Added check_mellanox.

check_mellanox

+#!/usr/bin/python
+
+# Author: Devon Merner (dmerner)
+# Date: November 17th, 2016
+# Purpose: To monitor Mellanox switches in similar fashion to the check_openmanage script that was made by Trond H. Amundsen.
+
+STATE_OK=0
+STATE_WARNING=1
+STATE_CRITICAL=2
+STATE_UNKNOWN=3
+
+WARNING_TEXT=""
+CRITICAL_TEXT=""
+
+import sys
+import argparse
+import netsnmp
+
+parser = argparse.ArgumentParser()
+parser.add_argument("-c", default="public", dest="community",
+                    help="Community to query the agent")
+parser.add_argument("-v", default="2", dest="version",
+                    help="SNMP version to use")
+parser.add_argument("host", default="localhost", nargs="?",
+                    help="Agent to retrieve variables from")
+parser.add_argument("-d", action="store_true", dest="debug",
+                    help="Print debug information")
+
+# Warning/critical thresholds
+parser.add_argument("-tW", default="40", dest="tempwarn",
+                    help="Board temperature monitor warning threshold in degrees celsius (returns warning if equivelant or greater than this number) Default: 40")
+parser.add_argument("-tC", default="50", dest="tempcrit",
+                    help="Board temperature monitor critical threshold in degrees celsius (returns critical if equivelant or greater than this number) Default: 50")
+
+parser.add_argument("-fW", default="2000", dest="fanwarn",
+                    help="Fan monitor warning threshold in RPM (returns warning if equivelant or less than this number) Default: 2000")
+parser.add_argument("-fC", default="1000", dest="fancrit",
+                    help="Fan monitor critical threshold in RPM (returns critical if equivelant or less than this number) Default: 1000")
+
+options = parser.parse_args()
+args = {
+    "Version": int(options.version),
+    "DestHost": options.host,
+    "Community": options.community
+}
+
+dataReturned = 0
+
+# Collect fan/temperature sensor data
+for idx in netsnmp.snmpwalk(netsnmp.Varbind(".1.3.6.1.2.1.99.1.1"), **args):
+    dataReturned = 1
+
+    # Mellanox indexes start with 200 for some reason.
+    if not idx.startswith("200"):
+        continue
+    
+    sensortype, desc, units, value  = netsnmp.snmpget(
+        netsnmp.Varbind(".1.3.6.1.2.1.47.1.1.1.1.7", idx),
+        netsnmp.Varbind(".1.3.6.1.2.1.47.1.1.1.1.2", idx),
+        netsnmp.Varbind(".1.3.6.1.2.1.99.1.1.1.6", idx),
+        netsnmp.Varbind(".1.3.6.1.2.1.99.1.1.1.4", idx),
+        **args)
+
+    if (sensortype is None):
+	continue
+
+    # Convert into actual units if using SNMP v1
+    if (int(options.version) == 1) and sensortype.startswith("Temperature"):
+        value = float(value) / 10
+
+    if options.debug:
+        print("{} {} {} {}".format(sensortype, desc, value, units))
+
+    # Check Thresholds
+    if sensortype.startswith("Temperature") and desc.lower().startswith("mgmt/board"):
+        if value >= int(options.tempcrit):
+            CRITICAL_TEXT += desc + " temperature is " + str(value) + " " + units + ". "
+        elif value >= int(options.tempwarn):
+            WARNING_TEXT += desc + " temperature is " + str(value) + " " + units + ". "
+
+    elif sensortype.startswith("Fan"):
+        if int(value) <= int(options.fancrit):
+            CRITICAL_TEXT += desc + " speed is " + value + " " + units + ". "
+        elif int(value) <= int(options.fanwarn):
+            WARNING_TEXT += desc + " speed is " + value + " " + units + ". "
+
+
+if dataReturned == 0:
+    print("UNKNOWN - SNMP returned no data (Host is unreachable or SNMP is not responding) | ")
+    sys.exit(STATE_UNKNOWN)
+
+# Collect power supply information
+for idx in netsnmp.snmpwalk(netsnmp.Varbind(".1.3.6.1.2.1.47.1.1.1.1.1"), **args):
+
+    sensortype, desc, alarm = netsnmp.snmpget(
+        netsnmp.Varbind(".1.3.6.1.2.1.47.1.1.1.1.7", idx),
+        netsnmp.Varbind(".1.3.6.1.2.1.47.1.1.1.1.2", idx),
+        netsnmp.Varbind(".1.3.6.1.2.1.131.1.1.1.5", idx),
+        **args)
+
+    if sensortype is None:
+        continue
+    if not sensortype.startswith("Power"):
+        continue
+
+    alarmnice = "OK"
+    if alarm == "\x10":
+        alarmnice = "CRITICAL"
+        CRITICAL_TEXT += "Power Supply " + desc + " is CRITICAL "
+
+    if options.debug:
+        print("{} {} {}".format(sensortype, desc, alarmnice))
+
+# Critical/Warning Think Logic
+if CRITICAL_TEXT and (CRITICAL_TEXT is not None):
+    print("CRITICAL - " + CRITICAL_TEXT + "|")
+    sys.exit(STATE_CRITICAL)
+elif WARNING_TEXT and (WARNING_TEXT is not None):
+    print("WARNING - " + WARNING_TEXT + "|")
+    sys.exit(STATE_WARNING)
+else:
+    print("OK - No problems detected |")
+    sys.exit(STATE_OK)