| Index: trunk/debs/wikimedia-raid-utils/check-raid.py |
| — | — | @@ -1,8 +1,42 @@ |
| 2 | 2 | #!/usr/bin/python |
| 3 | 3 | |
| 4 | | -import sys, os, re, subprocess |
| | 4 | +import sys, os, os.path, re, subprocess |
| 5 | 5 | |
| 6 | 6 | def main(): |
| | 7 | + osName = os.uname()[0] |
| | 8 | + if osName == 'SunOS': |
| | 9 | + utility = 'zpool' |
| | 10 | + elif osName == 'Linux': |
| | 11 | + utility = getLinuxUtility() |
| | 12 | + else: |
| | 13 | + print 'WARNING: Operating system "%s" is not supported by this check script' % (osName) |
| | 14 | + sys.exit(1) |
| | 15 | + |
| | 16 | + try: |
| | 17 | + if utility == None: |
| | 18 | + print 'OK: no RAID installed' |
| | 19 | + status = 0 |
| | 20 | + elif utility == 'arcconf': |
| | 21 | + status = checkAdaptec() |
| | 22 | + elif utility == 'tw_cli': |
| | 23 | + status = check3ware() |
| | 24 | + elif utility == 'MegaCli': |
| | 25 | + status = checkMegaSas() |
| | 26 | + elif utility == 'zpool': |
| | 27 | + status = checkZfs() |
| | 28 | + elif utility == 'mdadm': |
| | 29 | + status = checkSoftwareRaid() |
| | 30 | + else: |
| | 31 | + print 'WARNING: %s is not yet supported by this check script' % (utility) |
| | 32 | + status = 1 |
| | 33 | + except: |
| | 34 | + error = sys.exc_info()[1] |
| | 35 | + print 'WARNING: check-raid.py encountered exception: ' + str(error) |
| | 36 | + status = 1 |
| | 37 | + |
| | 38 | + sys.exit(status) |
| | 39 | + |
| | 40 | +def getLinuxUtility(): |
| 7 | 41 | f = open("/proc/devices", "r") |
| 8 | 42 | regex = re.compile('^\s*\d+\s+(\w+)') |
| 9 | 43 | utility = None |
| — | — | @@ -23,29 +57,39 @@ |
| 24 | 58 | break |
| 25 | 59 | elif name == 'megaraid_sas_ioctl': |
| 26 | 60 | utility = 'MegaCli' |
| 27 | | - |
| | 61 | + break |
| | 62 | + |
| 28 | 63 | f.close() |
| | 64 | + if utility != None: |
| | 65 | + return utility |
| 29 | 66 | |
| | 67 | + # Try mdadm |
| | 68 | + devices = getSoftwareRaidDevices() |
| | 69 | + if len(devices): |
| | 70 | + return 'mdadm' |
| | 71 | + |
| | 72 | + return None |
| | 73 | + |
| | 74 | +def getSoftwareRaidDevices(): |
| | 75 | + if not os.path.exists('/sbin/mdadm'): |
| | 76 | + return [] |
| | 77 | + |
| 30 | 78 | try: |
| 31 | | - if utility == None: |
| 32 | | - print 'OK: no RAID installed' |
| 33 | | - status = 0 |
| 34 | | - elif utility == 'arcconf': |
| 35 | | - status = checkAdaptec() |
| 36 | | - elif utility == 'tw_cli': |
| 37 | | - status = check3ware() |
| 38 | | - elif utility == 'MegaCli': |
| 39 | | - status = checkMegaSas() |
| 40 | | - else: |
| 41 | | - print 'WARNING: %s is not yet supported by this check script' % (utility) |
| 42 | | - status = 1 |
| | 79 | + proc = subprocess.Popen(['/sbin/mdadm', '--detail', '--scan'], |
| | 80 | + stdout=subprocess.PIPE) |
| 43 | 81 | except: |
| 44 | | - error = sys.exc_info()[1] |
| 45 | | - print 'WARNING: check-raid.py encountered exception: ' + str(error) |
| 46 | | - status = 1 |
| 47 | | - |
| 48 | | - sys.exit(status) |
| | 82 | + return [] |
| 49 | 83 | |
| | 84 | + regex = re.compile('^ARRAY\s+([^ ]*) ') |
| | 85 | + devices = [] |
| | 86 | + for line in proc.stdout: |
| | 87 | + m = regex.match(line) |
| | 88 | + if m != None: |
| | 89 | + devices.append(m.group(1)) |
| | 90 | + proc.wait() |
| | 91 | + |
| | 92 | + return devices |
| | 93 | + |
| 50 | 94 | def checkAdaptec(): |
| 51 | 95 | # Need to change directory so that the log file goes to the right place |
| 52 | 96 | oldDir = os.getcwd() |
| — | — | @@ -195,4 +239,96 @@ |
| 196 | 240 | print 'OK: State is %s, checked %d logical device(s)' % (state, numDrives) |
| 197 | 241 | return 0 |
| 198 | 242 | |
| | 243 | +def checkZfs(): |
| | 244 | + try: |
| | 245 | + proc = subprocess.Popen(['/sbin/zpool', 'list', '-Honame,health'], |
| | 246 | + stdout=subprocess.PIPE) |
| | 247 | + except: |
| | 248 | + error = sys.exc_info()[1] |
| | 249 | + print 'WARNING: error executing zpool: %s' % str(error) |
| | 250 | + return 1 |
| | 251 | + |
| | 252 | + regex = re.compile('^(\S+)\s+(\S+)') |
| | 253 | + status = 0 |
| | 254 | + msg = '' |
| | 255 | + for line in proc.stdout: |
| | 256 | + m = regex.match(line) |
| | 257 | + if m != None: |
| | 258 | + name = m.group(1) |
| | 259 | + health = m.group(2) |
| | 260 | + if health != 'ONLINE': |
| | 261 | + status = 2 |
| | 262 | + |
| | 263 | + if msg != '': |
| | 264 | + msg += ', ' |
| | 265 | + msg += name + ': ' + health |
| | 266 | + |
| | 267 | + ret = proc.wait() |
| | 268 | + if ret != 0: |
| | 269 | + print 'WARNING: zpool returned exit status %d' % (ret) |
| | 270 | + return 1 |
| | 271 | + |
| | 272 | + if status: |
| | 273 | + print 'CRITICAL: ' + msg |
| | 274 | + else: |
| | 275 | + print 'OK: ' + msg |
| | 276 | + return status |
| | 277 | + |
| | 278 | +def checkSoftwareRaid(): |
| | 279 | + devices = getSoftwareRaidDevices() |
| | 280 | + if len(devices) == 0: |
| | 281 | + print 'WARNING: Unexpectedly checked no devices' |
| | 282 | + return 1 |
| | 283 | + |
| | 284 | + args = ['/sbin/mdadm', '--detail'] |
| | 285 | + args.extend(devices) |
| | 286 | + try: |
| | 287 | + proc = subprocess.Popen(args, stdout = subprocess.PIPE) |
| | 288 | + except: |
| | 289 | + error = sys.exc_info()[1] |
| | 290 | + print 'WARNING: error executing mdadm: %s' % str(error) |
| | 291 | + return 1 |
| | 292 | + |
| | 293 | + deviceRegex = re.compile('^(/[^ ]*):$') |
| | 294 | + statRegex = re.compile('^ *(Active|Working|Failed|Spare) Devices *: *(\d+)') |
| | 295 | + currentDevice = None |
| | 296 | + stats = { |
| | 297 | + 'Active': 0, |
| | 298 | + 'Working': 0, |
| | 299 | + 'Failed': 0, |
| | 300 | + 'Spare': 0 |
| | 301 | + } |
| | 302 | + for line in proc.stdout: |
| | 303 | + m = deviceRegex.match(line) |
| | 304 | + if m == None: |
| | 305 | + if currentDevice == None: |
| | 306 | + continue |
| | 307 | + else: |
| | 308 | + currentDevice = m.group(1) |
| | 309 | + continue |
| | 310 | + |
| | 311 | + m = statRegex.match(line) |
| | 312 | + if m == None: |
| | 313 | + continue |
| | 314 | + |
| | 315 | + stats[m.group(1)] += int(m.group(2)) |
| | 316 | + |
| | 317 | + ret = proc.wait() |
| | 318 | + if ret != 0: |
| | 319 | + print 'WARNING: mdadm returned exit status %d' % (ret) |
| | 320 | + return 1 |
| | 321 | + |
| | 322 | + msg = '' |
| | 323 | + for name in ('Active', 'Working', 'Failed', 'Spare'): |
| | 324 | + if msg != '': |
| | 325 | + msg += ', ' |
| | 326 | + msg += name + ': ' + str(stats[name]) |
| | 327 | + |
| | 328 | + if stats['Failed'] > 0: |
| | 329 | + print 'CRITICAL: ' + msg |
| | 330 | + return 2 |
| | 331 | + else: |
| | 332 | + print 'OK: ' + msg |
| | 333 | + return 0 |
| | 334 | + |
| 199 | 335 | main() |
| Index: trunk/debs/wikimedia-raid-utils/debian/changelog |
| — | — | @@ -1,3 +1,10 @@ |
| | 2 | +wikimedia-raid-utils (1.8) lucid; urgency=low |
| | 3 | + |
| | 4 | + * Added Linux software RAID support |
| | 5 | + * Added ZFS pool support (on Solaris) |
| | 6 | + |
| | 7 | + -- Tim Starling <tstarling@wikimedia.org> Mon, 12 Jul 2010 19:34:16 +1000 |
| | 8 | + |
| 2 | 9 | wikimedia-raid-utils (1.7) hardy; urgency=low |
| 3 | 10 | |
| 4 | 11 | * Removed files erroneously installed in the root directory |