disk_subsystem.rb.erb

#!/usr/bin/env ruby
###
# RAID Card / Drive Testing
###

# Error Tracking
$errors = Array.new


<% if node[:kernel][:modules].include? 'megaraid_sas' -%>
#####
## Check status of Dell megaraid_sas cards
#####

    omreport = '/opt/dell/srvadmin/bin/omreport'

    # check each disk
    %x{#{omreport} storage vdisk}.scan(/[Ss]tatus\s*:\s(.*)/).each do |status|
        if not status.first.eql? 'Ok'
            $errors.push("VDisk state: #{status.first}")
        end
    end
    
    # check storage (raid) controller
    num_cont = 0
    %x{#{omreport} storage controller}.scan(/[Ss]tatus\s*:\s(.*)/).each do |status|
        num_cont += 1
        if not (status.first.eql? 'Ok' or status.first.eql? 'Non-Critical')
            $errors.push("Controller state: #{status.first}")
        end
    end
    
    # check all physical drives
    (0...num_cont).each do |controller|
        res = %x{#{omreport} storage pdisk controller=#{controller}}
        res.scan(/^[Ss]tatus\s*:\s(.*)/).each do |status|
            if not (status.first.eql? 'Ok' or status.first.eql? 'Non-Critical')
                $errors.push("PDisk state: #{status.first}")
            end
        end
        res.scan(/^Failure\sPredicted\s*:\s(.*)/).each do |status|
            if not (status.first.eql? 'No')
                $errors.push("PDisk Failure Predicted: #{status.first}")
            end
        end
    end

    # check battery
    #%x{#{omreport} storage battery}.scan(/^[Ss]tatus\s*:\s(.*)/).each do |status|
    #    if not (status.first.eql? 'Ok' or status.first.eql? 'Non-Critical')
    #        $errors.push("Battery status: #{status.first}")
    #    end
    #end


<% elsif node[:kernel][:modules].include? 'aacraid' -%>
#####
## Check status of aacraid cards
#####
# NOTE: Storman doesn't add itself to the path properly, use full pathing

    controllers = %x{bash -c "LD_PRELOAD=/usr/StorMan/libstdc++.so.5 /usr/StorMan/arcconf getversion"}
    controllers.scan(/Controller #([0-9]+)/).flatten.each do |controller|
    
        storman = %x{bash -c "LD_PRELOAD=/usr/StorMan/libstdc++.so.5 /usr/StorMan/arcconf getconfig #{controller}"}
        
        #Check status of stuff
        items = ["status of logical device", "controller status"]
        items.each do |item|
            storman.scan(/^\s*#{item}\s*:?\s?(\S*)$/i) do |status|
                if not status.first.downcase.eql? 'optimal'
                    $errors.push("#{item}: #{status}")
                end
            end
        end
        
        #Check for failed/degraded devices
        storman.scan(/Logical devices\/Failed\/Degraded[\s:]*?[0-9]+\/([0-9]+)\/([0-9]+)/) do |status|
            status.each do |fd|
                if not fd.eql? '0'
                    $errors.push("Logical device failed or degraded")
                end
            end
        end

        #Check controller battery
        #(Turned off per colin's request)
        #storman.scan(/Controller Battery Information.*?Status[\s:]*([^\n]+)/m) do |status|
        #    if not (status.first.downcase.eql? 'optimal' or status.first.downcase.eql? 'charging')
        #        $errors.push("Controller Battery Information: #{status}")
        #    end
        #end

    end
    
<% else -%>
# Doesn't seem to be a managed RAID device.  Assume everything is okay (or 'sploded).
<% end -%>

#These 45 and 47 are vm's, they don't really have ipmi. 32 is being dumb
<% if not node[:disk_subsystem][:ipmi_exclude].include? node[:hostname] -%>

#IPMI stuff
%x{ipmitool sdr type Fan}.scan(/[^\n]+/).each do |fan|
    if not fan.match(/\|\s*ok\s*\|/)
        $errors.push("Fan status: #{fan}")
    end
end

%x{ipmitool sdr type 'Power Supply'}.scan(/[^\n]+/) do |supply|
    if not supply.match(/\|\s*ok\s*\||\|\s*lnc\s*\||\|\s*lcr\s*\|/) or supply.match(/Redundancy Lost/)
        $errors.push("Power supply status: #{supply}")
    end
end

<% end -%>


# we haz errors?
if not $errors == []
    $errors.each do |thing| 
        puts thing
    end
end