| Server IP : 79.170.40.229 / Your IP : 3.138.122.118 Web Server : Apache System : Linux web231.extendcp.co.uk 4.18.0-513.24.1.el8_9.x86_64 #1 SMP Mon Apr 8 11:23:13 EDT 2024 x86_64 User : 1stforcarhirealicante.com ( 296923) PHP Version : 5.6.40 Disable Function : NONE MySQL : ON | cURL : ON | WGET : ON | Perl : ON | Python : ON | Sudo : ON | Pkexec : ON Directory : /proc/thread-self/root/opt/icinga2-hardware-checks/checks/ |
Upload File : |
#!/usr/bin/env ruby
require 'rubygems' if RUBY_VERSION < '1.9.0'
require 'sensu-plugin/check/cli'
require_relative 'CheckHelper'
require_relative 'Plugincommands'
require 'socket'
class ControllerCheck < Sensu::Plugin::Check::CLI
def run
# if this is a virtul machine, we dont want to proceed
if CheckHelper.is_virtual()
ok
end
vendor = CheckHelper.get_manufacturer()
if vendor == ''
warning = 'Vendor couldnt be determined'
ok warning
end
# parse the hw config file to figure out what tools to use for the hw checks
@settings = CheckHelper.settings()
hwmap = @settings['tools']
@hwmatch = 0
@hwtype = ''
hwmap.each do |name,info|
if vendor.include?(info["vendor"])
# found a match in the config file
@hwmatch = 1
@hwtype = name
break
end
end
# if this is a vendor not found in the config file, lets run the default check
if @hwmatch == 0
@hwtype = 'default'
end
tools_to_run = hwmap[@hwtype]['raid_match']
run_status = {}
tool_found = false
tools_to_run.each do |tool|
# lets run a test against every tool by the order they are configured and if the results are inconclusive, go on to the next
if tool == 'default'
path = ''
else
path = Plugincommands.send('which', tool)
end
if path != nil
tool_found = true
# go ahead and run the test
case
when tool.downcase().include?('omreport')
run_status = perform_omreport_check()
when tool.downcase().include?('sas3')
run_status = perform_sas3ircu_check()
else
run_status = perform_default_check()
end
end
if tool_found
if run_status['exit_code'] == 0
# if the tool is valid, but the results are not concrete, we want to try other configured tools
unless run_status['msg'].include?('Warning')
ok
end
elsif run_status['exit_code'] == 2
critical run_status['msg']
end
end
end
if tool_found == false
# run the default test
begin
run_status = perform_default_check()
rescue => exception
warning 'No tools found to run the RAID tests. Tools checked: omreport, sas3ircu'
end
end
case run_status['exit_code']
when 0
ok
when 1
warning run_status['msg']
when 2
critical run_status['msg']
else
unknown
end
end
def perform_megacli_check
# use megacli/megacli64 to determine controller status
mcbin = ''
retstat = 0
errcount = 0
retmsg = ''
mccmd = Plugincommands.send('which', 'megacli')
if !mccmd.nil?
mcbin = mccmd
elsif File.exist?('/opt/MegaRAID/MegaCli/MegaCli64')
mcbin = '/opt/MegaRAID/MegaCli/MegaCli64'
elsif File.exist?('/opt/MegaRAID/MegaCli/MegaCli')
mcbin = '/opt/MegaRAID/MegaCli/MegaCli'
else
mcbin = '/opt/icinga2-hardware-checks/bin/MegaCli64'
end
if !mcbin.nil? and File.exist?(mcbin)
raidstat = `sudo #{mcbin} -LDInfo -Lall -aAll | grep ^State`
raidstat = raidstat.split(/\r?\n/)
if raidstat.length > 0
raidstat.each do |line|
if !line.include? 'Optimal'
retstat = 2
errcount += 1
end
end
else
retstat = 1
retmsg = mcbin + " could not determine raid status"
end
end
if !retstat.zero?
if retmsg == ''
retmsg = mcbin + " found " + errcount.to_s + " raid problems"
end
end
out = {'exit_code' => retstat, 'msg' => retmsg}
return out
end
def perform_omreport_check()
# uses omreport for the raid check; mainly for dell servers
# get the application path
omreport = Plugincommands.send('which', 'omreport')
# Sample output:
# Status : Ok
begin
# on c-series, omreport is known not to work for this
# where we find omreport doesn't work and megacli does,
# we should use megacli instead
unless CheckHelper.validate_omreport
if CheckHelper.validate_megacli
disk_status = perform_megacli_check
return disk_status
end
end
# If MegaCli is installed and working, check for punctures
if CheckHelper.validate_megacli
exit_code = 0
msg = ''
puncout = `sudo #{Plugincommands.megacli} -AdpAllInfo -aALL -NoLog`
if puncout.include? 'FATAL:Punctur'
msg = 'RAID Puncture detected:' + line
exit_code = 2
end
unless exit_code.zero?
out = {'exit_code' => exit_code, 'msg' => msg}
return out
end
end
# Check OMSA for failures
out = `#{omreport} storage controller | grep -i status`
cmdout = out.split(/\r?\n/)
if cmdout.length == 0
msg = 'Warning: Couldnt perform the check with this tool'
end
for line in cmdout
status = line.split(':')[1].strip().downcase()
# if status anything other than ok/okay/online, error out
unless ['ok','okay','online'].any? { |word| status.include?(word) }
msg = "omreport raid check failed with status #{status}"
out = {'exit_code' => 2, 'msg' => msg}
return out
end
end
rescue => exception
msg = exception.backtrace
out = {'exit_code' => 2, 'msg' => msg}
return out
end
out = {'exit_code' => 0, 'msg' => 'Tests passed'}
return out
end
def perform_sas3ircu_check()
# as with any tool, this only works for certain models (ceph/hadoop models)
path = Plugincommands.send('which', 'sas3ircu')
exit_code = 0
msg = "All checks passed"
begin
ok_list = ['okay','online']
# checking for 'status of volume' in the sas3 output will help us determine the controller status
# Sample output:
# Status of volume : Okay (OKY)
out = `sudo #{path} 0 display | grep -i 'status of volume'`
cmdout = out.split(/\r?\n/)
if cmdout.length == 0
msg = 'Warning: Couldnt perform the check with this tool'
end
for line in cmdout
status = line.split(':')[1].strip().downcase()
# anything other than okay/online, exit with error
unless ok_list.any? { |word| status.include?(word) }
msg = "sas3 raid checks failed with status: #{status}"
exit_code = 2
end
end
rescue => exception
msg = exception.backtrace
exit_code = 2
ensure
out = {'exit_code' => exit_code, 'msg' => msg}
return out
end
end
def perform_default_check()
# default when nothing else 'works'. The op will display an error in case there is one
cmd = "grep -E 'U_|_U' /proc/mdstat"
overall_raid_status = true
op = `#{cmd}`
if op.length > 0
overall_raid_status = false
end
if overall_raid_status == false
out = {'exit_code' => 2, 'msg' => 'RAID Check failed based on /proc/mdstat'}
return out
else
out = {'exit_code' => 0, 'msg' => 'Tests passed'}
return out
end
end
end