def pwQuery(ip):
"""run the proware query
"""
status = nagios.OK
message = snmp.get(ip, pwOID["defaultMsg"]).strip()
errMessages = []
# check proware events
events = snmp.walk(ip, pwOID["event"])
for name, sStat in events:
if sStat != "Good(1)":
status = nagios.CRITICAL
name = name.rstrip(".0").replace("event", "")
errMessages.append("%s-%s" % (name, sStat))
# disk status
sTable = snmp.walkTable("192.168.0.112", pwOID["slotTable"])
dStatCount = iterCount(sTable["slotStatus"])
# hot spare check
if "Free(2)" not in dStatCount:
dStatCount["Free(2)"] = 0
if dStatCount["Free(2)"] != 1:
status = nagios.CRITICAL
errMessages.append("%s Hot Spares" % (dStatCount["Free(2)"]))
# indivdual slot check
slotStat = dict(zip(sTable["slotDiskIndex"], zip(sTable["slotStatus"], sTable["slotBadBlockNumber"])))
sIdx = slotStat.keys()
sIdx.sort()
for s in sIdx:
sStat, sBB = slotStat[s]
if sStat not in ("Arranged(1)", "Free(2)"):
status = nagios.CRITICAL
errMessages.append("Slot %s Status-%s" % (s, sStat))
if sBB > 0:
errMessages.append("Slot %s-%s Bad Blocks" % (s, sBB))
if status == nagios.OK:
status = nagios.WARN
if len(errMessages) > 0:
message = ";".join(errMessages)
return status, message
def countList(ip,oid):
"""Go through a list (specified by oid) and
return a dict of { value:count,...} for allt values in list.
"""
return iterCount(walkList(ip,oid))
def aceraQuery(ip):
"""run the proware query
"""
status = nagios.OK
message = snmp.get(ip, aceraOID["defaultMsg"]).strip()
errMessages = []
# check voltages
for mv in ACREA_MV:
sMV = snmp.get(ip, aceraOID[mv])
if float(abs(sMV - mv)) / float(mv) >= WARN_MVOLT_PCT_TOL:
status = nagios.WARN
errMessages.append("%s mV line is %s mV" % (mv, sMV))
# temperatures
sBT = snmp.get(ip, aceraOID["boardTemp"])
if sBT >= WARN_TEMP:
status = nagios.WARN
errMessages.append("Controller Temp-%s C" % (sBT))
dskTbl = snmp.walkTable(ip, aceraOID["dskTbl"])
dskDict = dict(zip(dskTbl["hwHddIndex"], dskTbl["hwHddTemp"]))
for dsk, tmp in dskDict.items():
if tmp >= WARN_TEMP:
status = nagios.WARN
errMessages.append("Disk %s Temp-%s C" % (dsk, tmp))
# fan RPM
minRPM = min(snmp.walkList(ip, aceraOID["fanTbl"]))
if minRPM <= WARN_LOW_FAN_RPM:
status = nagios.WARN
errMessages.append("Fan RPM Low-%s" % (minRPM))
# CRITICAL tests follow
if minRPM <= CRIT_LOW_FAN_RPM:
status = nagios.CRITICAL
errMessages.append("Fan RPM Low-%s" % (minRPM))
# power supplies
pwrCt = snmp.countList(ip, aceraOID["pwrTbl"])
for stat, n in pwrCt.items():
if stat != "Ok(1)":
status = nagios.CRITICAL
errMessages.append("Power Supply Status-%s" % (stat))
# disk roles
dskStates = snmp.walkList(ip, aceraOID["hddStates"])
dskIdx = snmp.walkList(ip, aceraOID["hddIdx"])
dsCt = iterCount(dskStates)
if "Hot Spare" not in dsCt:
dsCt["Hot Spare"] = 0
if dsCt["Hot Spare"] != 1:
status = nagios.CRITICAL
errMessages.append("%s Hot Spares" % (dsCt["Hot Spare"]))
for n, s in enumerate(dskStates):
if s not in ("RaidSet Member", "Hot Spare"):
status = nagios.CRITICAL
errMessages.append("Slot %s Status-%s" % (dskIdx[n], s))
# raid+volume status
rNames = snmp.walkList(ip, aceraOID["raidNames"]) + snmp.walkList(ip, aceraOID["volNames"])
rStates = snmp.walkList(ip, aceraOID["raidStates"]) + snmp.walkList(ip, aceraOID["volStates"])
for n, s in enumerate(rStates):
if s != "Normal":
status = nagios.CRITICAL
errMessages.append("%s Status-%s" % (rNames[n], s))
if len(errMessages) > 0:
message = ";".join(errMessages)
return status, message