x.%01x", seg, bus, dev, fun)
+        end
+      end
+
+      IO.popen(["lsiutil", "-p", port.to_s, "-a", "21,1,0,0"]).each do |line|
+        if line =~ /^Volume (\d+) is/
+          array = {
+            :id => devices[:arrays].count,
+            :controller => controller[:id],
+            :number => Regexp.last_match(1),
+            :disks => []
+          }
+
+          devices[:arrays] << array
+          controller[:arrays] << array[:id]
+        elsif line =~ /^  Member \d+ is PhysDisk (\d+) /
+          array[:disks] << disks[Regexp.last_match(1).to_i][:id]
+          disks[Regexp.last_match(1).to_i][:arrays] << array[:id]
+        end
+      end
+    end
+
+    disks.each do |disk|
+      slot = controllers[disk[:controller]][:pci_slot]
+      sas_address = "0x#{disk[:sas_address]}"
+
+      Dir.glob("/sys/bus/pci/devices/#{slot}/host*/port-*:*/end_device-*:*/sas_device/end_device-*:*").each do |sas_device|
+        if read_sysctl_file("#{sas_device}/sas_address") == sas_address
+          if device = Dir.glob("#{sas_device}/device/target*:0:*/*:0:*:0/scsi_generic/sg*").first
+            disk[:device] = "/dev/#{File.basename(device)}"            
+          end
+        end
+      end
+    end
+  end
+
+  def find_mpt2_disks(devices)
     controllers = []
 
     IO.popen(%w(sas2ircu list)).each do |line|