Add WWN and /dev/* node for PDs and support for toggling the fault LED

bin/check_megacli

@@ -2,6 +2,7 @@
 
 # Example script using pymegacli which is suitable for invocation by nagios
 
+from __future__ import print_function
 import argparse
 import os
 import sys
@@ -31,11 +32,39 @@ def main():
         action='append',
         help='Subsystems to check. If not passed, defaults to all'
     )
+    parser.add_argument(
+        '--led',
+        action='store_true',
+        help='Use libstoragemgmt (megaraid plugin) to toggle the fault led on unhealthy disks'
+    )
+    parser.set_defaults(led=False)
     args = parser.parse_args()
 
     if os.geteuid() != 0:
         parser.error('Must run as root!')
 
+    if args.led:
+        try:
+            import lsm
+            try:
+                import lsm.plugin.megaraid
+                HAVE_LSM_MEGARAID = True
+            except:
+                HAVE_LSM_MEGARAID = False
+                print(
+                    "--led specified but libstoragemgmt megaraid plugin was not found, install like so on Fedora/CentOS... :\n"
+                    "\tyum install libstoragemgmt-megaraid-plugin",
+                    file=sys.stderr
+                )
+        except ImportError:
+            HAVE_LSM_MEGARAID = False
+            print(
+                "--led specified but libstoragemgmt module was not found, install like so on Fedora/CentOS... :\n"
+                "\tyum install libstoragemgmt libstoragemgmt-megaraid-plugin",
+                file=sys.stderr
+            )
+
+
     checks = {}
     for check in CHECKS:
         checks[check] = (not args.check or check in args.check)
@@ -52,12 +81,14 @@ def main():
     def check_component(component):
         if component.healthy:
             messages[OK].append('%s is healthy' % component.identifier)
+            return OK
         else:
             for message in component.health_messages:
                 messages[CRITICAL].append('%s %s' % (
                     component.identifier,
                     message
                 ))
+            return CRITICAL
 
     controllers = list(connection.controllers)
     if not controllers:
@@ -66,10 +97,19 @@ def main():
         for controller in controllers:
             if checks['PD']:
                 for disk in controller.PDs:
-                    check_component(disk)
+                    disk_health = check_component(disk)
+                    if args.led and HAVE_LSM_MEGARAID:
+                        try:
+                            if CRITICAL == disk_health:
+                                lsm.LocalDisk.fault_led_on(disk.devnode)
+                            else:
+                                lsm.LocalDisk.fault_led_off(disk.devnode)
+                        except:
+                            print('Couldn\'t manipulate enclosure LED', file=sys.stderr)
+
             if checks['LD']:
                 for logical_device in controller.LDs:
-                    check_component(disk)
+                    check_component(logical_device)
                     if 'WriteBack' not in logical_device['Current Cache Policy']:
                         messages[WARNING].append('%s has cache policy %s, which does not include WriteBack' % (
                             logical_device.identifier,
@@ -82,16 +122,16 @@ def main():
                         messages[WARNING].append('%s is in learn cycle' % bbu.identifier)
 
     if messages[CRITICAL]:
-        print 'CRITICAL: %s' % '; '.join(messages[CRITICAL])
+        print('CRITICAL: %s' % '; '.join(messages[CRITICAL]))
         return CRITICAL
     elif messages[WARNING]:
-        print 'WARNING: %s' % '; '.join(messages[WARNING])
+        print('WARNING: %s' % '; '.join(messages[WARNING]))
         return WARNING
     elif messages[UNKNOWN]:
-        print 'UNKNOWN: %s' % '; '.join(messages[UNKNOWN])
+        print('UNKNOWN: %s' % '; '.join(messages[UNKNOWN]))
         return UNKNOWN
     else:
-        print 'OK: %s' % '; '.join(messages[OK])
+        print('OK: %s' % '; '.join(messages[OK]))
         return OK
 
 

pymegacli/components.py

@@ -1,6 +1,8 @@
 import pipes
 import subprocess
 import re
+import os
+import glob
 
 from .parser import BlockParser
 from .parser import bail_on
@@ -131,20 +133,29 @@ class Disk(Component):
         'Predictive Failure Count',
     )
     ERROR_BOOL_KEYS = ('Drive has flagged a S.M.A.R.T alert', )
-    REQUIRED_FIELDS = ('Enclosure Device ID', 'Slot Number')
+    REQUIRED_FIELDS = ('Enclosure Device ID', 'Slot Number', 'WWN')
 
     PARSER = BlockParser(rules=[
         once_per_block(colon_field('Enclosure Device ID', int_or_na)),
         rule(colon_field('Slot Number', int)),
+        rule(colon_field('WWN', str)),
         rule(colon_field('Other Error Count', int)),
         rule(colon_field('Predictive Failure Count', int)),
         rule(colon_field('Media Error Count', int)),
         rule(colon_field('Drive has flagged a S.M.A.R.T alert', yesnobool)),
     ], default_constructor=colon_field(None, str))
 
-    def __init__(self, enclosure_id, slot_number, parent, props=None):
+    def __init__(self, enclosure_id, slot_number, wwn, parent, props=None):
         self.enclosure_id = enclosure_id
         self.slot_number = slot_number
+        self.wwn = wwn.lower()
+        disk_by_id_path_glob = "/dev/disk/by-id/wwn-0x%s?" % self.wwn[:-1]
+        disk_by_id_path_glob_results = glob.glob(disk_by_id_path_glob)
+        if len(disk_by_id_path_glob_results) == 1:
+            self.linux_disk_by_id = disk_by_id_path_glob_results[0]
+        else:
+            self.linux_disk_by_id = None
+
         self.thresholds = dict(
             (k, 0)
             for k
@@ -157,7 +168,14 @@ def set_threshold(self, key, value):
 
     @property
     def identifier(self):
-        return 'PhysDrv [%d:%d]' % (self.enclosure_id, self.slot_number)
+        return 'PhysDrv [%d:%d] WWN: %s Dev: %s' % (self.enclosure_id, self.slot_number, self.wwn, self.devnode)
+
+    @property
+    def devnode(self):
+        if self.linux_disk_by_id:
+            return os.path.realpath(self.linux_disk_by_id)
+        else:
+            return None
 
     @property
     def health_status(self):